报告：超过10亿人生活在贫困地区.docx

Poverty hotspots and the correlatesof subnational developmentRaj M. Desai12 Homi Kharas Walsh School of Foreign Service and Department of Government, 3700 O Street, NW, Washington, DC 20057. The Brookings Institution, 1775 Massachusetts Avenue, NW, Washington, DC 20036. Selen Ozdogan2December 2020Global working paper #149AbstractEconomic prosperity is unevenly distributed across geography, even within national boundaries. As national incomes converge, many subnational areas within countries show widening disparities. Much of the evidence of subnational growth is hampered by inadequate attention to the spatial clustering of economic development. We seek to explain the determinants of subnational growth by taking into account possible neighborhood and sp川over effects whereby growth and development are influenced by growth rates in proximate geographic areas. Using data from around 3,000 first-level, subnational areas across 169 countries, we find that spatial autocorrelation is a critical factor in explaining growth at the subnational level. We also find that certain characteristics of these areas affect growth independently of national economic policy, including soil suitability for agriculture and malaria ecologies. We also show that legacies of conflict exert a consistent, negative effect on subnational growth. Our findings carry implications for identifying and for spatial targeting of poverty hotspots.AcknowledgementsThe authors thank Tarek Ghani, Brad Parks, and participants in the uNo One Left Behind" seminar at the Brookings Institution.The Brookings Institution is a nonprofit organization devoted to independent research and policy solutions. Its mission is to conduct high-quality, independent research and, based on that research, to provide innovative, practical recommendations for policymakers and the public. The conclusions and recommendations of any Brookings publication are solely those of its author(s), and do not reflect the views of the Institution, its management, or its other scholars.Brookings gratefully acknowledges the program support provided by the Bill & Melinda Gates Foundation.Brookings recognizes that the value it provides is in its commitment to quality, independence, and impact. Activities supported by its donors reflect this commitment.areas experienced faster growth. Meanwhile, tropical temperatures that facilitate the reproduction of Anopheles mosquito larva reduce economic growth: a one-unit increase in the p. falciparum suitability index decreases growth by 1 percent. Lower elevations are also associated with faster growth. Meanwhile, exposure to state authority shows a small but significant effect on growth, indicating that historical proximity to central political powers does somewhat benefit subnational administrative areas contemporaneously. The presence of oil or gas deposits, the average distance from ports, and travel time to urban areas have no effect. Human capital, as measured by expected years of schooling, by contrast, has a strong positive effect on subnational development. Finally, the presence of a conflict resulting in deaths lowers subnational growth by 0.3 percent.Disaggregating development by subnational units allows us to examine growth processes that, as local indicators of spatial autocorrelation show, may ignore national boundaries. Nevertheless, the effects of national policies on subnational growth cannot be discounted. National development strategies, as well as country-level characteristics, may exert an influence on subnational growth. Rather than controlling for multiple country-level factors in spatial regression, we add country-level fixed effects. With the inclusion of country dummy variables, the effect of the spatial lag of growth falls from 0.65 to 0.18 (p< 0.001), indicating that neighborhood effects are reduced by two-thirds when controlling for national-level factors, but that cross-border sp川overs still exist and are significant.The resulting, within-country relationships are shown in panel (B) of Figure 5 and Table 5. Baseline per capita income, agricultural suitability, and malaria retain their effects; these are the subnational growth factors that operate within-country as well as cross- nationally. The coefficient on one factor that was insignificant in the earlier regression- travel time to major citiesis now moderately negative, suggesting that proximity to a city in another country where a national border has to be crossed is not a significant driver of growth, but higher travel time to a city within the same country adversely affects growth. The expected years of schooling exerts no effect on growth when controlling for all country-level characteristics. State exposure, elevation, and conflict do not show any significant within-country effect on subnational development as well. All other covariates remain insignificant.RobustnessInstrumenting conflict with droughtIt is possible that conflict may be affected by local economic development or that subnational growth and conflict may be driven by common factors in proximate areas. It is possible, for example, that armed insurgent groups may be drawn to poorer areas in order to see more willing recruits. In addition, negative economic shocks may exacerbate endemic tension in multi-ethnic or multi-religious regions, precipitating group conflict. These dynamics make conflict potentially endogenous to growth.Environmental shocks can precipitate political shocks that result in conflict-related deaths, often in interrelated ways (Smith 2015). Both factors have been associated with persistent underdevelopment and poverty at the national level, with some analyses showing that the two are relatedthat drought can increase the likelihood of conflict over resources (Miguel, Satyanath, and Sergenti 2004). Climate change poses an increasing risk to the global community: concentrations of CO2 and other long-lived greenhouse gases continue to increase; biodiversity is declining; tropical reefs and oceanic habitats are facing profound losses, and land degradation covers about 29 percent of the global land area (UNEP 2019). Addressing environmental shocks is central to accelerating development and ensuring that places are not left further behind.As have others, by instrumenting conflict with an indicator of drought or the standardized evapotranspiration index (SPEI), we attempt to resolve the endogeneity issue (SPEIbase v.2.5. 2017). Environmental factors strongly influence conflict, particularly through drought, and it is unlikely that climatic factors that affect droughts can be directly influenced by growth in the short-term. In sum, SPEI fulfills the standard variability and exclusion requirements for an instrumental variable (IV): it affects the potentially endogenous variable conflict, and it exerts no direct or confounding effect on the outcome of interest, i.e., subnational growth.Figure 6 and Table 6 show IV results (see appendix for raw regression tables). For IV spatial regressions, we rely on generalized spatial two-stage least squares estimation incorporating all regressors from the main model, with conflict treated as an additional endogenous covariate instrumented by SPEI. The IV results are largely symmetric with the main results. Baseline GDP per capita, soil suitability, state exposure, and expected years of schooling all retain their direction and significance. However, the effects of malaria and elevation are not robust to the IV specification. The negative effect of conflict, controlling for its endogeneity, increases tremendously from 0.3 percent to 2.5 percent. When including country-fixed effects, as with the previous estimation, baseline GDP per capita, soil, malaria, and travel time to urban areas retain their effects. Meanwhile, state exposure, expected years of schooling, and conflict are no longer significant.Sensitivity to spatial weighting matrix choiceWe test the sensitivity of our results to the choice of the spatial weighting matrix. We utilize alternative contiguity and inverse distance matrices to obtain global Moran5s / statistics. More specifically, we include first-order and rook contiguity matrices, in addition to inverse distance matrices with different distance cutoffs. The results, shown in Table 7, are consistent with the results from the queen-contiguity matrix.We use these alternative matrices to rerun our spatial regression analyses as well. The results, visualized in Figure 7, show that the derived effects from our main model are robust to the choice of the spatial weighting matrix. Almost all variables retain their effects, with the exception of conflict losing its significance when inverse distance matrices are used. A detailed table of the derived effects and regression results can be found in the appendix.Heterogeneity of treatment effectsFor policy purposes, it is important to understand if there is non-random variability in the magnitude of treatment effects across different places. As a check, we run our benchmark spatial regressions separately by income groups using World Bank income classifications for 2000. There is no theoretical reason to believe that growth drivers are the same at all income levels. For example, agricultural soil suitability is probably not important for high-income countries. The results for low- and lower-middle income countries are presented in Table 8. The rest of the results and regression tables can be found in the appendix. The convergence retains its significance for lower-middle-income countries, with low-income countries exhibiting significance only within countries. Malaria stands out as the most important factor that inhibits growth in low- and lower- middle-income countries. On the other hand, we find evidence that agricultural suitability has a positive effect on the income growth for low and lower-middle-income countries. Another interesting finding from this comparison is the effect of conflict. While conflict does not appear to affect income growth in low-income countries, it does affect lower- middle income growth. State exposure, oil or gas deposits, elevation, and expected years of schooling all matter for low-income countries, while they are not significant for lower- middle-income countries. These findings make intuitive sense.ConclusionIn 2015, over 1 billion people still lived in places where average income levels have been low for a prolonged period of time. As we have shown in this paper, these poverty hotspot regions are located within low-income, middle-income, or even high-income countries. A new toolkit of advanced geospatial technologies now permits an ever-more granular understanding of where the most vulnerable reside, even in places where the national averages are relatively high, and what can be done to get them back on track. Our local spatial autocorrelation analysis has shown that the poverty hotspots that require the most immediate attention in the world are in the Amazon basin, sub-Saharan Africa, and Central-South Asia.It is a fallacy to argue that natural migration will move people from poor areas to places that offer more opportunity and subsequently provide a solution for the regional inequalities. At least for the time being, higher fertility rates are pushing population growth rates in poorer places, above those in more prosperous places, even within each country. It is also short-sighted to overlook the current political tension around crosscountry migration, where the number of forcibly displaced people reached 79.5 million worldwide by the end of 2019 (UNHCR 2020).Recent advances in geospatial technology provide granular data on many indicators, including income, that can be used for policymaking. In order to leave no one behind, we need to be able to identify lagging regions and explore the factors that prevent these regions from developing.Using the data that is currently available, we find strong empirical evidence that while certain places indisputably face geographic constraintssuch as extreme temperatures, inhospitable soil, and proximity to the national borderother variables within the purview of policymaking also hold significant explanatory power. Human capital, infrastructure, and connectivity, shock-readiness, and governance all impact the extent to which a region develops or lags, suggesting that public officials have at their disposal a powerful antidote to poverty: inclusive local policies and institutions. We hope that as additional location-specific data becomes available, it can be used to (i) highlight underserved areas, (ii) encourage public officials to allocate resources to areas identified and underserved, and (iii) provide citizens with domestic accountability mechanisms that help ensure that resource allocation is more responsive to local needs (BenYishay and Parks 2019).Figure 1: Subnational GDP per capita growth, 2000-2015Subnational GDP per capita growth (%), .2000-2015Source: Authors' calculations based on Kummu, Taka, and Guillaume (2020), CIESIN (2018), and World Bank income classifications for 2000.Figure 2: Subnational GNI per capita, Atlas method (current US$), 2000 and 2015Source: Authors1 calculations based on Kummu, Taka, and Guillaume (2020), CIESIN (2018), and World Bank income classifications for 2000 and 2015.Figure 3: Moran scatterplot of subnational GDP per capita growth(Moran's 1=0.4214 and P-value=0.0010)-50510GDP per capita growth (2000-2015, In)15Figure 4: Local spatial autocorrelationHigh-High Cluster Low-High Outlier High-Low Outlier Low-Low Cluster Not significantFigure 5: Benchmark spatial regression resultsNotes: Graphs show direct, indirect, and total effects based on coefficients from spatial autoregressive regressions, with 95% confidence intervals. The dependent variable is real per capita GDP growth by subnational area, 2000-2015. Figure (B) shows the within-country effects.Figure 6: IV spatial regression results章 TW Direct IndirectNotes: Graphs show direct, indirect, and total effects based on coefficients from generalized spatial two-stage least squares regressions, with 95% confidence intervals. The dependent variable is real per capita GDP growth by subnational area, 2000-2015. Conflict is instrumented by the standardized evapotranspiration index (SPEI). Figure (B) shows the within-country effects.IntroductionEconomic activity is unevenly distributed across space. Even as national incomes converge, many areas within countries show widening disparities in per capita incomes. The poorest places in the world are barely growing in terms of average income but are growing quite rapidly in terms of population, despite out-migration. What explains persistent stagnation in some places and rapid development in others? Clearly, national economic growth explains a part of this, b