The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction

World Development Vol. xx, pp. xxx–xxx, 2017 0305-750X/Ó 2017 Published by Elsevier Ltd. www.elsevier.com/locate/worlddev

http://dx.doi.org/10.1016/j.worlddev.2017.06.001

The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction RICHARD DAMANIA a, JASON RUSS a, DAVID WHEELER b and ALVARO FEDERICO BARRA a,* a World Bank, USA b Center for Global Development, USA Summary. — Roads bring significant economic benefits that are vital for development. But they are often also the precursors to deforestation and other adverse environmental impacts. This paper examines the road-induced tradeoffs between economic growth, deforestation, and biodiversity loss in the Democratic Republic of Congo (DRC). Decades of conflict have left the DRC’s transport infrastructure among the sparsest and most dilapidated in the world. Most of the provincial capitals are unconnected to the capital city, and improving road connectivity could lead to a significant boost in trade and economic growth. At the same time the DRC is also home to the second largest rainforest in the world. The iconic Congo forests are a trove of ecological value—some monetizable and most that is not. So the destruction of the DRC’s forests will have significant environmental ramifications. We provide empirical estimates of the economic benefits of improving market access and reducing transportation costs. We then estimate a forest destruction function to assess the impact that new or improved roads have on forest clearing. In addition, a novel biodiversity index is developed to identify forests of high biodiversity significance. Two simulations are performed to quantitatively demonstrate the impacts of road improvement projects in terms of increased GDP, forest loss, and biodiversity that are put at risk. To our knowledge, this is the first study to jointly examine the economic benefits and ecological risks to infrastructure investments. It is envisioned that the methods employed here can be used to guide future infrastructure investments toward designs which have a large economic impact while minimizing ecological risks. Ó 2017 Published by Elsevier Ltd. Key words — infrastructure, economic growth, forests, biodiversity

1. INTRODUCTION

of conflict and neglect have left the DRC’s transport infrastructure among the sparsest and most dilapidated in the world, even by the standards of other low-income countries (Africa Infrastructure Country Diagnostics 2008). In many parts of the country, traveling to the capital, Kinshasa, by road is impossible and most of the provincial capitals are unconnected to Kinshasa. Were economic activity evenly distributed across the country this may not matter significantly, but as Figure 1.1 illustrates, GDP in the country is highly geographically concentrated around the capital Kinshasa, with peaks in income around a few other areas such as Lubumbashi, Mbuji-Mayi, and Kivu. Connecting regions that flourish with those that lag (relatively) could provide a significant boost to economic growth. Given the vast distances and extreme variations in the spatial distribution of GDP, there are compelling reasons for improving inter-provincial as well as intra-provincial connectivity to promote trade and economic cohesion. However, the DRC is also home to the second largest rainforest in the world. The iconic Congo forests are a trove of ecological value—some monetizable and much that is not. The DRC’s forests are distinguished by the unusually high number of endemic and endangered species (UNESCO, 2010). 1 The carbon sequestered by these forests (a stock of about 30–40 gigatons) corresponds to about 3–5 years of CO2 equivalent emitted globally (as a flow). So the destruction of the DRC’s forests could have global ramifications. Roads often catalyze a process of deforestation and land conversion. In addition, they are also accompanied by a litany of other

Roads bring significant economic benefits that are vital for development. But they are often also the precursors to deforestation and other adverse environmental impacts. The response of conservation managers in tropical forests has typically been reactive. Attempts are made to limit damage through the demarcation and protection of areas that are deemed critical for biodiversity conservation. This strategy seeks to minimize ecological impacts by preventing or severely restricting road improvements that increase the profitability of forest clearing within protected areas. Potential conflict over the desirability of road improvements is particularly high in forested regions with significant agricultural potential. When conventional protected-area strategies confront this conflict, they may fail to protect critical ecosystems for two reasons. First, governments may seek to minimize economic opportunity costs by siting protected areas in remote regions with low agricultural potential that may not coincide with the areas of highest ecological value. Second, attempts to restrict road improvements in protected areas with strong agricultural potential may fail because economic interests overwhelm the limited resources and political support of conservation managers. This paper presents an empirical approach that seeks to mitigate such conflicts by developing rigorous tools that can help steer infrastructure development toward sites where economic benefits can be realized, while ecological damage is avoided and minimized. We illustrate this approach by estimating the potential impacts of road upgrading in the Democratic Republic of Congo (DRC). The DRC, with its immense forests and woeful road network, presents an apt case study for this exercise. Decades

* We are grateful to 3 reviewers for helpful and incisive remarks. The views expressed in this manuscript are those of the authors and not the institutions to which they are affiliated. Final revision accepted: June 16, 2017. 1

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

2

WORLD DEVELOPMENT

Figure 1.1. Local GDP in DRC, 2006. Source: Ghosh et al. (2010).

forms of environmental degradation, especially in the DRC, where enforcement of regulations is immensely challenging. For instance, poaching, illegal trapping of exotic wildlife for the pet trade, fuel wood collection, and forest fires are among the plethora of problems that accompany roads in biologically sensitive forests. This suggests the need to establish procedures that preempt adverse and often irreversible consequences of road construction and yet allow for benefits of development to be realized. The exercise outlined in this paper draws on a variety of disciplines—GIS analysis, econometrics, and conservation biology—to create an approach that could guide the location and level of investments in roads. The methodology involves four steps. First the benefits of transport infrastructure are estimated using regression analysis. We provide what we believe are the most accurate estimates of transport costs that are available for the DRC and carefully address potential sources of endogeneity bias arising from the non-random placement of roads and the spatial sorting of cities. 2 Next, a disaggregated spatial data set of forest loss is used to estimate the effects of roads on forest cover. Recognizing that not all forests are of identical ecological significance, a new composite metric of biodiversity is developed to identify forests of high (and low) biodiversity significance. In the final stage the spatial estimates are combined to simulate the effects of different policies and identify hotspots where risks are high and benefits relatively low, areas where the reverse holds, and regions where there are large trade-offs between economic and ecological goals. To our knowledge, the present paper represents the first attempt to combine these unconnected strands of research to enable better informed approaches to road infrastructure investments. The remainder of the paper is organized as follows. Section 2 motivates the analysis with a brief review of prior empirical research on the econometric estimates of road infrastructure benefits, the economics of forest clearing, and measures of biodiversity. Section 3 describes several of the datasets employed in the analysis. Section 4 presents the results from estimating the benefits of road construction in the DRC. Section 5 presents estimates of a deforestation model that incorporates the impact of road improvement. Section 6 derives a biodiversity index of the Congo Basin that incorporates four distinct measures of biodiversity. In Section 7, we explore the implications of our results for local, regional, and national forest clearing. Section 8 concludes the paper with caveats.

2. PRIOR RESEARCH The purpose of this section is to briefly review several hitherto distinct strands of literature: the literature on estimating the economic benefits of roads, a different body of work on the empirical drivers of deforestation, the environmental damage from development, and research on biodiversity indicators. The objective is not to provide a comprehensive assessment of these burgeoning areas of research, but merely to highlight some of the more relevant contributions. The empirical literature on the economic benefits of roads is vast and rapidly evolving. Much of the recent analysis is concerned with identifying causal relationships between investments in roads and consequent economic impacts, with approaches that have varied considerably over time. Researchers have examined the effects of road infrastructure and investments in transport on aggregate productivity, usually measured by GDP (Aschauer, 1989; Ihori, Doi, & Kondo, 2001), with ambiguous results. To a large extent the contradictory evidence and the ensuing debates are a consequence of identification and reverse causality problems. Recent papers have used more rigorous and compelling identification strategies to shed light on these issues (Donaldson, 2010; Datta, 2012; Faber, 2014). One solution is to use panel data methods (Khandker & Koolwal, 2011), but the approach is limited by a lack of suitable time series data, especially in developing countries. Other papers exploit natural experiments (Donaldson, 2010), comparing regions where infrastructure investments were made with regions where they were planned but never completed. However, such assessments are uncommon since the fortuitous circumstances for a natural experiment are rarely encountered. Instead, much of the literature uses a difference-in-difference (Datta, 2012), or difference-in-difference with an instrumental variable (Faber, 2014) approach, or some exogenous geographic features to exploit natural differences in a sample (Jacoby & Minten, 2009; Shrestha, 2012; Emran & Hou, 2013). The approach used in this paper is most closely related to that of Faber (2014), Damania, Berg, Russ, Federico Barra, Nash, & Ali (2017), and Russ, Berg, Damania, Barra, Ali, & Nash (2017) which rely on an instrumental variable based on exogenous variation in geography. A distinct theoretical and empirical research on the determinants of forest clearing has provided many useful economic policy insights. The von Thunen model offers a convenient framework for understanding the process and links between

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

the spatial determinants of deforestation (Angelsen, 2007). Briefly, interventions such as (say) investments in roads that increase the payoffs to alternative activities such as agriculture, will induce a decline in forest cover. Eventually, when diminishing returns set in, this trend may be reversed and some of the lost forest may recover as suggested by forest transition theories. Empirical results are generally consistent with this model in which the conversion of forested land varies with rents and potential profitability. Nelson and Chomitz (2009) and Rudel, Defries, Asner, and Laurance (2009) have studied this relationship across countries over multi-year intervals. Within countries, numerous econometric studies have estimated the impact of economic, social, and geographic drivers on deforestation during multi-year intervals. Some studies have used aggregate data for states, provinces, or subprovinces (e.g., studies for Brazilian municipios by Goeschl and Igliori (2006), and Mexican states by Barbier & Burgess, 1996). Many studies have also used GIS-based techniques to obtain multi-year estimates at a higher level of spatial disaggregation (e.g., Cropper, Puri, Griffiths, Barbier, & Burgess, 2001 for Thailand; Agarwal, Gelfand, and Silander (2002) for Madagascar; and Vance & Geoghegan, 2002 for southern Mexico). In rarer cases, studies have used annual national or regional aggregate time series for extended periods (e.g., Zikri, 2009 for Indonesia; Ewers, Laurance, & Souza, 2008 for Brazil). In perhaps the most comprehensive survey of the literature, Geist and Lambin (2002) show that tropical deforestation is driven by broad macro-level factors such as institutions as well as factors that drive agricultural expansion, wood extraction, and infrastructure. Their analysis suggests that too much emphasis may have been given to other factors such as population growth and shifting cultivation as primary causes of deforestation. The most recent research has exploited higher resolution spatial panel data for more precise identification of deforestation and forest degradation drivers. In a synthesis of nationallevel studies, Hosonuma et al. (2012) find that deforestation drivers are similar in Africa and Asia, while degradation drivers are more similar between Latin America and Asia. Globally, commercial and subsistence agriculture are the primary and secondary drivers of deforestation, while degradation is driven, in decreasing order of importance, by timber extraction and logging, fuelwood collection, charcoal production, and livestock grazing. In a multi-country study for South America and Southeast Asia, Henders, Persson, and Kastner (2015) find that production of four commodities—beef, soybeans, palm oil, and wood products—accounts for 40% of deforestation. In Southeast Asia, Richards and Friess (2016) augment previous research on moist tropical forests by focusing on clearing of coastal mangrove systems. Finding a lowerthan-expected role for coastal aquaculture, they identify rice agriculture as a major driver of mangrove loss in Myanmar, and oil palm expansion as a critical threat in Malaysia and Indonesia. In South America, Sy et al. (2015) find that the primary deforestation drivers are pasturage (71% of cleared area) and commercial agriculture (14%). In a related assessment for Latin America using MODIS imagery, Graesser, Aide, Grau, and Ramankutty (2015) find forest-clearing contributions of 57% for new pastureland and 17% for cropland. Of particular relevance for the current study are recent findings on deforestation drivers in Sub-Saharan Africa (SSA). DeFries, Rudel, Uriarte, and Hansen (2010) find that deforestation rates in SSA remained significantly below rates in Latin America and Southeast Asia during the period 2000–05.

3

Rudel (2013) and Fisher (2010) find that population-driven expansion of lands for tropical food crops explains a significant proportion of moist forest clearing in many SSA countries. However, Rudel (2013) identifies accelerated urbanization as a deforestation-reducing factor in countries where resource extraction is a dominant source of government revenue. Mayaux et al. (2013) modify this finding by linking accelerated urbanization to more rapid deforestation around cities and an emerging spatial pattern of deforestation along transportation corridors between cities. Rudel (2013) stresses the importance of the Mayaux results for the DRC, while noting the additional role played by conflict-related population displacement. A somewhat smaller and distinct literature explores the consequences of road building and other development activities on wildlife poaching and hunting. In an empirical analysis of hunting pressures Abernethy, Coad, Taylor, Lee, and Maisels (2013) find that the devastating decline in Central African elephant numbers is highly correlated with proximity to roads as well as other country-level factors such as governance indicators, and population density. The most commonly hunted species in African rainforest are small ungulates, monkeys, and rodents, usually trapped with wire and snares (Abernethy, Coad, Taylor, Lee, & Maisels, 2013). Evidence suggests that roads increase bushmeat hunting not only by improving access for hunters, but also by stimulating local demand and facilitating wildlife trade out of local villages. When roads improve access to markets, larger and more commercially valuable species such as forest elephants and apes are targeted using more expensive hunting techniques, resulting in local extinctions as predicted by open access models of species harvesting (Laurance et al., 2006; Damania, Milner-Gulland, & Crookes, 2005). Finally there is an extensive literature on measuring and defining biodiversity. Most contributions acknowledge that biodiversity is too complex to be adequately quantified in a single measure that is suitable for all purposes. Instead, a variety of measures have been developed that vary in spatial scale and level of detail, with the suitability of each determined by the problem under consideration. For instance, Noss (1990) distinguishes between biotic indicators that assess the status of particular species and abiotic ones that refer to the broader habitat and seek to capture the level of environmental stress in an area. Measures also differ by spatial scale. Alpha biodiversity refers to the range of species found within a given area (such as an ecological compensation area), whereas beta biodiversity compares the abundance and distribution across a larger spatial scale, often between areas, and gamma diversity is the total species diversity in a landscape. There is debate on how each of these should be measured and how they mathematically relate to each other. The widely used IUCN Red List of Species is one of the best known examples of this approach. Comparisons of biodiversity richness between regions are difficult and often require judgements to be made regarding the complementarity or uniqueness of particular species and groups. To capture these attributes, measures of species endemism, spatial turnover, and genetic distance on the phylogenetic tree have also been developed (Kinzig & Harte, 2000; Duelli & Obrist, 2003). In the spirit of measuring ecosystems rather than individual species, WWF has developed maps of ecoregions that contain distinct assemblages of species and biota across a large range of flora and fauna (Groves et al., 2002). In Section 6, this paper develops a composite biodiversity measure that seeks to capture the main features of these metrics—species abundance, genetic rarity, endemicity, and ecoregions.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

4

WORLD DEVELOPMENT

3. DATA In this section we present a brief discussion of several of the spatially explicit datasets used in this analysis. Much of our data is calculated from raster datasets which are extracted into a gridded framework. For our analysis on the economic impacts of roads, grid cells are 5 arcmin2, or approximate 10 km2 at the equator. When we analyze the impact of roads on deforestation, we use a smaller aggregation level, 2.7 km2, as the forestry data are available at a much finer level. (a) Calculating transport costs In order to estimate the cost of transporting goods around the DRC road network, we first compile the road network. Since no up-to-date and complete road network exists, we assemble it from two different sources. We begin with a GIS vector of the transportation network obtained from Delorme, which provides a thorough data set of both major trunk roads, as well as rural roads throughout the country. This network is overlaid with another road network constructed for the African Infrastructure Country Diagnostic (AICD), which includes quality attributes such as whether the road is paved/unpaved, the road type (primary, 7-m-wide roads; secondary, 6-m-wide roads; and tertiary, 5-m-wide roads), and road quality (good; fair; poor). 3 These attributes are transferred to Delorme’s vector data. Finally, the road network is updated by making adjustments based on information obtained from transport experts familiar with DRC. To our

knowledge this provides the most complete data set of the network that is currently available. In order to calculate the costs of traveling along the road network, we apply the Highway Development Management Model (HDM-4), a standard model frequently used by road engineers. This model takes as inputs the road attributes available in the AICD dataset, the roughness of terrain along the road, as well as country-level information on various factors which can affect the price of transporting goods (i.e., price of fuel, labor costs, etc). The output is the cost per kilometer of transporting a ton of goods in a heavy truck, for every possible road classification combination. 4 We use this model to calculate the least-cost route for traveling from the centroid of every grid cell within the DRC to every market, where a market is defined as a city of 50,000 people. The least cost market is then determined by the location that is cheapest to travel to. 5 (b) Economic outcome variables In order to estimate the economic impact of improving road infrastructure, we utilize two separate datasets. The first, our main outcome variable, measures local GDP at the grid cell level. To test the robustness of our estimates, we also employ a dataset which measures crop production for several important crops in the region. Our measure of local GDP is obtained from Ghosh et al. (2010). This dataset uses a model to spatially disaggregate 2006 GDP into a gridded framework by utilizing nighttime

Table 4.1. Effects of transport costs on local GDP Dependent Variable:

ln(Cost to Market), USD ln(Conflict Fatalities) ln(Population) ln(Population)^2 ln(Distance to Mine), km ln(Distance to Mine)^2, km ln(Natural Path), hours ln(Euclidean Distance), km ln(Distance to East Border), km Other variables included: N R-sq Hansen J p-value Kleibergen–Paap LM p-value Kleibergen–Paap F-stat

(1) OLS

(2) 2SLS

(3)

(4)

(5) Spatial 2SLS

ln(GDP)

ln(Cost to Market)

ln(Conflict Fatalities)

ln(GDP)

ln(GDP)


0.141** (
2.39)
0.122* (
1.77) 0.525*** (59.89) 0.0356*** (50.70)
0.300*** (
3.22) 0.0371*** (3.25)


0.083*** (
8.15)
0.049*** (
4.29) 0.681*** (118.64) 0.023*** (49.89)
0.164*** (10.40) 0.02*** (10.40)

23,489

23,489


0.0330*** (
4.25) 0.00392* (1.93) 0.526*** (59.11) 0.0356*** (51.66)
0.193*** (
2.96) 0.0212*** (3.21)


0.0045***
0.0032 (
4.04) (0.735)
0.0009*** 0.0002 (
7.04) (0.867)
0.1626***
0.8210*** (
5.18) (
4.60) 0.0149*** 0.1253*** (4.64) (6.45) 0.1160 0.2833*** (24.15) (1.24) 0.6288***
0.8854*** (50.78) (
8.97) 0.0845***
0.2801*** (13.78) (
5.90) Quadratic terms for agricultural potential yields of cassava, bananas/plantains, groundnuts, maize, and rice; and marketshed fixed effects 23,489 0.930

23,489

23,489

0.2236 0.0000 10.023

Note: Sample includes all non-urban cells. Columns 1 presents coefficients from an OLS regression with ln(local GDP) as the dependent variable. Columns 2–4 present 2SLS regressions, with the first stage for ln(cost to market) and ln(conflict fatalities) in columns 2 and 3, respectively, and the second stage in column 4. Robust t-statistics are in parentheses. Column 5 presents spatial 2SLS estimates with bootstrapped standard errors in parentheses. *p < 0.1, ** p < 0.05, ***p < 0.01.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

light satellite imagery collected by the National Oceanic and Atmospheric Administration (NOAA), as well as data on agricultural GDP. The methodology is predicated on the fact that brighter lights at night have been shown to be associated with higher levels of economic activity, particularly in urban areas (Elvidge et al., 2001; Ebener, Murray, Tandon, & Elvidge, 2005; Sutton, Elvidge, & Ghosh, 2007; Chen & Nordhaus, 2011). Indeed, several recent studies have either directly used nighttime lights, or employed modified functions thereof as a proxy for economic activity (e.g., Doll, Muller, & Elvidge, 2000; Henderson, Storeygard, & Weil, 2012; von Uexkull, Croicu, Fjelde, & Buhaug, 2016; Henderson, Storeygard, & Deichmann, 2017). The Ghosh et al. dataset offers an important improvement over directly using nighttime lights to proxy for economic activity. Nighttime lights have been shown to significantly underestimate rural economic activity, where access to electricity can be relatively sparse (Mellander, Lobo, Stolarick, & Matheson, 2015). This is a particular concern in a country like DRC, where GDP from rural agriculture makes up a significant percentage of total GDP (fluctuating between 20 and 25% of total GDP over the past decade (World Bank, 2017)). In the Ghosh et al. dataset, non-agricultural GDP is modeled separately from agricultural GDP, thus avoiding the urban bias intrinsic to raw nighttime lights data. To test the robustness of our local GDP estimates, we also estimate the impact of improving transport infrastructure on agricultural productivity. To do so, we utilize the Spatial Production Allocation Model (SPAM) dataset from HarvestChoice. (2012), which contains gridded data on the

5

production of several crops for DRC. Five crops, which are particularly important in DRC are selected: cassava, bananas/plantains, maize, groundnuts, and rice. According to the Food and Agriculture Organization (FAO), in 2011 these 5 crops accounted for approximate 75% of the value of all crops produced in DRC (FAO & Faostat, 2017). (c) Forest cover and forest loss To measure deforestation, this study uses the first highresolution, consistently derived estimates of global forest clearing, published by Hansen et al. (2013). The data are tiff panels at 30 m2 spatial resolution for 2000–12. These have been converted to annual files in which cleared pixels are assigned the value 1 in the year when most clearing occurred. Uncleared pixels are assigned the value 0. For tractability the pixels are aggregated to 2.7 km2 grid cells. However, each grid cell contains 8,100 pixels each of 30 m2; so total counts of cleared pixels within grid cells are equivalent to deforestation rates. This aggregation therefore still allows us to observe the impact of small-level, artisanal deforestation which is quite common in DRC. Henceforth, the 30 m2 pixels are referred to as ‘‘Hansen pixels” and the aggregated grid cells are referred to as ‘‘Hansen grid cells”. 4. ESTIMATING THE BENEFITS OF ROADS In this section, we estimate the economic benefits of improving road infrastructure. In order to do so, we estimate

Table 4.2. Effect of transport costs on Cassava Production (1) OLS Dependent Variable:

(2) 2SLS

(3)

(4)

(5) Spatial 2SLS

ln(Cassava Production) ln(Cost to Market) ln(Conflict Fatalities) ln(Cassava Production) ln(Cassava Production)


0.203*** (
2.70) ln(Conflict Fatalities)
0.0293 (
1.26) ln(Population) 0.227*** (15.80) ln(Population)^2 0.0356*** (14.21) ln(Distance to Mine), km 2.767*** (4.00) ln(Distance to Mine)^2, km
0.286*** (
3.97)

ln(Cost to Market), USD


0.0046*** (
4.36)
0.0011*** (
9.65)
0.0792*** (
2.60) 0.0066** (2.10)

0.0069 (0.69)
0.0013 (
1.31) 0.6984*** (4.36)
0.0131 (
0.77)


1.876*** (
3.96)
1.805*** (
3.44) 0.227*** (9.19) 0.0320*** (9.61) 4.074*** (4.75)
0.315*** (
3.93)


1.126*** (
11.14)
1.087*** (
9.51) 0.124*** (5.31) 0.044*** (20.67) 2.54*** (16.77)
0.198*** (
14.35)

0.1510* 0.2791*** (24.08) (1.67)
0.9568*** ln(Euclidean Distance), km 0.6329*** (53.30) (
10.11) ln(Distance to East Border), km 0.0551***
0.3120*** (8.86) (
7.28) Other variables included: Quadratic terms for agricultural potential yields of cassava and marketshed fixed effects ln(Natural Path), hours

N 26,384 R-sq 0.308 Hansen J p-value Kleibergen–Paap LM p-value Kleibergen–Paap F-stat

26,384

26,384

26,384

N per sample: 414–443 # Samples: 3,900

0.2153 0.0000 16.821

Note: Sample includes all non-urban cells with positive cassava potential yield, according to FAO/GAEZ. Columns 1 presents coefficients from an OLS regression with ln(Cassava production) as the dependent variable. Columns 2–4 present 2SLS regressions, with the first stage for ln(cost to market) and ln (conflict fatalities) in columns 2 and 3, respectively, and the second stage in column 4. Robust t-statistics are in parentheses. Column 5 presents spatial 2SLS estimates with bootstrapped standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

6

WORLD DEVELOPMENT

a cross-sectional regression where we test the relationship between market access via the road network, and several economic outcomes. Specifically, we estimate the benefits of reducing the costs of traveling to the cheapest market as defined in (a). In order to test this relationship, we estimate the following equation: lnðY i Þ ¼ b0 þ b1 lnðT i Þ þ b2 C i þ X 0i c þ ei

ð4:1Þ

where Y i is output or welfare in location i, T i is the cost of traveling to the cheapest market from location i, C i is the conflict intensity around cell i, and X i is a vector of control variables. In our main set of results, Y i represents local GDP (in millions USD) in grid cell i, as described in (b). In subsequent results, we also replace local GDP with agricultural production (in tonnes) of several staple crops to DRC. Conflict intensity around cell i, C i , is calculated using data from the Armed Conflict Location Events Dataset (ACLED) version 4 (Raleigh, Linke, Hegre, & Karlsen, 2010). Since the impacts of conflict spill over beyond the precise location where a violent encounter may have occurred, we transform points of conflict into conflict intensity, using a standard kernel density function. To calculate the intensity of conflict around any point, the kernel density function takes a weighted average of the fatalities of all the conflicts around that point. The magnitude of the weight declines with distance from the point, according to the chosen kernel function. 6 Control variables included in X i enter in quadratic form, and include: population of cell i, obtained from LandScan

(2006); agro-ecological potential yield of cell i, obtained from GAEZ (FAO and IIASA 2000) 7; and Euclidean distance to the nearest mine, calculated from data available from the National Minerals Information Center of the USGS (Matos, Miller, & Barry, 2015). Because of the granularity of the pixels, and the focus on measuring the effect of reducing the transport costs to the nearest markets, all of which are cities, we remove urban areas from the dataset, 8 as well as grid cells which have zero agro-ecological potential (such as lakes and rivers, however, we do not omit cells which have positive agro-ecological potential, but zero agricultural production). When estimating Eqn. (3.1), there are several identification and estimation challenges that arise. These include: (1) estimating the cost of transporting goods around the road network; (2) endogeneity bias arising from non-random placement of roads; (3) endogeneity bias arising from spatial sorting of people (i.e., historic migration) and cities; (4) endogeneity bias arising from two-way causality between conflict and incomes/production; and (5) bias arising from spatial autocorrelation. We will briefly discuss each of these challenges and how they are handled. Recognizing that roads are non-randomly placed, and often sited where they will have the biggest economic impact, ordinary least squares (OLS) estimates will be biased. Hence, we adopt an instrumental variable (IV) strategy to eliminate this bias. The literature on the economic benefits of roads typically relies on one of two types of IVs; straight line, or ‘‘Euclidean distance” IVs, and historical route IVs. Of late, thanks in part to greater access to digitized historical maps and books, the

Table 4.3. Effect of Transport Costs on Banana/Plantain Production (1) OLS

(2) 2SLS

(3)

(4)

(5) Spatial 2SLS

Dependent Variable:

ln(Banana/Plantain Production)

ln(Cost to Market)

ln(Conflict Fatalities)

ln(Banana/Plantain Production)

ln(Banana/Plantain Production)

ln(Cost to Market), USD


0.785*** (
13.05) 0.0158 (0.84) 0.263*** (11.45) 0.0404*** (15.96)
7.073*** (
7.27) 0.455*** (4.77)


0.0038 (
0.41)
0.0003 (
0.34)
0.6850*** (
3.85) 0.108*** (5.55)


1.751*** (
4.97)
0.862** (
2.22) 0.251*** (10.50) 0.0384*** (14.35)
7.615*** (
7.52) 0.548*** (5.29)


1.532*** (
21.90)
0.786*** (
10.03) 0.059*** (2.93) 0.059*** (35.35)
10.35*** (
78.84) 0.806*** (54.729)

ln(Conflict Fatalities) ln(Population) ln(Population)^2 ln(Distance to Mine), km ln(Distance to Mine)^2, km ln(Natural Path), hours ln(Euclidean Distance), km ln(Distance to East Border), km Other variables included: N R-sq Hansen J p-value Kleibergen–Paap LM p-value Kleibergen–Paap F-stat


0.0039*** (
3.59)
0.0009*** (
7.37)
0.1674*** (
5.41) 0.0154*** (4.85)

0.2796*** (24.38) 0.6372*** (53.12) 0.0572*** (9.08) Quadratic terms for agricultural potential 23,755 0.411

0.1987** (2.20)
1.0068*** (
10.64)
0.333*** (
7.81) yields of bananas and marketshed fixed effects 23,755

N per sample: 414–443 # Samples: 3,900

0.0000 0.0000 19.968

Note: Sample includes all non-urban cells with positive banana/plantain potential yield, according to FAO/GAEZ. Columns 1 presents coefficients from an OLS regression with ln(Banana/Plantain production) as the dependent variable. Columns 2–4 present 2SLS regressions, with the first stage for ln(cost to market) and ln(conflict fatalities) in columns 2 and 3, respectively, and the second stage in column 4. Robust t-statistics are in parentheses. Column 5 presents spatial 2SLS estimates with bootstrapped standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

use of historical road IVs has been growing. While these two types of IVs are very different in their formulation, they are both attempts at estimating the same thing; namely, the natural way for humans to travel over land, absent the presence of a road network. Historical routes are useful as IVs, in that they represent the easiest path to travel over land. As these historical paths were constructed with little or no technology, they generally follow a smoother terrain and have been used for hundreds of years, and are thus the most cost effective routes to construct a road. At the same time, they are not correlated with the current economic benefits that lead to the endogeneity bias, given that in many cases, these routes were constructed well over 100 years ago. 9, 10 Following this reasoning, we generate a new instrument, called the Natural–Historical path (NHP). As suggested by its name, the NHP takes into consideration historical data on caravan routes from the 19th century used to transport ivory and other goods, as well as the terrain and historical land cover within DRC to estimate the quickest path, on foot, to and from anywhere within DRC’s borders, in the absence of any transport infrastructure. The NHP is affected by the terrain and land cover of the land between location i and the market, and not geographic attributes at either of those points themselves, suggesting that the exclusion restriction will not be violated. For more information on how the NHP was constructed, see the Natural–Historical Path Appendix. A combination of both the natural path and historical caravan data represents an improved estimate of how people traveled over land in prior centuries, and thus, it is arguably

7

a suitable IV for transportation cost. A major problem with only using historical path data is that people now live in areas that may have been uninhabited, or not a part of the trade network, many years ago. Historical path data will therefore not be able to identify the likely paths that would have been used to travel to and from those locations. Using natural path data, we are able to fill in gaps in the historical caravan data to get a complete picture of the optimal historical travel paths. In addition to the NHP, we also use Euclidean distance to the nearest market as an instrument for travel costs. Historic migration (i.e., spatial sorting of people), and nonrandom placement of cities (i.e., spatial sort of cities), may also bias regression estimates if not properly accounted for. If people and cities tend to locate where economic potential is highest, either because of natural endowments or available infrastructure, then cross-sectional estimates of the effect of market access on economic outcomes will be biased. In order to account for this, we include two sets of control variables in tandem. The first is the agro-ecological potential of the land, which will account for natural differences in land fertility and climate. Secondly, we also include marketshed fixed effects, which are fixed effects based on the different markets that each location travels to (i.e., is their cheapest market). This controls for other natural as well as human-made (e.g., infrastructure) endowments that might contribute to migration into and out of the area, or the formation of cities. Including both of these variables in the regression should account for regional differences that might lead one area to outperform another economically.

Table 4.4. Effect of Transport Costs on Maize Production (1) OLS

(2) 2SLS

(3)

(4)

(5) Spatial 2SLS

Dependent Variable:

ln(Maize Production) ln(Cost to Market) ln(Conflict Fatalities) ln(Maize Production) ln(Maize Production)

ln(Cost to Market), USD


0.0515 (
0.83)
0.0322* (
1.66) 0.190*** (15.67) 0.0220*** (10.60) 1.665*** (2.60)
0.193*** (
2.93)

ln(Conflict Fatalities) ln(Population) ln(Population)^2 ln(Distance to Mine), km ln(Distance to Mine)^2, km


0.0049*** (
4.61)
0.0011*** (
9.38)
0.0782** (
2.55) 0.0066** (2.10)

0.2874*** (25.18) ln(Euclidean Distance), km 0.6233*** (53.28) ln(Distance to East Border), km 0.0512*** (8.88) Other variables included: Quadratic terms for agricultural potential

ln(Natural Path), hours

N R-sq Hansen J p-value Kleibergen–Paap LM p-value Kleibergen–Paap F-stat

26,914 0.268

26,914

0.0053 (0.54) 0.0004 (0.46) 0.6661*** (4.19)
0.0062 (
0.36)


0.968*** (
3.52)
0.979*** (
3.45) 0.188*** (11.82) 0.0221*** (9.64) 2.332*** (3.31)
0.202*** (
2.92)


0.655*** (
10.81)
0.732*** (
11.58) 0.181*** (9.58) 0.002*** (15.33) 1.25*** (10.51)
0.108*** (
10.37)

0.2221** (2.49)
1.050*** (
11.31)
0.3701*** (
8.99) yields of maize and marketshed fixed effects 26,914

26,914

N per sample: 414–443 # Samples: 3,900

0.3471 0.0000 13.43

Note: Sample includes all non-urban cells with positive maize potential yield, according to FAO/GAEZ. Column 1 presents coefficients from an OLS regression with ln(Maize production) as the dependent variable. Columns 2–4 present 2SLS regressions, with the first stage for ln(cost to market) and ln (conflict fatalities) in columns 2 and 3, respectively, and the second stage in column 4. Robust t-statistics are in parentheses. Column 5 presents spatial 2SLS estimates with bootstrapped standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

8

WORLD DEVELOPMENT

The conflict variable also has the potential to bias the regression due to the fact that causality between economic production and conflict can run in both directions. Conflict, for obvious reasons, can lead to lower investment levels, lower incomes, and lower welfare in general. At the same time, lack of economic opportunities and poor institutions (e.g., rule of law) can also lead some to join rebel or insurgent militias. 11 This implies that conflict is likely to occur in poorer areas, and is also likely to depress these areas further, resulting in a two-way causality. We therefore address this potential bias by including an additional instrument which measures the distance from the grid cell centroid to the eastern border with Uganda, Rwanda, and Burundi, as there is greater conflict along this border. Finally, spatial autocorrelation has the potential to bias both the estimates and their standard errors. In order to account for this, we run a spatial bootstrapping technique in which we re-sample the entire dataset, randomly selecting a single-grid cell, and removing all grid cells within 65 km 12 of it. We continue to randomly sample the dataset until there are no remaining grid cells. The end result is a dataset with between 414 and 443 observations which are spatially independent. We generate 3,900 of these datasets and use them to calculate bootstrapped estimates and standard errors which are not biased by spatial autocorrelation. We now present results from estimating Eqn. (4.1). Estimates when the dependent variable is local GDP are shown in Table 4.1. Column 1 presents coefficients when Eqn. (4.1) is estimated via OLS, and columns 2–4 present two-stage

least-squares (2SLS) regressions, with the first stage for ln(cost to market) and ln(conflict fatalities) in columns 2 and 3, respectively, and the second stage in column 4. Column 5 then presents estimates from a spatial 2SLS regression. Because all variables are in logarithmic form, we can interpret coefficients as elasticities. We see in column 4 that reducing transport costs to the local market by 10% would lead to an increase in local GDP by 1.41%. As one would expect, conflict tends to reduce local GDP. We also see that local GDP increases convexly in population, and decreases concavely with distance to mine. First-stage estimates in columns 2 and 3 show that our instruments strongly pass the inclusion restriction, with higher natural path and Euclidean distance values associated with increased costs to travel to the market, and conflict fatalities increasing with distance to the Eastern border. We also fail to reject the Hansen J null hypothesis that our overidentifying restrictions are valid, we reject the null hypothesis of the Kleibergen–Paap LM test of under identification, and the Kleibergen–Paap F-stat is above 10, which implies that this regression likely does not suffer from weak instruments. The diagnostic tests therefore give confidence in our chosen instruments. Finally, column 5 confirms that these coefficients and errors terms are not significantly biased due to spatial auto-correlation, with the coefficients changing only slightly (becoming slightly smaller for cost to market and conflict variables), and retaining their statistical significance. Tables 4.2–4.6 display results when production of cassava, bananas/plantains, maize, groundnuts, and rice, respectively, are the dependent variables. The 2SLS results in column 4,

Table 4.5. Effect of Transport Costs on Groundnuts Production (1) OLS

(2) 2SLS

(3)

(4)

(5) Spatial 2SLS

Dependent Variable:

ln(Groundnut Production)

ln(Cost to Market)

ln(Conflict Fatalities)

ln(Groundnut Production)

ln(Groundnut Production)

ln(Cost to Market), USD


0.630*** (
11.33) 0.0106 (0.60) 0.251*** (14.67) 0.0288*** (13.68) 2.412*** (4.04)
0.311*** (
5.10)


2.086*** (
4.93)
1.506*** (
3.14) 0.239*** (11.77) 0.0259*** (10.16) 3.748*** (4.64)
0.362*** (
4.99)


1.485*** (
20.58)
0.90*** (
10.88) 0.32*** (16.42) 0.022*** (13.18) 2.51*** (18.80)
0.266*** (
22.98)

ln(Conflict Fatalities) ln(Population) ln(Population)^2 ln(Distance to Mine), km ln(Distance to Mine)^2, km ln(Natural Path), hours ln(Euclidean Distance), km ln(Distance to East Border), km Other variables included: N R-sq Hansen J p-value Kleibergen–Paap LM p-value Kleibergen–Paap F-stat


0.0051*** (
4.76)
0.0011*** (
9.65)
0.0819*** (
2.65) 0.0069** (2.19)

0.1313 0.2767*** (24.26) (1.47) 0.6350***
0.9225*** (53.93) (
9.83)
0.2888*** 0.0663*** (12.64) (
6.26) Quadratic terms for agricultural potential yields of groundnuts and marketshed fixed effects 26,178 0.344

26,178

26,178

26,178

N per sample: 414–443 # Samples: 3,900

0.0004 0.0000 12.421

Note: Sample includes all non-urban cells with positive groundnut potential yield, according to FAO/GAEZ. Column 1 presents coefficients from an OLS regression with ln(Groundnut production) as the dependent variable. Columns 2–4 present 2SLS regressions, with the first stage for ln(cost to market) and ln(conflict fatalities) in columns 2 and 3, respectively, and the second stage in column 4. Robust t-statistics are in parentheses. Column 5 presents spatial 2SLS estimates with bootstrapped standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

and the spatial 2SLS results in column 5 of the tables show that the coefficient on ln(cost to market) is negative and significant, implying that reducing the cost to market would increase the production of these 5 staple crops. 13 The results in the spatial model are highly statistically significant, and largely consistent with the 2SLS results, albeit a little smaller in magnitude. We also note that the elasticities on ln(cost to market) for the agricultural production estimates are significantly larger than that for the local GDP regressions. There are at least two reasons why we might expect this. First, it is possible that the agricultural sector disproportionately benefits from reducing transportation costs. In both regressions we are only examining rural locations, and it may be the case that other rural industries are not as reliant on gaining access to markets. Second, recall that the agricultural models use total production in tonnes as the dependent variable, while local GDP is measured in dollars. While increasing market access may lead to a large increase in total agricultural production, this may reduce prices, leading to a smaller, but still significant, increase in production value. 5. ESTIMATING IMPACTS OF ROADS ON FORESTS Figure 5.1 displays gridded deforestation estimates for the DRC with roads overlaid upon the map. The figure reveals a striking pattern of forest loss—deforestation has mainly occurred along the roads in our database and declines exponentially with distance from a road. In other cases, however,

9

deforestation is much less pronounced for reasons that are better captured in the regression estimates presented below. Drawing upon previous research, the empirical model incorporates seven critical determinants of forest clearing in road corridors: road quality, distance from the road, travel cost to the nearest market center, the agricultural opportunity value of the land, terrain elevation, legal protection status, and the incidence of violent conflict. To avoid spatial autocorrelation problems, the estimates use mean pixels cleared within grid cells whose distance from a road segment is aggregated to 2.7-km intervals on both sides of the segment. The estimating equation is specified as: lnðhit Þ ¼ a0 þ a1 lnðdij Þ þ a2 pi  lnðdij Þ þ a3 ln qi þ a4  lnðmi Þ þ a5 lnðci Þ þ a6 lnðei Þ þ a7 lnðvi Þ X þ bt yt þ it

ð5:1Þ

where hit is pixel cleared in grid cell i, road segment j, year t; dij is distance of cell i from road segment j; pi is the legal protection status of cell i; qi is the condition of road segment j; mi is either travel cost from cell i to the nearest urban center; ci is agricultural opportunity value of cell i; ei the elevation of cell i; vi the conflict incidence in cell i; yt a dummy variable with value 1 in year t and 0 otherwise; Ɛit an error term. Precise definitions of terms are relegated to the Data Definition Appendix for brevity. All variables are calculated as centroid values for 2.7-km grid cells.

Table 4.4. Effect of Transport Costs on Maize Production (1) OLS

(2) 2SLS

(3)

(4)

(5) Spatial 2SLS

Dependent Variable:

ln(Maize Production) ln(Cost to Market) ln(Conflict Fatalities) ln(Maize Production) ln(Maize Production)

ln(Cost to Market), USD


0.0515 (
0.83)
0.0322* (
1.66) 0.190*** (15.67) 0.0220*** (10.60) 1.665*** (2.60)
0.193*** (
2.93)

ln(Conflict Fatalities) ln(Population) ln(Population)^2 ln(Distance to Mine), km ln(Distance to Mine)^2, km


0.0049*** (
4.61)
0.0011*** (
9.38)
0.0782** (
2.55) 0.0066** (2.10)

0.2874*** (25.18) ln(Euclidean Distance), km 0.6233*** (53.28) ln(Distance to East Border), km 0.0512*** (8.88) Other variables included: Quadratic terms for agricultural potential

ln(Natural Path), hours

N R-sq Hansen J p-value Kleibergen–Paap LM p-value Kleibergen–Paap F-stat

26,914 0.268

26,914

0.0053 (0.54) 0.0004 (0.46) 0.6661*** (4.19)
0.0062 (
0.36)


0.968*** (
3.52)
0.979*** (
3.45) 0.188*** (11.82) 0.0221*** (9.64) 2.332*** (3.31)
0.202*** (
2.92)


0.655*** (
10.81)
0.732*** (
11.58) 0.181*** (9.58) 0.002*** (15.33) 1.25*** (10.51)
0.108*** (
10.37)

0.2221** (2.49)
1.050*** (
11.31)
0.3701*** (
8.99) yields of maize and marketshed fixed effects 26,914

26,914

N per sample: 414–443 # Samples: 3,900

0.3471 0.0000 13.43

Note: Sample includes all non-urban cells with positive maize potential yield, according to FAO/GAEZ. Column 1 presents coefficients from an OLS regression with ln(Maize production) as the dependent variable. Columns 2–4 present 2SLS regressions, with the first stage for ln(cost to market) and ln (conflict fatalities) in columns 2 and 3, respectively, and the second stage in column 4. Robust t-statistics are in parentheses. Column 5 presents spatial 2SLS estimates with bootstrapped standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

10

WORLD DEVELOPMENT

Figure 5.1. Forest clearing and road networks, 2000–2012. Source: Hansen et al. (2013) and authors’ calculations.

Eqn. (5.1) is distilled from numerous experiments that tested the interactions of distance to road with road quality, travel cost, agricultural opportunity value, elevation, and protection status. These revealed that road surface type has no significance for forest clearing, controlling for road condition, so the former is excluded from the final regressions. Simultaneity bias may be significant in this context, since forest clearing and road placement are jointly determined in a properly specified spatial economic model. Hence any measure of travel cost will be endogenous. To address estimation bias, Euclidean distance is used as an instrumental variable. Table 5.1 presents final estimates using several estimation techniques. The first two columns present OLS results for travel cost and Euclidean distance. Columns (3–6) present four results that employ Euclidean distance as an instrument for travel cost: standard 2SLS, GLS (IV) with covariance matrix

adjustments for road-specific error variances, GLS (IV) with standard errors corrected for spatial dependence (Conley, 1999), and robust regression (IV). The estimated coefficients in Table 5.1 have generally high significance, and their signs are consistent with prior expectations. We use the Conleycorrected GLS (IV) estimates for our discussion of results. The most critical variable for this exercise is distance from the road. In Table 5.1, the results for this variable are strong and consistent across estimators: Ceteris paribus (in GLS (IV)), forest clearing intensity declines 3.1% with each 10% increase in distance. The result for protection status indicates that the relationship between clearing intensity and distance steepens significantly in protected areas, suggesting that protected areas in the DRC are having some of the desired effects. The results for road quality are also consistent across specifications, and for numerous experiments that test interactions

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

11

Table 5.1. Deforestation Regression Results Dependent variable: ln(No. Pixels cleared) ln(Distance from Road) Protected area x ln(Distance from road) Road condition Ln(Cost to market)

(1) OLS

(2) OLS

(3) 2SLS

(4) GLS (IV)

(5) GLS (IV)(Conley)a

(6) Robust (IV)


0.298*** (29.97)
0.165*** (18.56) 0.544*** (17.56)
0.481*** (34.15)


0.309*** (30.43)
0.204*** (22.79) 0.455*** (13.76)


0.309*** (30.43)
0.204*** (22.79) 0.455*** (13.76)
0.952*** (23.54)


0.309*** (8.80)
0.204*** (2.91) 0.455 (1.57)
0.952*** (3.28)


0.309*** (18.28)
0.204*** (5.94) 0.455*** (3.25)
0.952*** (4.29)


0.296*** (32.77)
0.152*** (19.18) 0.513*** (17.49)
0.876*** (24.39)


0.023*** (2.63)
0.13*** (3.97) 0.029*** (9.96) 0.934*** (21.46) 1.269*** (29.15) 1.558*** (35.81) 1.826*** (41.97) 1.982*** (45.58) 2.154*** (49.52) 2.278*** (52.37) 2.462*** (56.60) 2.65*** (60.94) 2.746*** (63.14) 2.831*** (65.09) 6.747*** (25.94) 13,758


0.023 (0.37)
0.13 (0.59) 0.029 (1.23) 0.934*** (19.61) 1.269*** (22.76) 1.558*** (27.75) 1.826*** (32.71) 1.982*** (36.81) 2.154*** (38.55) 2.278*** (40.65) 2.462*** (44.19) 2.65*** (48.34) 2.746*** (48.95) 2.831*** (49.73) 6.747*** (4.07) 13,758


0.023 (0.61)
0.13 (0.71) 0.029*** (3.01) 0.934*** (19.65) 1.269*** (23.13) 1.558*** (35.00) 1.826*** (35.79) 1.982*** (35.41) 2.154*** (39.91) 2.278*** (38.87) 2.462*** (43.04) 2.65*** (41.58) 2.746*** (40.42) 2.831*** (41.06) 6.747*** (4.51) 13,758

0.004 (0.46)
0.238*** (8.18) 0.000 (0.1) 0.925*** (23.92) 1.246*** (32.24) 1.542*** (39.91) 1.81*** (46.86) 1.972*** (51.07) 2.145*** (55.55) 2.271*** (58.81) 2.456*** (63.62) 2.647*** (68.56) 2.738*** (70.92) 2.821*** (73.06) 7.255*** (31.42) 13,758

Euclidian distance to nearest urban center Land opportunity value Elevation Conflict intensity (1997–2007) D2002 D2003 D2004 D2005 D2006 D2007 D2008 D2009 D2010 D2011 D2012 Constant Observations R-squared


0.091*** (10.32)
0.066** (2.06) 0.016*** (5.52) 0.935*** (21.93) 1.269*** (29.78) 1.559*** (36.58) 1.827*** (42.87) 1.983*** (46.56) 2.155*** (50.59) 2.278*** (53.50) 2.462*** (57.81) 2.651*** (62.25) 2.747*** (64.50) 2.832*** (66.48) 4.81*** (23.39) 13,758 0.49


0.279*** (23.54)
0.023*** (2.63)
0.13*** (3.97) 0.029*** (9.96) 0.934*** (21.46) 1.269*** (29.15) 1.558*** (35.81) 1.826*** (41.97) 1.982*** (45.58) 2.154*** (49.52) 2.278*** (52.37) 2.462*** (56.60) 2.65*** (60.94) 2.746*** (63.14) 2.831*** (65.09) 4.806*** (22.07) 13,758 0.47

Note: Sample includes cells within 2.7 km of a road segment. Columns 1 and 2 present coefficients from an OLS regression with ln(No.Pixels cleared) as the dependent variable. Columns 3–5 present 2SLS regressions, which employ Euclidean distance and NHP as instruments for travel cost: standard 2SLS, GLS (IV) with covariance matrix adjustments for road-specific error variances, and robust regression (IV). t-statistics are in parentheses, *p < 0.10, ** p < 0.05, ***p < 0.01. a Standard errors corrected for spatial dependency. See Conley (1999).

with distance from the road. Road surface (earth, gravel, asphalt) never has a significant effect, 14 but the impact of road condition (1 = very poor; 2 = poor; 3 = fair; 4 = good) is large and highly significant. Neither road surface nor road condition interacts significantly with distance from the road. Among other regression variables, forest clearing is negatively related to travel cost to the nearest market center and elevation, and positively related to land opportunity value and conflict intensity. As expected, the IV results for travel cost to the nearest market differ markedly from the OLS results. The yearly dummy variables increase steadily in size, reflecting the rise in cumulative clearing, but the increments decrease markedly during the period.

To illustrate the implications of these estimates we simulate the effect of upgrading a road using parameters from the Kasese region of the DRC. To ensure consistency with the initial pattern of deforestation, we form the ratio [q = (f^1 )/(f^0 )] and multiply by actual clearing (f) to obtain the final prediction result [f^ = qf] (where (f^1 ),(f^o ) are predicted levels of clearing under the improved and unimproved road respectively). Figure 5.2 presents the results for the Kasese road segment as its condition improves from very poor (1) to good (4). With the road in very poor condition (value 1), 34% of previously forested land is cleared within 200 m of the road. Further away, clearing declines to 11% at 1 km, 7% at 2 km and 2% at 10 km. With further upgrading to good condition (value

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

12

WORLD DEVELOPMENT

Figure 5.2. Effect of road quality on forest clearing intensity. Source: authors’ calculations.

4), clearing at the four distances increases to 91%, 30%, 19%, and 6%. Two patterns are noteworthy in these results. First, upgrading from very poor to good produces near-complete deforestation within a narrow 1-km corridor straddling the road. Second, deforestation intensity falls rapidly as distance from the road increases. This finding has important policy implications. If an area under consideration is of high ecological significance, a relatively small deviation in the location of the road (2 km–10 km) could yield significant protective environmental benefits. 6. GRADIENTS OF BIODIVERSITY IMPACTS Not all forest land is of uniform ecological value, nor is it of uniform economic value. This section develops a variety of metrics to identify areas that are of high ecological value and at higher risk of degradation. 15 This approach has practical policy merit. Some road corridors will be built in areas of modest ecological concern, while others pass through areas of higher value. If a sufficiently robust and representative indicator of biodiversity significance is developed it can be used to

minimize ecological damage by favoring road improvements in areas of limited consequence. However, efforts to develop a single measure and concept of ecological value remain elusive and beset with difficulties of enumeration and measurement. Nor is there a consensus in the literature on the weights that should be assigned to different sub-components of an index. To address these issues, this study develops a composite index that encompasses the main concerns in the literature. Species density is clearly central to any measure of biodiversity significance. Spatial data on distribution among species classes is provided by IUCN and Birdlife International and summarized in Table 6.1. The Congo Basin is distinguished by the large number of birds and mammals that reside within its ecosystem. However, while species density is important, it is not the only issue or indicator of relevance. There are at least four other elements that are important. Endemism is another significant area of concern. Species that reside in very few areas (grid cells) will be more vulnerable to extirpation, ceteris paribus. This is captured through an index of endemism, which measures the percentage of each species’ range that is found in each grid cell. Total endemism for each grid cell –defined as the sum of its species endemism measures—assigns higher values to cells inhabited by species whose ranges are relatively limited. By implication, forest clearing in higher value cells may be particularly destructive for remaining critical habitat. Extinction risks arise for several reasons that are not captured by the endemism measure. To incorporate these factors, the threat status code assigned to each species by the IUCN is used with extinction probabilities using the methodology of Mooers et al. (2008). Table 5.1 tabulates conversions from Red List codes to normalized species weights, using four probability assignments. Three IUCN estimates are employed to derive measures of extinction probability over the next 50, 100, and 500 years. Biodiversity conservation is largely about protecting the level of genetic diversity on the planet. This suggests that a species on an isolated branch of a phylogenetic tree is rarer and hence more important than another with more common antecedents. Recent work by Isaac, Turvey, Collen, Waterman, and Baillie (2007) combines the IUCN extinction risk measure with a measure of each species’ isolation on a phylogenetic tree. 16 This is yet another indicator that is used and shown in the third column of Table 6.1.

Table 6.1. Normalized species aggregation weightsa Normalized Extinction Probabilities IUCN: Future Years

a b

IUCN Code

Status

Isaacb

50

100

500

CR EN VU NT LC

Critically Endangered Endangered Vulnerable Near Threatened Least Concern Rounded Weight Ratios CR:EN CR:VU CR:NT CR:LC

1.00000 0.50000 0.25000 0.12500 0.06250

1.00000 0.43299 0.05155 0.00412 0.00005

1.00000 0.66770 0.10010 0.01000 0.00010

1.00000 0.99600 0.39000 0.02000 0.00050

2 4 8 16

2 19 243 20,000

1 10 100 10,000

1 3 50 2,000

Data source: Mooers, Faith and Maddison (2008). From calculations by Mooers et al., based on Isaac et al. (2007).

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

13

Figure 6.1. Composite species–ecoregion index, Congo Basin countries. Source: authors’ calculations.

Finally, measures based on species vulnerability alone provide an incomplete accounting of ecological values and functions. A comprehensive measure would need to incorporate biomes and ecosystems. Using the World Wildlife Fund (WWF) classification of ecoregions 17 a vulnerability index is derived which measures the amount of an eco-region in a given area. The WWF ecoregions serve as a general proxy for distinctive plant, inspect, and animal species that are not represented in the range maps provided by IUCN and BirdLife International. 18 Since each of these indexes measure different issues they inevitably lead to very divergent ranking of priorities. Consider for instance the data in Table 6.1. Suppose there are two areas A and B. Area A contains 20,000 very diverse species, all in the ‘‘Least Concern” category, while area B contains only 2 species, but they are both ‘‘Critically Endangered Species”. Area A would receive a very low score by the IUCN extinction risk measures, since there are no species at ‘‘high” enough risk. But by the Isaacs measure it would be deemed much more important since there is greater genetic diversity of species, even when none are Critically Endangered. To be precise, by the IUCN 50-year extinction scores, each Critically Endangered species carries the weight of 20,000 Least Concerned species, while by the Isaac scores each Critically Endangered species is equivalent to 16 Least Concerned species. The differences are thus significant. To accommodate these diverse perspectives, a conservative strategy is adopted whereby priority is given to the index that generates the highest threat level. This is done by normalizing the indices for comparability using ranks measured as per-

centiles in each index and selecting the maximum index (risk) value as the risk measure for the cell. This approach gives parity to alternative vulnerability indicators and always picks the indicator that generates the highest threat level. Figure 6.1 illustrates the outcome for the Congo Basin countries. One striking feature is the blue/green (0–50) band that arcs from northern Cameroon to eastern DRC and back to southern DRC. Another is the prominent clustering of very high values in western Cameroon, along the border between Congo and the DRC, and along the eastern margin of the Basin. And finally there are the highly vulnerable ‘‘red strips” that identify the habitat of Critically Endangered species. 19 The most important and often neglected message from this exercise is the non-uniformity of ecological vulnerability across forested areas. This highlights the importance of going beyond simple measurement of forest loss to an assessment of the potential impact of that loss on biological diversity. Note, however, that the composite species-ecoregion index grades areas by their relative biodiversity value. However, this is not to say that the lowest graded areas have low value on an absolute, global scale. On the contrary, areas of relatively low value in the biodiversity-rich Congo Basin might well be high value relative to areas in other countries. 7. THE STAKES FOR VULNERABLE AREAS The purpose of this section is expositional. It joins the strands of the previous sections to show how priorities for road investment can be established and vulnerabilities to for-

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

14

WORLD DEVELOPMENT

Figure 7.1. Major urban center road linking network.

ests and biodiversity identified to avoid damage and conflict. The exercise entails combining the estimates of the benefits of road upgrading with predicted deforestation from road upgrading and the composite index of ecological vulnerability. The outcome is a highly varied pattern of economic benefits, deforestation levels, and biodiversity threats, suggesting wide scope for damage avoidance through prudent planning. 20 To demonstrate the approach, two simulations are described in this section. Since the analysis is carried out a high level of spatial of disaggregation (10 km  10 km), the simulations are illustrated using GIS maps for expositional ease. The first simulation considers the consequences of a project linking some of the provincial urban centers to Kinshasa—the national capital. The second simulation at a finer spatial scale estimates the effects of a smaller project in the environmentally fragile northeastern part of the country, near Virunga National Park.

Figure 7.1 illustrates the scope and scale of the project linking the main provincial urban centers (capitals) to Kinshasa. This project would involve 6,500 km of roads that traverse much of the country, and connect many areas which are currently only accessible by river or air travel. It is assumed that this network would be improved from its current quality, to ‘‘good” (paved) condition status, with several impassable missing links filled in. The baseline scenario (current quality) has only 20% of the network that is currently paved, and about 75% of the roads are in poor condition. To compute the change in GDP from the road improvement project, the regression estimates from Table 4.1 are used in the following formula: DGDP i ¼ gM  siM  y i ;where DGDP i is the total increase to local GDP in grid cell i, gM is the local GDP elasticity of transportation costs to the

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

15

Figure 7.2a and b. GDP Change and Forest Cover Loss from the Major Urban Center Road Project. Note: 7.2a (left) shows estimated changes in local GDP from the road improvement project, and 7.2 b (right) shows estimated additional deforestation. When calculating the local GDP increase simulation, the choice of coefficient used will not affect the figures displayed here. The map shows relative differences in local GDP. Different coefficients will change the absolute magnitude of the benefit, but not the relative differences. Source: Authors’ calculations.

local market, siM is the percentage change in transportation costs to the local market in cell i, and y i is baseline local GDP in cell i. The total increase in local GDP is then obtained by summing the increase in each grid cell. The aggregate increase in local GDP from the project linking the major urban centers is estimated to be between US$32.70 and US $55.55 million per year, depending on whether the coefficient from Table 4.1 column 4 (
0.141) or column 5 (
0.083) is used. This is a lower bound estimate. 21 In a similar manner, simulations were done which estimate the total deforestation due to the major urban center road improvement project. The estimates suggest that much of the additional deforestation will occur near the major cities of Kananga, Kisangani, Maniema, as well as South Kivu and Maniema provinces. A comparison of GDP and forest loss is presented in the Figures 7.2a and 7.2b. Figure 7.3 combines these to provide a clearer visual depiction and location of the economic and ecological impacts. Changes in local GDP and deforestation are overlaid to identify areas which would see the most benefits, and face the highest risks of loss. Areas in green are ‘‘pure benefit” regions, where local GDP gains are very significant, and deforestation increases are very low. Red areas are the riskiest regions, which are estimated to have very low local GDP gains, but significant deforestation as a result of the project. These are the regions which would be most beneficial to protect, given that there would be little loss in terms of economic activity, and there is a significant risk to deforestation. The intermediate zone is in yellow. A policy implication of this pattern of impacts is that areas of high concern are relatively few and well defined, on the

other hand the trade-off zones and low-hazard areas seem larger, suggesting scope for considerable win–wins for the economy and the environment. As was shown by the composite species–ecoregion index, not all forested areas are equally important. In order to further prioritize the areas that would be in most need of protection, the regions at risk of high deforestation are further dissected to reveal those that have the most ecological importance. This is accomplished by intersecting the composite species–ecoregion index with the simulated deforestation due to the road improvement project. This is shown in Figure 7.4. Note that the red, high-risk areas in Figure 7.4 are a subset of those in Figure 7.3. These red areas represent the regions that are most important to protect, while also having a low potential economic impact from the project. A benefit of this exercise is that it shows that the truly highrisk areas appear to be small when threats from the road are considered. It also suggests where conservation efforts ought to be directed. To illustrate the utility of this approach at finer spatial scales the same techniques are used to examine the costs and benefits of a much smaller road improvement project, situated around Virunga National Park. Perhaps the most notable aspect of Virunga National Park is that it is home to the extremely rare mountain gorillas that are listed as one of the most Critically Endangered species in the world. The hypothetical road project would improve a 525-km road which connects the city of Goma, situated just south of Virunga National Park, between the park and Lake Kivu, to Bunia, approximately 100-km north of the park, near Lake Albert. Despite being a very populated area (approximately 4.5 million Congolese live within a small area around the

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

16

WORLD DEVELOPMENT

Figure 7.3. Economic and Deforestation Benefit and Risk Assessment. Source: Authors’ calculations.

road), the current condition of the road is quite poor, and in many areas, impassable. The surrounding area has significant deposits of mineral wealth including gold and the rare-earth mineral coltan. The land also contains fertile soils, with theoretical maximum yields that are orders of magnitude greater than current agricultural yields. This road thus appears to be a major candidate for significant investments to spur economic activity. Nevertheless, road infrastructure development in this region may come with deep trade-offs. The land around the potential road project is heavily forested, and includes one of the world’s most important national parks. Virunga National Park was established in 1924 and was the first designated national park in Africa. Apart from the Critically Endangered mountain gorillas, Virunga hosts an immense variety of endemic biodiversity. Environmental factors aside, Virunga National Park has the potential to become one of the greatest tourist attractions on the continent if the conflict and security issues in eastern DRC could be resolved. Destroying this asset could thus extinguish a significant

source of future income for the country’s impoverished inhabitants. Given the immense tradeoffs that come with this project, this is an example of a project that would benefit from the analysis developed in this report. The benefits, in terms of the increase in local GDP, are calculated at the pixel level, and aggregated to arrive at a final range of $5.73 million–$31.9 million per year above the baseline, depending on whether one uses a local elasticity, or the national elasticity. Figure 7.5a shows the spatial distribution of these benefits. These are clustered around the road because the local GDP increase is the intersection of the baseline local GDP and the percentage change in transport costs, both of which are highly clustered around the road themselves. Multiplying these two together magnifies this clustering effect even further. Figure 7.5b shows the estimated annual deforestation that would occur due to the road improvement project. The biggest risks to deforestation are those regions which are nearest to the population centers and the improved road. This simulation shows that the areas that would be most stressed are near

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

17

Figure 7.4. High Ecological Risk Areas from Major Urban Road Project. Source: Authors’ calculations.

those near Lake Edward, the corridor between Goma and Rutshuru, and the corridor from Katwa, to Butembo, to Beni. As a final point, the estimated additional deforestation due to the project is layered on top of the current biodiversity index in Figure 7.6, to see which threatened areas have the most biodiversity, and are therefore worth the most to protect. The grid cells outlined in black are those in which deforestation is predicted to increase because of the project. Although this does not distinguish the intensity of deforestation, it allows one to compare the gradient of biodiversity within the areas affected, to identify areas of ecological vulnerability. It is clear that some of the regions with the highest ecological values also coincide with the regions predicted to experience the highest rate of deforestation from the project. The important conclusion is that this road project poses a significant risk to the forests and to high value biodiversity in the region. The estimates presented here suggest that a small deviation of this road may make an immense difference in generating more economic benefits while safeguarding vulnerable areas. This illustrates one possible use of the approach

described in the paper, which may assist in assuring more sustainable and less environmentally damaging paths to development. 8. CONCLUSIONS AND DISCUSSION Roads bring significant economic benefits and are essential for development, especially in densely populated rural areas that are unconnected to markets and economic activity. As a vast body of empirical research has established, roads are often also the precursors to deforestation and biodiversity loss. Conventional attempts to resolve this dilemma have relied on reactive approaches to limit damage. One common approach promoted by development agencies and bilateral donors involves the use of ‘‘environmental safeguard instruments” such as the Environmental Impact Assessment (EIA). An EIA typically seeks changes in structure and design of infrastructure to limit the level of damage. But this approach, while sometimes useful, has limits. Affordable

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

18

WORLD DEVELOPMENT

Figure 7.5a and b. GDP Change and Forest Cover Loss from the Virunga National Park Road Project. Note: 7.5a (left) shows estimated changes in local GDP from the road improvement project, and 7.5b (right) shows estimated additional deforestation. Source: Authors’ calculations.

Figure 7.6. High Ecological Risk Areas from Virunga National Park Road Project. Source: Authors’ calculations.

alternatives that prevent damage are not always available. More importantly there are concerns about the impartiality of the process. Since EIAs are financed and managed by developers and investors, there are fears that objectivity and impartiality of the process is compromised and rendered incentive incompatible (Wright, Dolman, Jasny, Parsons, Schiedek, & Young, 2013). Recognizing the limits of the EIA process, governments have relied on establishing protected areas to slow or limit the amount of development that can occur. This strategy too has had at best partial success. As with the EIA processes, attempts to restrict road improvements in areas with strong economic potential tend to fail because economic interests overwhelm the limited resources of conservation interests. Experience with managing such conflicts suggests the need for preemptive approaches that consider impacts at the very outset of the planning process. Drawing on a variety of disciplines—GIS, econometrics, and biology—this paper suggests how a set of tools can be developed and combined to identify areas of high ecological value and significant economic potential in order to steer development away from ecological hotspots toward areas where net impacts are more positive or benign. Using the DRC as an example, the exercise has highlighted the importance of considering not just levels of forest loss but also varied biodiversity attributes to prioritize investments and areas for protection. Finally it is worth noting that this paper has purposefully eschewed ascribing monetary values to biodiversity assets, recognizing the challenges of the task and the difficulties in gaining policy acceptance of the resulting estimates. Despite considerable advances in implementing contingent valuation surveys, there remains strong opposition to using estimates of the non-use benefits of biodiversity to guide economic and infrastructure decisions. For instance, non-use biodiversity benefits are ignored in the recent Natural Capital

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

Valuation Project citing methodological obstacles (Khan and Johnson, 2014). Global initiatives such as the Natural Capital Protocol also do not consider biodiversity values and instead focus on resources (such as water) where use values can be obtained. 22 Where non-use values are significant but

19

excluded, conventional cost-benefit assessments will be biased and inappropriate for evaluating environmental trade-offs. The techniques suggested in this paper provide one, albeit imperfect, approach that circumvents the need for valuing these elusive and typically non-monetizable attributes.

NOTES 1. There is very high endemism not only at the species level but also at the genus and even family levels in the Congo. On the other hand there is a lower level of species diversity (in aggregate) than the Amazon rain forests. The lowland forests of DRC contain around 10,000 plants, of which 30% are endemic, while the afro-montane forests contain around 4,000 species, of which 70% are endemic (including 2 endemic families). There are a host of endangered and charismatic mammals including the okapi, bongo, genet, gorilla, and bonobo. Many of the small primates and duikers are also unique to these forests. In addition to the endemic and famous Congo peacock the forests contain at least 5 bird families endemic to Africa. There are still discoveries of new species in the Congo. 2. Since roads are not randomly placed and usually connect nodes of economic opportunity, OLS estimates will be biased upward since it would capture the differentially higher opportunity of the connected regions. 3. Roads not appearing in AICD are assumed to be tertiary, unpaved, and of poor quality. Given the state of DRC’s road network, we believe this to be a very safe assumption. 4. See Highway Development and Management Model (HDM-4) Appendix for further detail. 5. This paper analyzes the combined effect of both large transport infrastructure (such as highways) and rural roads. Thus, it differs from Michaels (2008), Datta (2012), Faber (2014) and Banerjee, Duflo, and Qian (2012), that analyze the impact of large transport infrastructures, highways, and railways. It also differs from Jacoby and Minten (2009), Dorosh, Wang, You, & Schmidt (2009), Khandker and Koolwal (2011), Mu and van de Walle (2007) that analyze the impact of smaller rural roads. 6. See Conflict Kernel Function Appendix for more information. 7. When estimating crop production functions, the acro-ecological potential yield of the specific crop being estimated is included. When estimating local GDP, the acro-ecological potential yield of cassava, bananas/plantains, and groundnuts are all included as these are three of the most important crops in DRC in terms of total production value. 8. The methodology for determining which pixels of the Landscan dataset are urban areas is as follows. DRC’s urbanization rate as defined by the Central Intelligence Agency World Factbook was 34.3% in 2011 (see: https://www.cia.gov/library/publications/the-world-factbook/fields/ 2212.html). The total population in the Landscan dataset is approximately 68.8 million, implying an urban population of 23.6 million. The pixels with the largest number of people according to Landscan are marked as being urban pixels until the total number of people living in these marked pixels equals 23.6 million. These marked pixels are then omitted from the regressions. 9. While these IVs are desirable, they are not always feasible if data on historical paths are unavailable. In these cases, researchers often rely on straight line IVs. These variables are usually correlated with historical paths, given that the quickest path between two points is usually the path with the shortest length—the straight line. However, they cannot account

for the fact that the topography of the land may make traveling in a straight line impossible, or extra costly, making these IVs potentially quite weak. 10. Some recent examples of papers employing historical route IVs include: Duranton, Morrow, & Turner, 2014, which used routes from major exploration expeditions in the US between the 16th and 19th centuries as instruments for the US interstate highway system; GarciaLo´pez, Viladecans-Marsal, and Holl (2013), which used ancient Roman roads, among others, as exogenous sources of variation in Spain’s current highway system; and Martincus et al., 2013, which used the Incan road network to instrument for Peru’s current road infrastructure. 11. The causes of conflict are not always cut and dried, however. While those who create conflict may originate from areas with low economic opportunities, often conflict arises around areas with wealth, as there are more opportunities for theft. 12. The distance of 65km was determined by examining a spatial correlogram of the local GDP data. 13. We note that for some of the crops, the Hansen-J test rejects the null hypothesis, suggesting that the over-identifying restrictions are not valid. This is not surprising as one might not expect the same set of instruments to pass this test for several different models. Nevertheless, we do not rely on these results for simulations and merely test these models to ensure that they are not in disagreement with our local GDP estimates, which they are not. 14. We have tested the effect of road surface using categorical variables as well as a cardinal measure. 15. Spatial analysis of carbon stocks in the Congo and biodiversity habitats has largely confirmed that there are significant overlaps: Areas that store large amounts of biomass carbon may coincide with areas of biodiversity significance see Musampa, Mane, Betzky, Ravilious, & Miles (2012). 16. A phylogenetic tree is a branching tree diagram that traces the evolutionary descent of different species from a common ancestor. Species in sparse (isolated) branches of a phylogenetic tree are relatively unique, since they share common descent patterns with fewer other species. 17. Defined as ‘‘a large unit of land or water containing a geographically distinct assemblage of species, natural communities, and environmental conditions.” 18. The method for incorporating WWF ecoregions resembles this study’s treatment of species endemism. For the group of selected countries, the percent of total area accounted for by each ecoregion is computed. Its vulnerability index is then computed as the inverse of its area share and the appropriate index value is assigned to each pixel in the Congo Basin countries. This accounting assigns high values to pixels in smaller ecoregions, where clearing single pixels may pose more significant threats to biome integrity.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

20

WORLD DEVELOPMENT

19. Some of these are habitats for charismatic and better known species such as the bonobo monkeys (endemic to a narrow band in the DRC), the African forest elephants, gorillas, and chimpanzees. 20. In fact the correlation between the measures of clearing and ecological vulnerability is close to zero (q =
0.0356). 21. Keeping in mind that this is estimated using a partial equilibrium framework, and that these benefits are only a subset of the total benefits to reducing transportation costs (other benefits include those stemming from improved transport between cities, increased access to multiple cities rather than solely the cheapest one, and better access to ports), this estimate is likely a very conservative, minimum benefit. 22. http://www.sustainablebrands.com/news_and_views/new_metrics/sustainable_brands/natural_capital_protocol_will_demystify_business_value 23. Why Nations Fail. Acemoglu and Robinson (2013). 24. ‘‘The wealth of Africa, The kingdom of Kongo”. The British Museum 2011. 25. Congo Free State, 1885–1908 (http://www.yale.edu/gsp/colonial/belgian_congo/).

26. Carte du Congo Belge / e´dite´e par l’Office de Publicite´, anciens e´tablissements J. Lebe`gue & Cie. - E´diteurs, Bruxelles (1896). Stored at the Library of Congress and downloaded from http://www.wdl.org/en/item/ 59/ 27. Uchida, H. and Nelson, A. Agglomeration Index: Toward a New Measure of Urban Concentration. Background paper for the World Bank’s World Development Report 2009. 28. A raster is a geographic map with information (e.g., elevation, landcover) in a matrix of pixels. 29. http://www2.jpl.nasa.gov/srtm/ 30. ORNL is a multiprogram science and technology laboratory managed for the U.S Department of Energy by UT-Battelle,LLC. 31. ‘‘Three presentations on geographical analysis and modeling nonisotropic geographic modeling speculations on the geometry of geography global spatial analysis”. National Center for Geographic Information and Analysis. Technical Report 93–1. February 1993. 32. http://www.princeton.edu/achaney/tmve/wiki100k/docs/Walking. html

REFERENCES Abernethy, K. A., Coad, L., Taylor, G., Lee, M. E., & Maisels, F. (2013). Extent and ecological consequences of hunting in Central African rainforests in the twenty-first century. Philosophical Transactions of the Royal Society B: Biological Sciences, 368(1625), 20120303. Acemoglu, D., & Robinson, J. A. (2013). Why nations fail: The origins of power, prosperity, and poverty. Crown Business. Agarwal, D. K., Gelfand, A. E., & Silander, J. A. (2002). Investigating tropical deforestation using two-stage spatially misaligned regression models. Journal of agricultural, biological, and environmental statistics, 7(3), 420. Ali, R., Barra, A. F., Berg, C. N., Damania, R., Nash, J. D., & Russ, J. D. (2015). Infrastructure in conflict-prone and fragile environments: evidence from the Democratic Republic of Congo (No. 7273). World Bank. Angelsen, A. (2007). Forest cover change in space and time: Combining the von Thu¨nen and forest transition theories (Vol. 4117). World Bank Publications. Aschauer, D. A. (1989). Is public expenditure productive?. Journal of Monetary Economics, 23(2), 177–200. Banerjee, A., Duflo, E., & Qian, N. (2012). On the road: Access to transportation infrastructure and economic growth in China (No. w17897). National Bureau of Economic Research. Barbier, E. B., & Burgess, J. C. (1996). Economic analysis of deforestation in Mexico. Environment and Development Economics, 1(2), 203–239. Chen, X., & Nordhaus, W. D. (2011). Using luminosity data as a proxy for economic statistics. Proceedings of the National Academy of Sciences, 108(21), 8589–8594. Conley, T. G. (1999). GMM estimation with cross sectional dependence. Journal of Econometrics, 92(1), 1–45. Cropper, M., Puri, J., Griffiths, C., Barbier, E. B., & Burgess, J. C. (2001). Predicting the location of deforestation: The role of roads and protected areas in North Thailand. Land Economics, 77(2), 172–186. Damania, R., Berg, C., Russ, J., Federico Barra, A., Nash, J., & Ali, R. (2017). Agricultural technology choice and transport. American Journal of Agricultural Economics, 99(1), 265–284. Damania, Richard., Milner-Gulland, E. J., & Crookes, D. J. (2005). A bioeconomic analysis of bushmeat hunting. Proceedings of the Royal Society of London B: Biological Sciences, 272(1560), 259–266. Datta, S. (2012). The impact of improved highways on Indian firms. Journal of Development Economics, 99(1), 46–57. De Sy, V., Herold, M., Achard, F., Beuchle, R., Clevers, J. G. P. W., Lindquist, E., et al. (2015). Land use patterns and related carbon losses

following deforestation in South America. Environmental Research Letters, 10(12), 124004. DeFries, R. S., Rudel, T., Uriarte, M., & Hansen, M. (2010). Deforestation driven by urban population growth and agricultural trade in the twenty-first century. Nature Geoscience, 3(3), 178–181. Deveny, A., Nackoney, J., Purvis, N., Gusti, M., Kopp, R. J., Madeira, E. M., ... Obersteiner, M. (2009). Forest carbon index. Washington DC, USA: Resources for the Future (RFF). Doll, C. H., Muller, J. P., & Elvidge, C. D. (2000). Night-time imagery as a tool for global mapping of socioeconomic parameters and greenhouse gas emissions. AMBIO: A Journal of the Human Environment, 29 (3), 157–162. Donaldson, D. (2010). Railroads of the Raj: Estimating the impact of transportation infrastructure (No. w16487). National Bureau of Economic Research. Dorosh, P., Wang, H. G., You, L., & Schmidt, E. (2009). Crop Production and Road Connectivity in Sub-Saharan Africa: A Spatial Analysis. Africa Infrastructure Country Diagnostic Working Paper, 19. Duelli, P., & Obrist, M. K. (2003). Biodiversity indicators: The choice of values and measures. Agriculture, Ecosystems & Environment, 98(1), 87–98. Duranton, G., Morrow, P. M., & Turner, M. A. (2014). Roads and Trade: Evidence from the US. The Review of Economic Studies, 81(2), 681–724. Ebener, S., Murray, C., Tandon, A., & Elvidge, C. C. (2005). From wealth to health: Modelling the distribution of income per capita at the subnational level using night-time light imagery. International Journal of Health Geographics, 4(1), 1. Elvidge, C. D., Imhoff, M. L., Baugh, K. E., Hobson, V. R., Nelson, I., Safran, J., et al. (2001). Night-time lights of the world: 1994–1995. ISPRS Journal of Photogrammetry and Remote Sensing, 56(2), 81–99. Emran, M. S., & Hou, Z. (2013). Access to markets and rural poverty: Evidence from household consumption in China. Review of Economics and Statistics, 95(2), 682–697. Ewers, R. M., Laurance, W. F., & Souza, C. M. (2008). Temporal fluctuations in Amazonian deforestation rates. Environmental Conservation, 35(04), 303–310. Faber, B. (2014). Trade integration, market size, and industrialization: Evidence from China’s National Trunk Highway System. The Review of Economic Studies, rdu010. FAO and IIASA (Food and Agriculture Organization of the United Nations and the International Institute for Applied Systems Analysis).

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH (2000). ‘‘Global Agro-Ecological Zones (GAEZ)”. FAO, Rome; and IIASA, Laxenberg, Austria. FAO, Faostat (2017). Value of Agricultural Production Retrieved fromhttp://www.fao.org/faostat/en/#data/QV, . Fisher, B. (2010). African exception to drivers of deforestation. Nature Geoscience, 3(6), 375. ` ., Viladecans-Marsal, E., & Holl, A. (2013). SuburGarcia-Lo´pez, M. A banization and highways: When the Romans, the Bourbons and the first cars still shape Spanish cities. Geist, H. J., & Lambin, E. F. (2002). Proximate causes and underlying driving forces of tropical deforestation: Tropical forests are disappearing as the result of many pressures, both local and regional, acting in various combinations in different geographical locations. BioScience, 52(2), 143–150. Ghosh, Tilottama et al. (2010). Shedding light on the global distribution of economic activity. The Open Geography Journal, 3.1, 148–161. Goeschl, T., & Igliori, D. C. (2006). Property rights for biodiversity conservation and development: Extractive reserves in the Brazilian Amazon. Development and Change, 37(2), 427–451. Graesser, J., Aide, T. M., Grau, H. R., & Ramankutty, N. (2015). Cropland/pastureland dynamics and the slowdown of deforestation in Latin America. Environmental Research Letters, 10(3), 034017. Groves, C. R., Jensen, D. B., Valutis, L. L., Redford, K. H., Shaffer, M. L., Scott, J. M., et al. (2002). Planning for Biodiversity Conservation: Putting Conservation Science into Practice: A seven-step framework for developing regional plans to conserve biological diversity, based upon principles of conservation biology and ecology, is being used extensively by the nature conservancy to identify priority areas for conservation. BioScience, 52(6), 499–512. Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S. A., Tyukavina, A., et al. (2013). High-resolution global maps of 21stcentury forest cover change. Science, 342(6160), 850–853. HarvestChoice. (2012). ‘‘Spatial Allocation of Agricultural Production.” International Food Policy Research Institute, Washington, DC, and University of Minnesota, St. Paul, MN. http://harvestchoice.org/node/ 2248. Henders, S., Persson, U. M., & Kastner, T. (2015). Trading forests: Landuse change and carbon emissions embodied in production and exports of forest-risk commodities. Environmental Research Letters, 10(12), 125012. Henderson, J. V., Storeygard, A., & Deichmann, U. (2017). Has climate change driven urbanization in Africa?. Journal of Development Economics, 124, 60–82. Henderson, J. V., Storeygard, A., & Weil, D. N. (2012). Measuring economic growth from outer space. The American Economic Review, 102(2), 994–1028. Hosonuma, N., Herold, M., De Sy, V., De Fries, R. S., Brockhaus, M., Verchot, L., et al. (2012). An assessment of deforestation and forest degradation drivers in developing countries. Environmental Research Letters, 7(4), 044009. Ihori, T., Doi, T., & Kondo, H. (2001). Japanese fiscal reform: Fiscal reconstruction and fiscal policy. Japan and the World Economy, 13(4), 351–370. Isaac, N. J., Turvey, S. T., Collen, B., Waterman, C., & Baillie, J. E. (2007). Mammals on the EDGE: Conservation priorities based on threat and phylogeny. PLoS ONE, 2(3), e296. Jacoby, H. G., & Minten, B. (2009). On measuring the benefits of lower transport costs. Journal of Development Economics, 89(1), 28–38. Khandker, S. R., & Koolwal, G. B. (2011). Estimating the long-term impacts of rural roads: A dynamic panel approach. Kinzig, A. P., & Harte, J. (2000). Implications of endemics–area relationships for estimates of species extinctions. Ecology, 81(12), 3305–3311. Klein Goldewijk, K., Beusen, A., & Janssen, P. (2010). Long-term dynamic modeling of global population and built-up area in a spatially explicit way: HYDE 3.1. The Holocene, 20(4), 565–573. LandScan (2006). Global Population Database (2006 release). Oak Ridge, Tennessee: Oak Ridge National Laboratory, http://www. ornl.gov/landscan/. Laurance, W. F., Nascimento, H. E., Laurance, S. G., Andrade, A., Ribeiro, J. E., Giraldo, J. P., ... D’Angelo, S. (2006). Rapid decay of tree-community composition in Amazonian forest fragments. Proceedings of the National Academy of Sciences, 103(50), 19010–19014.

21

Matos, G. R., Miller, L. D., & Barry, J. J. (2015). Historical global statistics for mineral and material commodities (No. 896). US Geological Survey. Mayaux, P., Pekel, J. F., Descle´e, B., Donnay, F., Lupi, A., Achard, F., et al. (2013). State and evolution of the African rainforests between 1990 and 2010. Philosophical Transactions of the Royal Society B, 368 (1625), 20120300. Mellander, C., Lobo, J., Stolarick, K., & Matheson, Z. (2015). Night-Time Light Data: A Good Proxy Measure for Economic Activity?. PLoS ONE, 10(10), e0139779. Michaels, G. (2008). The effect of trade on the demand for skill: Evidence from the interstate highway system. The Review of Economics and Statistics, 90(4), 683–701. Mu, R., & Van de Walle, D. (2007). Rural roads and poor area development in Vietnam (No. WPS4340). World Bank. Musampa Kaungandu, C., Mane, Lola Amani, Betzky, P., Ravilious, C., & Miles, L. (2012) Mapping potential biodiversity benefits from REDD+. The DRC. Prepared by UNEP-WCMC, Ministry of Environment DRC and Satellite Observatory for Central African Forests. Mooers, Arne Ø., Faith, Daniel P., & Maddison, Wayne P. (2008). Converting endangered species categories to probabilities of extinction for phylogenetic conservation prioritization. PloS one, 3(11), e3700. Nelson, A., & Chomitz, K. M. (2009). Protected area effectiveness in reducing tropical deforestation. A global analysis of the impact of protection status. Noss, R. F. (1990). Indicators for monitoring biodiversity: A hierarchical approach. Conservation Biology, 4(4), 355–364. Raleigh, C., Linke, A., Hegre, H., & Karlsen, J. (2010). Introducing acled: An armed conflict location and event dataset special data feature. Journal of Peace Research, 47(5), 651–660. Richards, D. R., & Friess, D. A. (2016). Rates and drivers of mangrove deforestation in Southeast Asia, 2000–2012. Proceedings of the National Academy of Sciences, 113(2), 344–349. Rudel, T. K. (2013). The national determinants of deforestation in subSaharan Africa. Philosophical Transactions of the Royal Society B, 368 (1625), 20120405. Rudel, T. K., Defries, R., Asner, G. P., & Laurance, W. F. (2009). Changing drivers of deforestation and new opportunities for conservation. Conservation Biology, 23(6), 1396–1405. Russ, J., Berg, C., Damania, R., Barra, A. F., Ali, R., & Nash, J. (2017). Evaluating Transport Infrastructure Projects in Low Data Environments: An Application to Nigeria. The Journal of Development Studies, 1–20. Shrestha, S. A. (2012). Access to the north-south roads and farm profits in rural Nepal. University of Michigan. Sutton, P. C., Elvidge, C. D., & Ghosh, T. (2007). Estimation of gross domestic product at sub-national scales using nighttime satellite imagery. International Journal of Ecological Economics & Statistics, 8 (S07), 5–21. Tobler, W. (1993). Three presentations on geographical analysis and modeling. NCGIA. UNESCO (2010). World Heritage in the Congo Basin. New York: UNESCO. Vance, C., & Geoghegan, J. (2002). Temporal and spatial modelling of tropical deforestation: A survival analysis linking satellite and household survey data. Agricultural Economics, 27(3), 317–332. von Uexkull, N., Croicu, M., Fjelde, H., & Buhaug, H. (2016). Civil conflict sensitivity to growing-season drought. Proceedings of the National Academy of Sciences, 201607542. World Bank, World Development Indicators (2017). Agriculture, value added (% of GDP) Retrieved fromhttp://data.worldbank.org/indicator/NV.AGR.TOTL.ZS, . Wright, A. J., Dolman, S. J., Jasny, M., Parsons, E. C. M., Schiedek, D., & Young, S. B. (2013). Myth and momentum: A critique of environmental impact assessments. Zikri, M. (2009). An econometric model for deforestation in Indonesia. Working Paper in Economics and Development Studies, No. 200903. Center for Economics and Development Studies, Department of Economics, Padjadjaran University. July.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

22

WORLD DEVELOPMENT

APPENDIX A. HIGHWAY DEVELOPMENT AND MANAGEMENT MODEL (HDM-4) The Highway Development Management Model (HDM-4) considers several different variables in order to estimate the cost of traveling along each segment of the road network. The data used for the estimates used in this paper was collected specifically for DRC, to best characterize the transportation conditions one would find there. In order to estimate the unit cost (in ton per km), the cost of transporting a vehicle with an average weight of 25 tonnes, one kilometer, was first estimated. The unit cost per ton-km was derived from the costs per vehicle using a factor of 15 ton per vehicle (average net weight). This factor was obtained based on the assumption of a 30 ton gross vehicle weight, with a 10 ton tare weight and a 75% loading factor. A.1 Characterization of network type and terrain The road network of DRC includes three classes of roads: primary, secondary, and tertiary. Average vehicle speed and width of the main carriage road were used to characterize the differences among network types as follows: Paved Road Speed (km/h) by Network & Condition Road Condition

Primary 7m

Secondary 6m

Tertiary 5m

Flat Rolling Mountainous

100 80 60

80 70 50

70 60 40

Unpaved Road Speed (km/h) by Network & Condition Primary 7m

Road Condition

Secondary 6m

Tertiary 5m

80 70 60 Flat Rolling 60 50 40 Mountainous 40 30 20 where terrain type is defined using the following concepts and road geometry parameters:  Flat. Mostly straight and gently undulating  Rolling. Bendy and gently undulating  Mountainous. Winding and gently undulating

Terrain Type

Number Rise & Rise & Horizontal Super_ Fall Fall Curvature elevation (m/km) (#) (deg/km) (%)

Flat 10 Rolling 15 Mountainous 20

2 2 3

15 75 300

2.5 3.0 5.0

A.2 Characterization of network type and condition The International Roughness Index IRI (m/km) was used to define the differences in road condition by network as follows: Paved Road IRI (m/km) by Network & Condition Road Condition

Primary 7m

Secondary 6m

Tertiary 5m

Good Fair Poor

2 5 8

3 6 9

4 7 10

Unpaved Road IRI (m/km) by Network & Condition Primary Secondary Tertiary Road Condition 7m 6m 5m 6 8 10 Good Fair 12 13 14 Poor 16 18 20 Finally, using these parameters above, a final cost per ton-km for each road type is estimated ($/ton/km): Paved FLAT Road Condition

Primary

Secondary

Tertiary

Good Fair Poor Paved Rolling

0.1174 0.1226 0.1286

0.1192 0.1264 0.1299

0.1237 0.1293 0.1349

Road Condition

Primary

Secondary

Tertiary

Good 0.1190 Fair 0.1241 Poor 0.1305 Paved Mountainous

0.1191 0.1268 0.1315

0.1231 0.1302 0.1367

Road Condition

Primary

Secondary

Tertiary

Good Fair Poor Unpaved Flat

0.1283 0.1333 0.1410

0.1292 0.1318 0.1391

0.1312 0.1382 0.1449

Road Condition Good Fair Poor Unpaved Rolling

Primary 0.1401 0.1622 0.1976

Secondary 0.1463 0.1755 0.2133

Tertiary 0.1559 0.1901 0.2290

Road Condition

Primary

Secondary

Tertiary

Good 0.1348 Fair 0.1638 Poor 0.1991 Unpaved Mountainous

0.1453 0.1771 0.2147

0.1588 0.1921 0.2305

Road Condition

Primary

Secondary

Tertiary

Good Fair Poor

0.1390 0.1681 0.2014

0.1570 0.1857 0.2186

0.1806 0.2091 0.2379

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

APPENDIX B. NATURAL–HISTORICAL PATH Since the early arrival of the Portuguese Mariner Diego Ca˜o in 1483, the Congolese (Kingdom of Kongo at that time) has had cultural, social and economic connections with Europe. Western religions, literacy, the wheel, the plow, the gun and many other technologies were quickly adopted by the Congolese (Acemoglu & Robinson, 2013). 23 All these came at

23

very large expense: one of the principally traded goods in exchange were slaves 24. As their contact deepened other types of goods were introduced such as ivory, rubber, copper, diamonds, raffia cloth, and pottery among other natural resources. The European trade was based in the coastal cities of Sonyo and Pinda so it required an extensive trade network toward the eastern part of the country (primarily near present day Kivu and Katanga

Figure A.1. Terrain roughness and land cover circa 1900. Source: Authors’ calculations using NASA and ORNL.

Figure A.2. Walking time in DRC given land topography (hours per pixel). Source: Authors’ calculations.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

24

WORLD DEVELOPMENT

provinces) where much of the mineral deposits and other natural resources were mined. Fueled by the industrial revolution and new inventions such as the inflatable rubber tubes, the demand for goods increased dramatically. 25 By the end of the 19th century the Congo was a personal possession of King Leopold II (not an official Belgian Colony). The king was engaged in a vigorous publicity campaign aimed at convincing the other European powers to recognize the legitimacy of his rule, a difficult task in view of the notorious brutality of his administration in Africa. One of the products of King Leopold’s ‘‘Office of Publicity” is a very detailed ‘‘Carte du Congo Belge” 26 (Map of the Belgian Congo), which includes caravan routes and existing and projected railways (see figure A.4). This map, which shows the transport network constructed to move slaves, ivory and mineral resources between the interior of Congo and the coastal harbors, is the main input used to construct the natural–historical instrumental variable. The natural–historical IV is constructed by merging two sources: the historical caravan route map described above, and a natural walking path map calculated for this study. The natural walking path, or natural path, is created by estimating the time minimizing route a pedestrian would travel over land, absent the benefits of a road network. The subsequent section details how the natural path network was calculated. B.1 Natural path walking time calculation The first step is creating a GIS cost surface model that accounts for all the traffic off-road or outside the caravan network for the year 1900. For that purpose, this report followed a similar approach to that which was used to construct a global map of accessibility in the World Bank’s World Development Report 2009 Reshaping Economic Geography (Uchida and Nelson 2009). 27 The surface model, or off-path frictionsurface raster, 28 is a grid where each pixel contains the estimated time required to cross that pixel by walking. To create this raster two basic layers are combined: terrain slope and land cover. The slope raster was calculated from NASA’s Shuttle Radar Topography Mission (SRTM) 29 Digital Elevation Models (DEMs) with a resolution of 90 m. Even though the original topography data was obtained in February 2000, it was assumed that there has not been any drastic change in DRC’s terrain in the 20th century. Therefore, it is safe to assume that the SRTM 90-meters dataset provides a fairly good representation of the elevation terrain circa 1900. Land cover data is far more challenging. In the last few years, with the surge of remote sensing technology, several land cover and land use datasets have been created. These high-resolution datasets are a very accurate representation of the current state of physical material at the surface of the earth. However, these cannot be used in the analysis because they are not a good representation of the land cover for year 1900. Land has changed rapidly in the last 100 years in DRC: deforestation, open pit mining and urbanization in some cities have drastically transformed the surface. Therefore, other datasets were used. The Oak Ridge National Laboratory (ONRL) 30 has developed a Historical Land Cover and Land Use data estimate (Klein Goldewijk, Beusen, & Janssen, 2010). This dataset describes historical land use changes over a 300-year historical period (1700–1990) and was modeled based on a deep understanding of the global environment, historical statistical inventories on agricultural land (census data, tax records, land surveys, etc), and different spatial analysis techniques. A shortcoming of this dataset is

that the resolution is approximately 55 km per cell, making a clear tradeoff between space (resolution) and time (representative for 1900). Given the importance of obtaining an accurate picture of the historical land cover, this low resolution, but better account of the land surface types circa year 1900 was chosen (i.e., ORNL Historical Land Cover) over the better resolution but newer data. Figure A.1 displays the terrain roughness (left) and land cover (right) for DRC, circa 1900. The off-path friction-surface raster is created by combining the land topography raster and the land cover map. This is based on the guiding assumption that all travel in 1900 was on foot and walking speed is therefore determined by the land cover class and slope. The typical velocity of a hiker when walking on uneven or unstable terrain is 1 h for every 4 km (4 km/h) and diminishes on steeper terrain. A hiking velocity equation 31 (Tobler, 1993) was used to reflect changes in travel speed as a function of trail slope: W = 6 * exp(
3.5 * |S + 0.05|)where W is the hiking velocity in km/h and S is the slope or gradient. By applying the speed formula, he time it required to cross 1 pixel (92.5 m) was computed. In this way, the time (hours) that it takes to walk through any pixel— only taking into account the topography—was calculated, as shown below in Figure A.2. Note that the more mountainous regions of DRC near the Kivu provinces in the east have significantly higher walking times Figure A.3. Next, a delay factor to account for effect of walking through different land classes was estimated. The historical land cover raster resolution was changed from half degree to 90 m and each class was assigned a speed reducing factors according to the following table: Class # Biome Type 0 Oceans/Water 1 Cultivated land 2 Pasture/land used for grazing 5 Ice 6 Tundra 7 Wooded tundra 8 Boreal forest 9 Cool conifer forest 10 Temperate mixed forest 11 Temperate deciduous forest 12 Warm mixed forest 13 Grassland/Steppe 14 Hot desert 15 Scrubland 16 Savanna 17 Tropical woodland 18 Tropical forest 19 No data over land (e.g., Antarctica)

Delay Factor N/A 1.00 1.00 1.33 1.00 1.00 1.17 1.00 1.17 1.33 1.17 1.00 1.00 1.00 1.00 1.33 1.67 N/A

Source: Own calculations based on Uchida, H. and Nelson, A. (2009). Lastly, walking travel speed (slope variable) was multiplied by the delay factor (land cover variable) to obtain the off-path friction-surface raster that models the time that it takes to walk 92.5 m anywhere in DRC circa 1900. B.2 Historical path creation The second step was to digitalize a historical map of DRC to create shapefile that could be added as a layer for spatial analysis. A mix of image manipulator (open source GIMP—http:// www.gimp.org/) and GIS software (ESRI’s ArcGis Desktop)

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

25

Figure A.3. Final natural–historical path raster (hours per pixel). Source: Authors’ calculations.

Figure A.4. DRC historic trade routes circa 1896 and digitally manipulated map with routes. Source: Carte du Congo Belge/e´dite´e par l’Office de Publicite´, anciens e´tablissements J. Lebe`gue & Cie. - E´diteurs, Bruxelles (1896).

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

26

WORLD DEVELOPMENT

Figure A.5. Historical caravan route shapefile. Source: Authors’ calculations.

Figure A.6. Travel time to Kinshasa in 1900. Source: Authors’ calculations.

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001

THE ROAD TO GROWTH: MEASURING THE TRADEOFFS BETWEEN ECONOMIC GROWTH

were used to separate the routes from other map features and then digitize the map. Figure A.4 shows the historical map which was then converted into the shapefile shown in Figure A.5. Then the newly created shapefile was converted into a raster with a resolution of 92.5 m to match that of the natural path friction-surface raster. The pixel value assigned to every cell where there is a caravan route passing through is approximately 0.02 h or 1.2 min. This was arrived at by assuming the caravan route travel speed at 5 km/h 32; equivalent to the human average walking speed on a stable terrain. B.3 Cost-distance function: calculating travel time The third step was to merge the off-path and the caravan route friction surfaces. ArcGIS Desktop’s tool MERGE was used to combine the two rasters into a single one where the order of the input defines the order of precedence, in this case the caravan routes overlay the off-path walking. Then the friction surface was obtained to model the time that it takes to move around the entirety of DRC around the year 1900, taking into account terrain, land cover type, and transport infrastructure. To create the final variable, which was used as an instrumental variable in this study, the time that it takes to travel on foot from each pixel in the study area to different selected cities or target destinations, was estimated. ArcGIS Cost distance tools were used to calculate, for each pixel, the least cumulative amount of time it takes to walk to a specified locations (market). The algorithm utilizes the node/link cell representation, whereby the center of each cell is considered a node and each node is connected to its adjacent nodes by multiple links. Every link has an impedance derived from the costs (measured in units of time) associated with the cells from the natural path friction cost surface and from the direction of movement through the cells. See Figure A.6 for an example of a raster measuring travel time from each point to Kinshasa. The creation of a least cumulative cost raster was replicated for each of DRC’s 57 selected cities and then cell values from

27

the travel time raster at SPAM locations was retrieved. As a result an origin/destination travel time matrix of 27,500 rows (number of pixels in DRC) and 57 columns (selected cities) is obtained. Finally, this dataset was compared with the current travel cost dataset, and the appropriate city for each pixel was selected for the econometric model. B.4 Data definitions Hansen pixels cleared: per 2.7 km grid cell, the number of 30 m pixels identified by Hansen et al. (2013) as cleared, by year, for 2000–12. Distance from road segment: distance from the centroid of each grid cell to the nearest road, calculated in ArcGIS 10. Legal protection status: 1 if the parcel is in a protected area identified by the World Database on Protected Areas (WDPA); 0 otherwise. The WDPA shapefile has been downloaded from http://www.protectedplanet.net/. Condition of the road segment: Surface types are earth, gravel and asphalt. Road conditions are identified as good, fair, poor and very poor. We have translated each variable into ordered cardinal values for estimation: Surface: earth (1), gravel (2), asphalt (3); Condition: very poor (1), poor (2), fair (3), good (4). Travel cost: cost from a grid cell’s mean point to the nearest urban center with a population of 50,000 or greater. Raster resolution: 0.0083 decimal degrees. Agricultural opportunity value: mean value for a grid cell, calculated from the high-resolution global grid developed by Deveny et al. (2009). Raster resolution: 0.0025 decimal degrees. Elevation: Average elevation for a grid cell, calculated from the CGIAR-SRTM dataset (3 s resolution), aggregated to 30 s resolution by DIVA-GIS (http://www.diva-gis.org/gdata). Conflict incidence: Armed conflict fatalities per unit area, 1997–2007, calculated by Ali et al. (2015) at 0.017 decimal degrees resolution from data in the Armed Conflict Location Events Dataset (ACLED) (Raleigh et al., 2010).

Available online at www.sciencedirect.com

ScienceDirect

Please cite this article in press as: Damania, R. et al. The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction, World Development (2017), http://dx.doi.org/10.1016/j.worlddev.2017.06.001