UfL-zi 'Jnt ^ iiiiiiiiiiiiii DEWEY iiiiiiilFrTT^iiiiiiiiiiiiiiff^^ 3 9080 03316 3350 Massachusetts Institute of Technology Department of Economics Working Paper Series PRODUCTIVITY DIFFERENCES BETWEEN AND WITHIN COUNTRIES Daron Acemoglu Melissa Dell Working Paper 09-20 July 6, 2009 Room E52-251 50 Memorial Drive Cambridge, MA 021 42 This paper can be downloaded without charge from the Social Science Research Network Paper Collection http://ssrn.com/abstract = 143 1581 at J Productivity Differences Between and Within Countries* Daron Acemoglu Department of Economics, Massachusetts Institute of ''^^- Department Technology Melissa Dell of Economics, Massachusetts Institute of Teclinology July, 2009. Abstract We large document substantial within-country (cross-municipality) differences in incomes for a number of countries in the Americas. A significant fraction of the within-country differences cannot be explained by observed human capital. We conjecture that the sources of within- country and between-country differences are related. As a we propose a simple model incorporating both first step towards a unified framework, differences in technological know-how across countries and differences in productive efficiency within countries. Keywords: economic development, economic growth, institutions, Latin America, productivity differences, regional inequality, within-country differences. JEL Classification: 018, O40, Rll "Acemoglu: Department of Economics, Mas,sachusett,.s In.stitute of Technology, 50 Memorial Drive, Cambridge 02139. e-mail: daron&mit.edu. Dell: Department of Economics, Massachusetts Institute of Technology, 50 Memorial Drive, Cambridge MA 02139. e-mail: mdoll@mit.cdu. We thank conference participants at the 2009 American Economic Association meetings, an anonymous referee, and Steven Davis for useful suggestions. We also gratefully acknowledge fina.ncial support from the Air Force Office of Scientific Reseaich and the National Science Foundation. A number of individuals provided invaluable assistance in locating and accessing the many data sources used in this paper. We thank in particular Luis Bertola, Rafael Moreira Claro, Sebastian Galiani, Brett Huneycutt. Ana Maria Ibafiez, Oscar Landerretche, ,lo.se Antonio Mejia-Ouerra, Ezequiel Molina, Glenn Graham Hyrnan, Daniel Ortega, Carmen Pages, Pablo Querubin, Marcos Robles, Oscar Roba Stuart, Carmen Taveras, Jose Te.sada. .Jimmy Va.sq\iez, and Ga.ston ^'alonetzky. MA Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium Member Libraries http://www.archive.org/details/productivitydiffOOacem much Spatial differences in income per capita motivate economics. tries have been documented extensively, there magnitudes important growth theory and development systematic evidence on is little income differences (within countries) compare and relative of two related reasons. for at least income and productivity are national, in of local, and idiosyncratic growth and development Second, they signal the possible presence of important interlinkages should strive to match. local The they shed light on the extent to which sources First, nature, documenting patterns that comprehensive explanations between interregional and within-municipahty differences are major economic differences in how relate to inequality across countries. of cross-country, cross-municipality, of oretical some coun- While income differences across countries and across regions within and national determinants of productivity, which would necessitate a unified the- framework for analysis. This paper documents the magnitudes of cross-country, cross-municipality, and within- municipahty inequality in labor incomes and household expenditure for the Americas (Canada, Latin America, and the United States), a large geographic region containing almost one billion people and around 30% of global GDP. Our contribution is twofold. First, stantial within-country (and cross-municipality) differences in output for a large number of countries. For example, we have municipality level data, the among we document sub- and standards of living eleven Latin American countries for which between-municipality differences in individual labor income are about twice the size of between-country differences (when the United States ratio is About reversed). explained by observed half of both between-country human capital, the in physical capital across regions are is included, this and between- municipahty differences are remainder being due to "residual" factors. Disparities unhkely to be the primary factor explaining these ences, because of the relatively free mobility of capital within national boundaries. differ- Therefore, similar to the residual in cross-country exercises, these regional residual differences can least partially ascribed to differences in the efficiency of The dominant be at production across sub-national units. empirical approach for understanding differences in income per capita starts with the neoclassical (Solow) growth model. The neoclassical framework explains growth and output levels by human in the neoclassical on the dynamics capital, physical capital and technology. Since technology is exogenous model, the emphasis in empirical studies starting with this model of the capital stock. Our view model inside national boundaries, the neoclassical across regions within countries. is that, given the is often mobihty of physical capital offers limited insight into efficiency differences Thus, the second contribution of our paper is to take a first step towards developing a unified theoretical framework for the analysis of cross-country and within-country differences. Our framework emphasizes the importance of local differences in the efficiency of production, likely shaped by institutions (defined as the rules determining how is collective decisions are made). More specifically, within countries, productive efficiency determined, amongst other things, by local institutions. local and regional collective decisions are the national government, and • Examples of national how made, how lower political power is institutions include the structures Local institutions influence levels of how government interact with distributed at the local level. ^ Through imposed by the national constitution and laws. these channels, local institutions impact important determinants of the efficiency of production, such as the provision of local public goods and the security of local property rights. country productive efficiency level, is determined by the average of local institutions, by national and by the technology adoption and use decisions institutions, At the of profit-maximizing firms. A country where local institutions in several regions create inefficiencies will exhibit not only within-country differences, but also lower national income. Aggregate output due to the presence of these low income regions, and new regions will lead to a smaller market size for indirectly, is lowered directly, because low demand from poorer technologies, discouraging technology adoption at the national level. sometimes It is (explicitly or implicitly) may be important assumed that even though understanding cross-country differences in economic outcomes, they do for not play a major role in explaining interregional differences view is institutional differences (e.g., Guido Tabellini (2006)). This predicated on the notion that institutions are national and cannot explain within-country differences. However, both de jure and de facto institutions vary greatly within countries. In countries with federal systems, such as Mexico and Brazil, states have considerable authority changing laws and de jure institutions, and de facto institutions of national laws, the extent to of de facto control by which local elites, local — the degree of enforcement e.g., and regional elections are and the functioning of the judiciary free urban and rural areas fair, may have promoting industrial development the degree differential will likely affect differently). As preliminary evidence on the importance ment and —often vary substantially within national boundaries.^ Moreover, national institutions and policies effects in different regions (e.g., a tariff policy in large disparities in access to countries in the Americas. We paved roads show that of local public goods — a specific and institutions, we docu- and important public good — within differences in access to paved roads are highly cor- related with individual incomes (after controlling for various geographic and other observable factors). we Finally, also discuss several existing empirical studies that connect public and economic prosperity The remainder goods to specific local institutions. of the paper is organized as follows. In the next section, we discuss existing approaches to cross-country and cross-regional income differences. Section 2 describes the micro data sets we use for investigating within-country differences. Section 3 provides various decom- positions of cross-country and within-country differences in the Americas. Section 4 summarizes evidence on the extent of within-country differences in institutional quality and availability of public goods. Section 5 introduces our theoretical framework, and Section 6 concludes. The online appendix contains additional details on data sources and further results. E.xamples of local institution.s include the degree to which regional and local facto control of stiite some constitutions that among election.s are free and fair, the de regions by local economic and political elites and organized criminals, and in federal systems, may give regional !awmaker.s substantial powers to determine local laws and pohcies. Edward L Gib.son (2005) and Guillermo O'Donnell (1993). See also Daron Acemoglu and a model of de jure and de facto institutional differences, with a discussion of the importance of de facto institutional differences in the development of the US South. "See. others, James A. Robinson (2008) for Approaches to Cross-Country and Cross-Region Differences 1 The dominant empirical approach for examining differences in income and growth rates between and between regions within countries, begins with the neoclassical (Solow) growth countries, model. As known, the neoclassical model has no theory of technology differences and a well is minimal theory of differences capital. At the cross-country human in level, capital. N. Gregory Much of its focus is on the dynamics of physical Mankiw, David Romer and David N. Weil (1992) have argued in a seminal contribution that the neoclassical growth model provides a good account for cross-country differences in income per capita without significant technology differences. In another prominent series of contributions, Robert Barro and Xavier Sala-i-Martin (1995) suggest that convergence across OECD countries, convergence across US regions, and cross-country growth dynamics can be understood through the closed-economy neoclassical growth model. While by, influential, this amongst and Charles emphasis on physical and human capital has been called into question others, Peter J. Jones (1999). I. Klenow and Andres Rodriguez-Clare (1997) and Robert E. Hall These authors document that, with reasonable assumptions on aggregate production functions, large technology differences are necessary to account for the significant cross-country differences in for growth dynamics. Moreover, it is income per capita and output per worker, and to account difficult to see model could provide an informative framework how for the closed- economy neoclassical growth understanding within-country differences, given the absence of barriers to physical capital mobility within countries. After documenting the large within-country differences in the Americas, we develop a theoretical framework emphasizing (broadly construed) technology and cross-municipality income per capita 1. in levels. Our approach emphasizes differences at both the cross-country the following potential determinants of national and local economies: Technological know-how will potentially vary at the national level, thus influencing cross- national income differences. 2. Efficiency of production will vary both at the national emphasize variation due to institutions and freeness and outcomes (i.e., (i.e., enforcement of property fairness of elections for varying levels of We entry barriers, the availability of public goods necessary for production and market transactions). Our framework Human rights, levels. government) and the implied policy is sufficiently flexible that it can also be used to think about non-institutional determinants of local productivity, some of which are discussed 3. and the subnational capital of the workforce will differ in Section 4. both across countries and within countries, in part because of differences in institutions and policies that affect access to schooling and the costs and benefits of acquiring a marginal unit of education. In this theoretical framework, national factors, in particular, national institutions and their impact on technology adoption, influence local not only local outcomes but also the overall they are adopted at the national outcomes, while, demand for new in turn, local institutions affect technologies and the rate at which level. This framework motivates (and is motivated by) our empirics in Section micro data on individual earnings from 17 countries in the Americas. 3. We start with Micro data enable us decompose labor income inequality to and within-country components into between-country and provide us with a simple methodology for separating the effect of human capital from other factors by controlling for individual-level education and experience. This exercise enables a preliminary decomposition of municipality-level economic differences between those due to education (proxying for factors embedded Our work is in related to a large literature variation in incomes within a single country. workers) and those related to the locality on spatial inequality, most of which is itself. focused on Notably, for the United States, Antonio Ciccone and Hall (1996) estimate that doubling county employment density (county labor input divided by county landmass area) increases average labor productivity by 6 percent. This estimated degree of locally increasing returns would account for more than half of the variation in labor productivity across bound estimate US states. In light of this evidence, Section 4 provides a preliminary of the role of density in explaining spatial inequality in the Americas. set of relevant contributions are collected in in particular includes two studies for Ravi Kanbur and Anthony Latin America: Chris Elbers J. Another Venables (2005), which al et. upper combine non-income census data and household survey data on income for Ecuador to produce a measure of wellbeing, finding that across-census tract inequality in well-being explains around in well-being within Ecuador, and Javier Escobal and Maximo data to investigate the determinants of spatial inequality for variation in private and public assets, 25% of inequality Torero use household expenditure in Peru, estimating a predominant role such as roads. Several studies examine within-country inequality among various countries. Most notably, Leondardo Gasparini (2004) and Juan Luis Londoho and Miguel Szekely (2000) document the extent of income inequality over time within a large provide cross-country comparisons.'^ number of Latin American countries and Several features distinguish our paper from these previ- ous studies on Latin American inequality. Through our use of population censuses and living standards measurement surveys, we have access to larger samples and higher-quality labor come estimates than the in- existing literature, which tends to use lower quality sources in order to produce an income panel. Thus, we are able to provide a more systematic and accurate snapshot of inequality patterns. We expenditure data, which are also confirm our findings using carefully constructed more reliable than labor income data sample because a high fraction of the population works in the household for several countries in our we are informal sector. Finally, not aware of other studies conducting comparative decompositions of within municipality and cross-municipality inequality into 2 We human capital-related and residual components. ^ Data use data on labor income, geo-referenced to the municipality, for 11 countries in the icas (see Appendix Table Al for the list of countries). We also geo-referenced at the regional level for an additional six countries. There is also a literature on inequality acro.ss versus Amer- examine labor income data Our data within countries at the global level. are drawn from a While interesting and important, due to severe data limitations this literature makes a large number of assumptions when constructing within-country income."? from highly aggregative data. number A is list of recent censuses of sources and measurement surveys, living standards We limit our provided in Appendix Table Al. is Our sample reported than total income.^ typicall)' better all conducted since 2000. attention to labor income, includes all which individuals with positive incomes, and for some calculations we limit the sample to males between the ages of 18 and 55 to reduce selection based on labor force participation. To increase comparability across countries, we adjust each country's income data so that national dollars, taken from the 2003 Penn World it GDP averages to Tables. per worker in constant inter- Population weights are constructed using 2000 GIS population data (Center for International Earth Science Information Network, Summary 2004). Our statistics are provided in Appendix Table A2. baseline results do not deflate incomes for differences in regional purchasing power. Differences in costs of living are important for comparisons of living standards across regions, but given our focus on productivity differences (rather than welfare), nationally-deflated incomes are more informative. We nonetheless confirm the robustness of our general conclusions to deflating incomes using the state median of a household-specific Paasche index constructed from a number of household expenditure surveys. To examine variation we use in local public goods, by the International Center geospatial data on intercity roads compiled Agriculture (CIAT, 2008a), supplemented with for Tropical more recent (2006) data on road infrastructure from the Earth Science Research Institute (Mexico) and the Peruvian Ministry of Transport (Peru). These data roads as well as their surface type (paved, gravel, or 3 identify the geographic location of dirt). Within-Country and Between-Country Differences In this section, we perform two three components: exercises. we decompose inequality First, by observable As is well at each level of human class: the and the set of additively Mean Log overall inequality in the residual. decomposable inequality indices corresponds to the We focus on two commonly used Deviation Americas (MLD) MLD = index and the Theil index. income in the the labor income of individual Americas, and L is The MLD index of Mj Ljrn lny- -Y,Y.J2^nyjrm J y-,mi is measures within the General is: J where Second, we decompose labor geographic aggregation into two components: that explained capital variables known, the General Entropy class of measures. Entropy income into inequality between countries, inequality between municipalities or regions (within countries), and inequality within municipalities/regions. income inequality in labor i =l in 771=1 (1) 7=1 municipahty m in country j, j/ is mean labor total population in the Americas.^ Similarly, the Theil index "For Latin America, the data provide information on monthly labor income, .so for these countries there may be greater tran.sitory variability than annual labor income numbers for the United States and Canada. ^Since our focus is on labor income inequality, we would have preferred to weight the data by the size of the of overall inequality in the Americas is: j=l Let us further define j/jm as mean number of individuals in municipality and as the Z/j decomposed number into our three desired m MLDjm = in country j. ^^yjm The ^ i=l — first ^1=^1 term f \ / labor income in municipality va in country m country in of individuals in country MLD= Wy-Y.^\r.y\ where m=l components + ;^ ^H ^^'^Vj-mi/ Ljm j, as t/j Then j. mean the MLD as the j, and Theil indices can be of inequality as follows (see the Appendix): - Iny, is the f^ ^ MLD + Y. lny,„ j Y^^^^A index for inequality measures between-country inequality. in (3) j, L_,>„ labor income in country in ^^^ municipality The second and third terms are between-municipality (within-country) and within-municipality inequality indices, respectively, decomposed as L fr^ where T,m country j. weighted by country j's population share. Similarly, the Theil index can be = y \ 73,^7 y ^''T' r J In L j^^ ( t'^^^ ) y \r^ J-iJTn ^, Lj yjm rp ST^ J^jm yjm ''" yj ^1 , / yjm [ y, (4) Lj y, the Theil index for inequality in municipalitv is These expressions show that the MLD m in index weights by population shares, whereas the Theil index weights by income shares. We begin in Table income 1 by examining the distribution, as well as the decompose between the 90th and 10th percentiles of the labor and Theil indices, for its three all individuals in our sample. component We parts: inequality across between municipalities/regions (within countries), and inequality within When decomposing municipalities/regions. population weighting schemes. equal population in ratio Americas into overall inequality in the countries, inequality Brazil, MLD all The countries, first Western Hemisphere inequality, we consider two uses actual population, whereas the second assumes and thus reduces the influence Colombia, Mexico, and the United States. of large countries such as This latter scheme is similar in spirit to the convention in the growth literature where different countries are given equal weight. comparison purposes, we decompose overall inequality separately For for the eleven countries geo- referenced to municipalities and for the six geo-feferenced to regions.^ At the bottom of the table, we decompose inequality for all countries included in our sample. We also report cross-country labor force rather than the size of the overall population. Unfortunately, data on labor force participation are not much Americas at tiie municipality level. Guatemala, Honduras, Mexico, Panama, Paraguay, Peru, United States, and Venezuela have data geo-referenced to the municipality. For Canada, Chile, Colombia, Costa Rica, Ecuador, and Uruguay, the data are geo-referenced to larger regions. readily available for of the ''Bolivia, Brazil, El Salvador, inequality of 2000 GDP per worker and 2000 GDP per capita from the Penn World Tables for countries in the Americas for which data are available. all 33 for GDP per worker and 37 for GDP which we do not have labor income data. The pattern our sample in is This increases the sample size to per capita, primarily by adding Caribbean nations for GDP data show that the cross-country inequality similar to that for the entire Americas, confirming that cross-country inequality in the subset of countries we examine similar to that in the is Appendix Table A3 provides additional documentation Americas as a whole. of inequality patterns and shows the decomposition of between-municipality and within-municipality inequality separately for each country. Table is 1 documents that for our entire sample, inequality in labor income across countries about one half to one third of the magnitude of inequality within municipalities/regions and between two to four times as large as inequality across municipalities/regions, precise sample and weighting scheme. MLD For example, using the depending on the index, equal population weights and focusing on countries with municipality data, overall between-country inequality is 0.23, while within-country inequality The inequality. 0.66, 0,11 of is which is due contribution of the between-municipality component to between-municipality is index, which gives greater weight to the top of the distribution, that (recall that the Theil The larger magnitude of the between-country much to the United States uses population shares). relative to the between-municipality inequality driven by the presence of the United States (and geo-referenced to regions), which are is, MLD index weights by income shares whereas smaller with the Theil Canada when we include is countries with data richer than the remaining countries in our sample. For this reason, decompositions without these two countries might be more informative. These are reported in the third and eighth rows. For example, the third row shows that, again focusing on the data with municipality referencing, between-country differences are now about half of the between-municipality differences both with the MLD and 0.05 vs. 0.11 MLD and Theil indices A4 shows with Theil). Appendix Table (0.05 vs. 0.12 a similar pattern when with the data are deflated using regional price indices. One important concern with the above inequality decompositions and transitory income shocks may be To investigate this concern, we inflating the is that measurement error magnitude of within-municipality inequality. followed the methodology proposed by Angus Deaton and Salman Zaidi (2002) to carefully construct, item- by-item, a comparable measure of household expendi- ture for a number of countries in the Americas. Inequality decompositions for household expen- diture, similar to those reported for labor income in Table 1, are presented in Appendix Table A5. As expected given concerns about measurement error, the magnitude of within-municipality inequality in household expenditure all is less than that in labor income. comparative patterns are similar, as within country inequality in Nevertheless, the over- household expenditure is substantially greater than inequality between countries.'^ 'Results (available upon request) documenting educational inequality for 19 countrie.s in the Americas, 11 of which have micro level census data, show that inequality in education attainment across municipalities in Latin America is about twice as large as inequality in educational attainment across countries, and this pattern is reversed when the U.S. is included. Table 2 limits our sample to males between the ages of 18 and 55 to compare incomes across a more homogenous population (particularly in terms of hours worked). Using this subsample, we perform the decomposition between predicted and labor income of individual i municipality in m in residual incomes. country Recall that yjmi is Let Xjmt denote a vector of j. detailed education categories (in particular, zero to four years of schooling, four to eight years of schoohng, some high school, high school graduate, and one or more years of higher education). Let experjmi denote potential experience (defined as usual as age-schooling completed-6). We then decompose labor income into predicted and residual components by running the following flexible regression, which allows for a full set of interactions quartic in potential experience, separately for each country between education categories and a j: 4 In yjmi where 5-j is a country-specific constant and ejmi has zero Given estimates from in = 22 ^jmt(ea;perjrat)''/?jfc + (5), we examine inequality Sj + ejmi, mean and (5) country-specific variance. in overall labor income {Vjmi), inequality predicted labor income (exp(2^,_j Xj„jj(ea;perjmi)''/5jfc))i and inequality in residual income Notice that country-specific constants, which are unrelated to difTerences in (exp(Jj +ejmi))- human capital, are part of residual income. Table 2 uses the Theil index to decompose each of the components of income (overall, predicted, and residual) into inequality between countries, inequality between municipalities/regions (of countries), and inequality within municipalities, reporting results by country only for the six countries with micro census data (and hence large within-municipality samples).^ Inequality in overall labor income to prime-aged males. differences, is where we do not restrict the sample sizable between-country predicted labor income similar to that reported in Table The decomposition shows 1, which become much smaller when the United States between-country residual inequality in the sample).^ The magnitudes of is is excluded. somewhat smaller (when comparing between municipality inequalities The magnitude across in predicted all of countries and residual labor income are similar. Table 2 also shows that the bulk of the within-municipality differences are due to residual factors. This may reflect a greater dispersion market imperfections and discrimination, measurement error, or in unobserved some combination skills, labor thereof. Overall, our evidence suggests that years of schooling and the experience of the labor force can explain a significant fraction of income disparities across and within countries, but residual factors are also significant and generally of comparable magnitude. Although these residuals undoubtedly include a component of unmeasured human capital differences within countries, they also likely reflect the effects of local factors impacting productive efficiency. ''Results using the ' MLD index are similar and available >ipon request. Notice that the decomjjo.sition between predicted and residual incomes i.s not additive, since we are taking exponential transformations of predicted and residual log incomes before the decomposition. 4 Determinants of Interregional Differences The large differences in labor incomes and residual incomes documented in the previous section are unlikely to be entirely due to differences in the physical capital intensity of production, given the absence of barriers to capital mobility within countries. Instead, they likely reflect the We influence of certain local factors, the nature of which will be discussed further in this section. focus on determinants of cross-municipality differences. While within-municipality differences are also clearly ineresting and important, space constraints preclude their treatment here. One possibly relevant local characteristics as emphasized, for example, Table A6 is the density of economic activity across regions, by Ciccone and Hall (1996).'" To examine this issue, Appendix repeats the decomposition of predicted and residual inequality, adding a control for municipal-level population density to equation (5).'^ The patterns when population density included in the predicted component of income are very similar to those in Table is exercise therefore suggests that the bulk of the between-municipality differences in the Americas is not accounted density of employment, which Our argument is by differences for is that, in the likely to in We now why We in be strongly correlated with population density). '^ same way that technology investigate technology, broadly construed. This population density (or more generally, in the shaping cross-national economic differences, they also differentials. 2. income there would may be ideally differences play an important role in likely play a major role in within-country significant within-country differences in document the correlation between various aspects of local institutions and incomes, but unfortunately there do not yet exist uniformly constructed, municipality-level measures of institutions in the Americas. We instead use a novel dataset on road infrastructure to measure the within-country inequality in proximity to paved roads, an important form of public infrastructure, and the correlation between proximity to roads and labor income. Local institutions affect investments other public goods by influencing the incentives of government (related to corruption and accountability), the capacity of the to finance public investment (from Ideal taxes or transfers the incentives and opportunities of citizens to effectively in road infrastructure and officials to local provide public goods government to raise revenues from the central government), and demand public goods from local and national politicians. '"Other factors thnt liavc been cniphasized for explaining cross-<:oinil,i'y differences include geogiajjliy and Regarding natural resources, fVancesco Caselli and Guy Michaels (2008) find that the in Brazil, discovery of oil in a municipality increases municipal public exjienditures but has little impact on measured citizen wellbeing. Melissa Dell, Benjamin F. .lones. and Benjamin A. Oiken (2009) document a .statistically significant cross-sectional correlation between climate and income within a number of countries examined in this study using the same economic dataset. The quantitative magnitudes they report, thotigh not trivial, suggest that climate cannot explain the full spatial variation in the data. This leaves significant scojje for the institutional factors that culture. we examine. The effects of culture (belief .systems) have also been emphasized recently (e.g.Tabellini, 2006). is not available at tlie nnmicipal "ideally we would control for employment density, but this information for much '^Results (available upon request) are also similar areas. level of the Americas. when we estimate inequality .separately for urban and rural Table 3 documents substantial differences across municipalities in proximity to paved roads. ^"^ km We calculate each municipality's proximity to intercity paved roads by overlaying a km grid on the Americas, calculating the distance (allowing for changes in elevation) each grid cell's numbers by country only In Table in Appendix Table A7 and the overall inequality decompositions in the final three rows of Table 3 Latin American countries and the United of all we conserve space by reporting the our sample are given for all countries in 1 with census income data geo-referenced to the for the five countries The numbers 3, x from centroid to the closest paved road, and then averaging this distance over grid cells contained within the municipality.''^ municipality. 1 States. The is for this full and second columns first of sample Table 3 report the mean and standard for Brazil, Mexico, Panama, the United States, Venezuela, and the Americas overall; column deviation of municipal-level proximity to paved road networks presents the 90-50 percentile ratio; and columns (4) through (7) decompose inequality in (3) proximity to paved roads into inequality across countries and inequality across municipalities within countries.'^ Not surprisingly, the United States is on average the most proximate to paved roads, and of the five countries reported in Table 3, road networks across municipalities Latin America in Brazil is the least. is Inequality in proximity to paved about 2,5 times as large as inequality across countries and remains higher than inequality across countries even States roads is is Nevertheless, these patterns included. may reflect both easier and more useful on the coasts than unrelated to productivity and income differences. this issue in columns and (8) (9), in the building Amazon), and may thus be undertake a preliminary investigation of for males between the ages of 18 and 55 in the with census income data geo-referenced to the municipality.'^ Column state fixed effects. the United (e.g., by computing the partial correlations between proximity to paved roads and the log of individual incomes, five countries We when geographic factors States (of which there are 27 in Brazil, 32 in Mexico, 9 in (8) includes Panama, and 23 Venezuela) are geographically small, and thus including state fixed effects ensures that we in are comparing municipalities characteristics that may in close affect the density of level controls for elevation, slope, and 2000 (see the data appendix in column (9)). road networks, column for sources), as well as is a in Brazil, municipal full set of age dummies. The correlation negative and highly significant for all countries (except Translating these correlations into elasticities suggests that increasing a municipality's average distance from paved roads by males by 0.06% (9) also includes and mean annual temperature and precipitation between 1950 between distance to roads and incomes Venezuela geographic proximity. To further control for geographic by 0.09% in 1% reduces labor income of prime-aged Mexico, and by 0.14% in Panama. Naturally, these elasticities do not reflect the causal effect of proximity to roads on income. upon reqiie.st) are similar when we con.sider both pa.ved and gra,vel roads, between individual income.^ and road.s are .similar when we instead calculate road network density, the municipality's total kilometers of paved roads divided by the surfa.ce area of the municipality, '''Note that in contra-st to the previous tables, there is no within nnmicipality variation in road networks since proximity is measured at the municipal level, "'We limit the sample to prime-aged males to decrease selection on labor force participation. Results (available upon request) are similar when all individuals with positive incomes are included in the sample. ''^R.esults (available '""The correlations • 10 proximity to roads First, thus interpret it as a is proxy likely correlated for a with the availability of other public goods, and we bundle of public goods. Second, there are several other reasons, unrelated to the availability of public goods and local institutions, be correlated with incomes (even after impact of roads (or local public paper. we summarize Instead, we why proximity may to roads control for observable factors). Estimating the causal goods more generally) on incomes is beyond the scope of this several recent studies that provide detailed empirical evidence relating institutions to local public goods and economic prosperity within particular countries. Melissa Dell (2008) utilizes a regression discontinuity approach to examine the long-run impacts of the mita, an extensive forced mining labor system in effect in Peru and Bolivia during the colonial era. She estimates that a mtta effect lowers household consumption today by around one third in subjected districts. Mtta districts historically had fewer large landowners and lower educational attainment, today they are less integrated into road networks, and their residents, who face difficulties in transporting crops to markets stantially Lakshmi effects more likely to be subsistence farmers. Outside of Latin America, Abhijit Banerjee and Iyer (2005) similarly on investments a robust association due to poor road infrastructure, are sub- in health between show that colonial land revenue systems in India have long-run and education infrastructure, Daron Acemoglu political inequality in the 19th century in et al. (2008) find Cundinamarca, Colom- bia (measured by the lack of turnover of mayors in the municipalities) and economic outcomes today. They local public also provide evidence, consistent with Dell's (2008) findings, that the availability of goods might be a particularly important intervening channel. Similar correlations are obtained for Brazil by Joana Naritomi, Rodrigo B. Soares, and Our Juhano Assungao (2007).^' findings are also broadly consistent with a large literature on the impact of infrastructure on economic outcomes. In a series of studies on Peru (summarized in the Escobal and Torrero chapter in Kanbur and Venables (2005)), Javier Escobal and co-authors empirically connect poor local road infrastructure to higher transaction costs, lower market participation, and reduced household income. Other studies of the effects of local public goods include, among others, Esther Duflo and Rohini Pande (2007), which examines the effects of dams in India, and Dave Donaldson (2008), which analyzes the impact of railroads in colonial India. Towards a Framework 5 In this section, we provide a simple framework income differences and their dynamics, to interpret cross-country and cross-municipality and to highlight the two-way interaction between local and national outcomes.-'^ The framework builds on endogenous technological change models in Latin America. Gibexamining "subnational authoritarianism", defined as the persistence of authoritarian regional governments in a nationally democratic society, and has emphasized large differences across Argentine provinces and Mexican states in the extent to which elections are fair and competitive, elected local authorities are protected from arbitrary removal by regional authorities, and the judicial system is accessible and independent. Similarly, O'Donnell (1993) has classified regions of Latin America based on the functioning of rule of the law, documenting substantial differences within countries. '*While the within-municipality difTerences are important, they are not our focus here. ^''AIso relevant son (200.5) i.s the political science literature on subnational in.stitutional variation has developed a theoretical framework for 11 (e.g., Philippe Aghion and Peter Howitt, 1992, Gene M. Grossman and Elhanan Helpman, 1991, and Paul M. Romer, 1990) and on the model of international technological in Acemoglu we consider a model that in Section 4, we countries. In addition, embedded capital diffusion presented (2009, ch. 18). Motivated by the empirical results in Section 3 and the discussion explicitly distinguishes countries as well as regions within delineate productivity differentials resulting from technology, workers, and local differences in public goods and institutions. in main capital differences, which are the human Physical emphasized by the neoclassical growth model, are factor omitted from this framework. We Population that all = in 1, 2, ..., J, in and there is each country normalized to is 1 y,.m L-j^rn is [t) = in economy ^^^ 1-/3 j at time a number is institutions In equation assume is hj^rn is factors, a region-specific productivity term. 4, we {h,,mL,^mf (6) , the efficiency of labor, determined which can vary across regions and While the 7j,mS could incorporate focus on one interpreation of the 7j,mS: efficiency terms reflecting local and policies (e.g., the provision of local public (6), the variable 7 This is without any is raised to the functional form in (6) is = t. 1- /3 goods or security of property rights). in order to simplify the expressions that 7 has no natural scale. Our population For now, we also ignore migration across regions. similar to the standard Dixit-Stiglitz aggregator used in en- dogenous gi'owth models. Similarly, Nj country j at time power loss of generality, since normalization implies that J2m=i ^j.m The We Mj. of characteristics affecting regional productivity, in accordance with the empirical evidence in Section follow. t labor input, which varies across regions; and 7j,m 1, ..., /o by education, public goods and other institutional countries; m= no population growth. ty-^dv X,.„(^, / country j indexed by good denoted by Y, and the aggregate countries and regions produce a single final m There are J countries continuous time. in and Mj regions (municipalities) production function of region where economy consider an infinite-horizon world indexed by j (t) denotes the number of machine varieties available to This variable captures the technological know-how of country j. Technology diffuses slowly across countries and producers can only use the technologies available in their country. This implies that Nj — thus the available technology —varies across countries. once a country has a particular Finally, Xj^m [v, t. To t) is the total level of technology, amount of it can be used machine variety u used in in all regions in region However, the country.-'^ m in country j at time simplify the analysis, let us suppose that the xs depreciate fully after use. Each machine variety machines embodying within the country. in economy j is owned by a technology monopolist, which this technology at the profit-maximizing (rental) price Py We transport costs, thus machine prices will be the same across regions."" '"Thus there is no slow diffusion of technology specific productivity (i^, t) assume that there are no regional taxes on machines or embedded in the 77,™ s acros.s regions "°Again any such differences can be incorporated into the 7j,mS. 12 in all regions differences in The monopolist can within a country, though differences can capture such slow diffusion. will sell in region- produce each unit of the machine at a marginal cost of any loss of generality, We we normalize ijj =1 - i/' terms of the in final good, and without (3. assume that each country admits a representative household with constant (CRRA) aversion preferences, with the same discount substitution) and the are given by /q^" exp {-pt) same degree rate across countries. In particular, preferences at time - (Cj {t)^~^ l) / - {I 9) dt, > with p and ^ > Thus i = Although our 0. empirical investigation so far has emphasized income inequality, our focus here in productivity across regions. relative risk of risk aversion (intertemporal elasticity of on differences is and consumption behavior of differences in the saving households within a country are not central to our framework and the representative household assumption enables us to suppress these. In addition, there countries produce the same final good) or in assets (thus no international trade is Cj{t) where Xj X,{t) spending on inputs at time is (t) + may which take the form of t R&D each country j at each time for + QZ,{t)<Yjit), and Zj [t] is t: (7) expenditure on technology adoption The parameter or other technology expenditures. at time Cj measures country-level distortions or institutional and policy differences, and t, (all no international borrowing or lending). This implies that the following resource constraint must hold - goods in will be a key driver of potential technolog}- differences. Technology maximizing where country j evolves as a result of the technology adoption decisions of profit- firms. In particular, the innovation possibilities frontier takes the A'' (t) common in an index of the world technology is to all economies. technological know-how of country j advances as a result of the a country-specific constant, is rjj > 0, Each economy starts with some initial any firm can invest The Since world growth is we also rate per unit of R&D An rates (i)) > and and other technologydepends on N and there is {t) /Nj [t)). free entry into and adopt new technologies (8). 0, that is, = gN{t). (9) factor markets are competitive. The capital in country j are denoted, respectively, equilibrium consists of sequences of technolog>' levels, for (^ not the focus here, suppose that the world technology frontier advances assume that and wage rates > A^j (0) (denoted by Zj grow) at an exogenous rate g > human R&D effectiveness of these investments technology stock in N(t) Finally, and know-how (captured by the term according to the innovation possibihties frontier (or frontier varieties for all j, and more importantly, on how advanced the world technology relative to country j's technological research, so that > rjj This form of the innovation possibilities frontier implies that the related expenditures of firms in the country. frontier frontier, form interest rate by R&D levels, r^ (t) and the wage and Wj machine [t). prices, interest each country and machine demands and output levels for each region, 13 such that final good firms and technology monopolists maximize technology adoption, and the representative household A utility. differences, = will I - each country maximizes its discounted rate. comparisons of economies in BGP demand 0, be p^ {v,t) markup curves for their machines, will set a constant 1. Given (6), this also which are a natural starting point. and the equilibrium price of every machine — in and cross-region In thinking about the cross-country who straightforward to verify that in any equilibrium, technology monopolists, iso-elastic tl> entry into free is balanced growth path equilibrium (BGP) refers to an equilibrium path each country grows at a constant It is in profits, there implies that the face over marginal cost each country at each point in time in demand for machines in each region of each country that will maximize the profits of the final good producers will be This implies the intuitive result that there workers have greater human and when capital Consequently, a technology monopolist total between price and marginal cost machine Let us start with the BGP. The in (9). for business. fol- = /3X^„ii Tj.m^j.m-^j.m, where P is = 1 — /?), while the summation gives the i/) in country j will (u,t) and t/' CRRA the (10). straightforward to verify that, given It is grow at the same rate g as given (1 ttj when make the which follows from sales of this monopolist, more favorable local conditions are machine variety (for lowing level of profits at every point in time: difference be more intensive use of technologies will countries (8), all must preferences of the representative household imply the standard Euler equation, which gives the growth rate of consumption of each country at each point in in time as C {t) /C (i) = each country grow at the rate g, {vj (i) — so the interest rates r* = p + ^m=l = Combining this expression with the innovation j's relative technology with r* given by (11). jij = Nj [t) /N In addition, it {t) in must also thus consumption be constant and equal to ". 9g. Consequently, the value of a technology monopolist in V* BGP, output and p) /d. In the BGP in ,.••.",;, country j is 'yj,mf^j,m-L'j,Tn , will we obtain {Nj (0)} _j, there exists a 'The proof of that country be can also be proved that this BGP allocation saddle-path stable, in the sense that starting with any strictly positive vector of levels . , ' possibilities frontier in (8), BGP initial unique equilibrium path converging to this BGP.^^ this result follows the similar derivation in Aceinoglu (2009, ch. 18). 14 (11) is globally technology What does this BGP above derivation immediately establishes that the country j in the BGP level of income per capita m in region in is '^'^"'^'0^^''"'^''"' ^^-=r~^( = (nrnh,,m) N {0) (13) . ) Then, summing across regions, total income ^^* The and cross-region inequality? allocation imply for cross-country r^(c^j country j in the in BGP is ^'w- lE-^^^-^^-^^-l (14) These expressions give the theoretical counterparts of the regional (municipal) and national labor incomes we computed and compared in Section 3 and can be used to develop between our theoretical framework and between- and within-country more structural They differences. that countries that have better possibilities for adopting world technologies (higher links highlight 77^3), where firms face less severe barriers to adopting technologies (lower Cjs) and have on average better local institutions (higher 7j,mS) to be and workers with greater human capital (higher Within a country, richer. regions share the all regions that have better local institutions (higher human 7j,Tns) capital (higher /ij.ms) that will be richer. This same technology /zj.ms) will A'j, so it will framework also (j, emphasizes the two-waj' m) and [j\m') identical characteristics but are situated in different countries will have different number new will be those and those that have workers with higher interaction between national and local factors. First, two regions because they tend that have income levels, have access to different country-level technologies. Second, a country with a of regions with low 7j,mS technologies (as and /ij,mS will shown by equation (10)), generate a lower and this will demand for machines embodying reduce the profitability of adopting technologies from the world frontier at the national level and will tend to reduce national income. This channel also suggests that if within-country inequalities are caused by the failure of some regions to offer good business conditions, public goods and a workforce with the requisite then this and will tend reduce income indirectl}' in the country both directly (because (because technology adoption at the national level becomes The framework presented above does not question One may is some regions whether migration would are poorer) less profitable). allow for migration across municipalities. affect cross-municipality differences in skills, A ~- natural income and output. conjecture that differences due to local institutions and policies (the 7s in the model) would be arbitraged away when migration as a variety of factors "The magnitude make movement is possible. This is not necessarily the case, however, across municipalities costly. of the indirect channel in practice will depend on the extent First, in parts of to which firms produce Latin for the Openness, defined as imports plus export as a share of GDP (Heston et al., 2006) ranges considerably across the countries we consider in our empirical analysis, from around 20% in Argentina, 2.5% in Brazil and the United States, and 40% in Boliiva. Colombia and Venzuela to around two thirds in Mexico, Chile, domestic market. and El Salvador. 15 Second and more importantly, migration America, there are explicit barriers to migration."'^ due to the 7s only when there are no difTerences will arbitrage all differences living in the costs of and housing across municipalities. In practice, both housing costs and the prices of other goods and services differ significantly across regions. Recognizing that migration, even not lead to an equalization of could take place without any impediments, would if it non-human capital incomes (when not deflated regional differences will have two components; those due to factors mobile with workers) to the influences of the hs and those due to differences and 7s in the We model. human fully) implies These correspond in local conditions. can then map that capital differences (and other the differences due to the hs to those related to education and the differences due to the 7s to the residual differences obtained in Section 3 after we removed the influence of education and experience. decomposition suggests that the cross-municipality variation accounted for In particular, our by these two sources are broadly similar. Concluding Remarks 6 This study used a novel data set of labor municipality) income differences for a large America, between-municipahty differences We documented that about can be accounted factors. We also for incomes to document within-country (cross- number in of countries in the Americas. half of the between-country by differences in Within Latin incomes are greater than cross-country differences. human capital, proposed a simple unified framework and between-municipality differences the remainder being due to residual for the analysis of cross-country and within-country income differences, which emphasizes the importance of the efficiency of production. Productive efficiency is determined at the country level by the technology adoption decisions of profit-maximizing firms and by national institutions, and within countries by local institutions, such as the availability of local public goods and the security of property Future research could follow a number of promising paths for identifying the rights. underlying determinants of local productivity differences. The empirical and qualitative evidence suggests that differences in local public goods level — are —determined in part one source of within-country differences. measurement and empirical investigation level, as well as new theoretical by institutions at the local Such patterns call for and regional more systematic of specific institutional features at the subnational work modeling the impact and endogenous determination of these local forces. ^^ in regions where a .substantial portion of the land is held b)' indigenous communities, there are and traditional impediments to selling land to outsiders, making larger cities, which often have significant Notably, legal dis-amenities, the main viable destinations for migration. IG A. Online Appendices — Not for Publication Al. Decomposing Inequality Measures In this section, we derive the decompositions of the general entropy indices used in the text of this paper: the The mean Mean Log (MLD) Deviation the income of individual is m— income and Lj is " given by: . (A-1) 1=1 1 m in municipality i is — ^ ^Iny^™, - =lnj/j ^ where yjmi index and the Theil index. log deviation index for total inequality in country j MLDj main country in j, y^ is mean national total national population in country j. Now adding and subtracting Ylrn=i ~i~' ^''^yjm to both sides (where y^rn is mean income in municipality and L-jrn is the number of individuals in municipality m), we obtain: m MLD, = ""' / Iny, - m=\ \ \ L- J^ -^ In ""' ^ L- + J^ -^MLD.m. j/,,„ m=l j -^ where MLDjm = is the MLD measure MLDs - InVjm J —^ m index for inequality in municipality In yjmi country in and the second term of cross-municipality inequality, let us repeat the versus within countries. same exercise for decomposing the The Western Hemisphere MLZ? = In 2/ -- ^ J where yjmi is the income of individual Western Hemisphere, and L is i =l MLD add and subtract ^^^j term is the MLD a weighted average of the number of individuals in country j. where y, is is m in country j, = Inyj - y^Int/,, Lj X^ ' MLD index for inequality in country j. n m=l y is mean income in the Western Hemisphere, mean income Again, basic algebra yields: J then: In y^mi, /here MLD index into inequality between t=l in municipality -^lnj/_,, MLD index 2^ 771=1 total population in the Now the is first within each municipality, where the weights are municipality population shares, Now is The j. 1=1 in country j and Lj is the Plugging in MLDj from our decomposition of above yields: MLD =hny-J2^\nyA+J2j;^ 3=1 where, from above, municipality m in MLDjrn = country — 77— YLi^ In J/jm The j. m=\ J=l J ' In yjmi is the MLD term gives between country first 771 / =1 index inequality, -^ for inequality in and the second and third terms are between municipality (within country) and within-municipality inequality, respectively, weighted A by country decompose the Theil index similar exercise can be used to where y^rni, Vj sides, and Lj are defined for country j, given by: V V|^ln(^^), - r, both population share. j's (A-2^ and subtracting Ylim=\ 'ir'^^ Iny^, as above. Again, adding we obtain M rp ' __ ST^ ^jm yjm ' ^, m=l Lj '^ yj' ' ^ ^jm rp Lj \ Vj-m I , VjT In yj V] where yjrm — 7 Tjm , _, is ^jmyjm the Theil index for inequality in municipahty Now we decompose , m I yjr \ yjm in country j. the Theil index for overall Western Hemisphere inequality, given by: yjr j where yjmi is m= l the income of individual the Western Hemisphere, and same steps = as before, we can A'^ is i l 1 = (A-3) y " 1 in municipality the total population m in in country j, y is mean income in the Western Hemisphere. W'ith the also write +E ^-E^^^. L y ^Ly\y j=i j=i ^ ^ where Mj Ljm ^.=y:y. is the Theil index for inequality in country decomposition from above J Ljyj T = S" ~L^ j = L i y J tlLJll ] y In yjr Note that this is equal to (A-2). Plugging in the yields: /Inyj \ j. yjr ^Y^ j=i Mo Lj yj tl^l L y E 18 J J-'jm , yjm , / M, . yjm \ , V^ ^jm yjm rp where again T,m in country = The j. ^^,-7 r In ^'Z' component first ^f^^ I is ) the Theil index for inequality in municipaUty is cross-country inequality, the second component municipality inequality, and the third component m between is within-municipality inequality. is A2. Data Methodology We in discussed the construction of our nationally (but not regionally) deflated labor income data the text, so here our main focus is on the methodology used to create our household expen- Appendix Table A5). diture aggregate (examined in We construct our measure of household expenditure by aggregating expenditures on food and non-food items, durable goods, and housMarriages, births, and funerals, which are lumpy and relatively infrequent expenditures, ing. We are excluded. very similar when include expenditures on health, but patterns of household expenditure are these are instead excluded. Gifts and transfers made by the household are not included, as these will be counted as they are spent by their recipients. The method we data available in use to calculate the flow expenditure on durable goods varies according to the each household survey. Household surveys for Bolivia, Ecuador, Guatemala, Honduras, Nicaragua, Panama, Paraguay, and Peru include the current value and age of durable goods held by each household. Following the recommendations of Deaton and Zaidi (2002), we calculate the average age for each durable good recorded in the survey. t using data on the purchase dates of the good Then, we estimate the average lifetime of each good as 2i, assumption that purchases are uniformly distributed through time. The remaining good then calculated as is the flow of services The life. interest is — 2t t, where t is the current age of the good. of each rough estimate of derived by dividing the current value by the good's expected remaining component in the flow of services is ignored. In contrast, the household survey for Brazil includes a household and A under the life list by the of durable goods held their ages, but does not contain estimates of their current values. We estimate the purchase value of the good as the state median price for that good using data on purchase values from the expenditure section of the survey. good as Then we calculate the average To complete the estimate, we calculate the average user 2t. median purchase value divided by the average lifetime of the good. life of the cost of the good as the Finally, some household surveys (Chile, Colombia, Costa Rica, El Salvador, Mexico, and Uruguay) do not include any information about the stock of durables held by households. In these cases, we calculate the durables sub-aggregate as expenditures by the household in the previous year on durable goods. Some periods — surveys ask multiple questions on the same expenditures with different reference i.e., Deaton and ''the last Zaidi, two weeks" versus a "typical month". we use the latter. expenditure per equivalent adult We calculate both per capita household expenditure (in local prices, national prices, and 2005 international lowing Deaton (1997), we assume that children aged aged Following recommendations by to 4 are equal to 0.4 adults, $). and Fol- and children 5 to 14 are equal to 0.5 adults. Adjusting for differences in purchasing power is important for making regional and interna- tional comparisons of household welfare using expenditure data. Consider a Paasche price index 19 comparing the price vector faced by the household, where g'' the consumption vector and is Rearranging and the reference price vector, ph, q^ denotes the inner product of these p'' P where the value of the household's consumption defined our object of interest and x/i is income accounting practice, all To in /cth at reference prices, y^ = p° q^, is household expenditure. Note the convenient link with national which real national product is the lefthand side of (A-5) summed households. calculate y^ from our expenditure Pp where two vectors. yields: P over po- tt.'^ is data, we rewrite (A-4) as: ^ = (A-6) the share of household h's budget devoted to good k and component pjj (and p°) denotes the P^ involves not only the prices faced by house- of the corresponding vector. hold h in relation to the reference prices, but also household /I's expenditure pattern. Using a Paasche price index with household specific weights allows us to account for differences in expenditure patterns — prevalent across regions—when adjusting the information about food quantities and expenditures available prices. in We calculate P^ from our household surveys. We take the reference vector po to be the median of prices observed from individual households in the survey. To reduce the influence of outliers, households in the we replace the individual p^ by their medians over same municipahty. Calculating prices from our survey data requires converting all quantities (i.e., pieces, bottles, bundles) to constant units (kilograms or hters) for each commodity, and then dividing total expenditure by quantity purchased. In calculations before releasing the data. necessary conversion factors some cases, national statistics offices —by commodity— to convert to constant (most notably, Bolivia), surveys report some quantities factors. In these instances, we performed these In other cases, the survey documentation contains the in "pieces" units. In a couple of cases but do not provide conversion use the conversion factors for similar goods provided by other surveys. Constructing a meaningful consumption aggregate also required checking the survey data for obvious data entry errors and irregularities, most some —but not —surveys, national carefully all common statistical offices did a in reporting food quantities. In thorough job of error checking. After examining individual data points, we used the following procedure to correct data that were clearly recorded incorrectly. If the household's annual expenditure on a good was than four standard deviations above the mean expenditure on that good municipality (or state if the municipality is in the more household's not identified or had very few observations), the 20 observation was replaced by the municipality median. Less than The procedure cutoff. consumption is is used for all 1% of the sample The surveys in our dataset. meets this distribution of aggregate robust to instead dropping these items, or requiring the value to be more than standard deviations above the municipality mean. Results are also robust to using deviations five than in logs rather levels. We apply a similar procedure index, dropping quantities that are more than in calculating prices for the Paasche four standard deviations from the municipality mean. After deflating regional prices to national prices, we further adjust for differences in inter- national purchasing power by normalizing the data so that household expenditure within each country aggregates to 2005 national per capita consumption in international dollars from the World Development ' Indicators. ' . ' Income data are drawn from either population censuses The census or household surveys. data give a single measure of labor income, whereas labor income from household surveys constructed from responses to various questions about earned income pation, secondary occupation, etc.) is The methodology we similar to that used to construct the is from primary occu- use to construct the income aggregate consumption aggregate. PPP from the census or household survey with the value of (i.e. We compare per capita income GDP adjusted per worker in 2003, taken from the Penn World Tables. This allows us to calculate a factor to adjust income so that it in averages to GDP per worker in constant international dollars. To produce the decompositions Appendix Table A2, we also deflate When index discussed above. power pro\'ided by National by the state median of the household a Paasche index is unavailable, specific we use data on regional purchasing Statistics Offices. The climate and geography variables were constructed as follows. Municipal-level temper- ature and precipitation variables were calculated using 30 arc second resolution mean temperature and at (NASA and NGIA, municipal-level mean (1 kilometer) precipitation over the 1950-2000 period, as compiled by climatologists U.C. Berkeley (Hijmans, Robert et al, 2005). data Paasche We also use 30 arc second resolution terrain 2000), collected by the Shuttle elevation and slope. Radar Topography Mission, to construct The GIS municipality boundaries were produced by the International Center for Tropical Agriculture (CIAT, 2008b). A3. Further Results Table Al tics. lists Table the data sources used in this paper, and Table A3 examines A2 provides summar)' statis- labor income inequality, where labor incomes are not deflated for regional purchasing power. The overall decompositions of inequality into cross-country, cross- municipality, and within-municipality inequality are thus identical to those presented in Table of the main full set text, with Table of countries for which A3 presenting the inequality measures, country-by-country, for the we have data. Table A3 shows that the extent of inequality across countries depends on the measure being used, though Ecuador, Peru and Venezuela are the most unequal countries with 1 all three measures. Table where incomes have been deflated by the state level 21 A4 examines median of the among labor income inequality Paasche price index described above. The Table to the shown results are very similar to those A5 examines is less than that Not than the between country less 1 and A3. surprisingly, the extent of inequality in ones.^"^ MLD The ranking household Nevertheless, the overall comparative patterns in labor income. are similar. In particular, the within-country sometimes quite Tables inequality in equivalent household expenditure, constructed according methodology described above. expenditure in and Theil indices are substantially greater of countries in terms of inequality in Table different than that in Table A3. For example, A5 is Peru appears to have relatively within-municipality inequality. This likely reflects differences in the variability of monthly labor income relative to household expenditure and in the precise nature and quality of the various household questionnaires.'^^ Table A6 investigates the implications of controlling for population density (as a proxy for the density of economic activity). It uses the Theil index to of income (overall, predicted, tween municipalities/regions is and residual) into inequality (of countries), decompose each of the components between countries, inequality be- and inequality within municipalities, where income predicted using information on individual education, predicted individual experience, and nicipal level population density. Table to paved roads country- by-country. have municipal level Finally, Figures data A7 mu- presents the inequality decompositions for proximity Note that when considering proximity to paved roads, we for all countries presented."^ Al through A3 provide maps showing mean (non-deflated) labor income by municipality (or region) for North America, Mexico and Central America, and South America, respectively.^^ Note the difference in scale between the North America map and the Latin America maps. "^Note that because of tlie .small sample sizes of existing ronsumption data.sets for the United States and Canada, these countries are not included in the consumption inequality decompositions. Hence, the overall inequality numbers in Table A5 should be compared to those witliout tiie United States and Canada in Table 1. ''For example, the Peruvian consumption data provides somewhat larger samples than most of the other consumption data sets and contains many carefully detailed questions on expenditure and home production. To the extent that it is one of the highest quality data sets, it may have little measurement error in the withinmunicipality inequality component, relative to the within components from the other expenditure and income data sets. Canada is omitted because its administrative division involves extremely large swathes of territory grouped into single administrative units. This absence of municipality level data make it difficult to compare the decompositions for Canada to those for the rest of the Americas. 'Particularly for the countries where data are drawn from household surveys, labor income every mimicipality. In order to provide an approximate overall pict\ire of spatial patterns mi.ssiug municipality values by the median labor income in or department). 22 the municipality's first in is not available for income, we replace administrative unit (i.e. state References Acemoglu, Daron, _ and James , Introduction to {1), . Modem Economic A. Robinson, "Persistence Review. 2008. 98 _ Maria - ; Growth, Princeton University Press, 2009. of power, elites, and institutions," American Economic 267-293. Angelica Bautista, Pablo Qeurubin, and James A. Robinson, "Economic and political The Case of Cundinamarca, Colombia," NDER Working Paper. 2008. Inequality in Development: Aghion, Philippe and Peter Howitt, "A Model of Growth Through Creative Destruction," Econo- metrica. 1992, 60 (2), 323-351. Banerjee, Abhijit and Lakshmi Iyer, "History, Institutions, and Economic Performance: The Legacy of Colonial Land Tenure Systems in India," American Economic Review, 2005, 95 (4), 1190-1213. Barro, Robert Caselli, J. and Xavier Sala-i-Martin, Economic Francesco and Guy Growth, McGraw-Hill, New York, 1995. Michaels, "Resource Abundance, Development, and Living Standards: the Case of Brazil," mimeo, 2008. Center International Earth Science Information Network, Global Rural-Urban Mapping Alpha Version: Population Grids, Socioeconomic Data and Applications Center (SEDAC), for Project, Columbia CIAT, _ . Uni\'ersity, http://sedac.ciesin.columbia.edu/gpw (May 10th, 2007), 2004. Latin America and the Caribbean Roads Database, Center for Tropical Agriculture, 2008. Sub National Boundaries of South and Central America, Center for Tropical Agriculture, 2008. Ciccone, Antonio and Robert E. Hall, "Productivity and the density Am.erican Economic Review, 1996, pp. 54-70. of economic activity," The of Household Surveys: A Microeconometric Approach Hopkins University Press, 1997. Deaton, Angus, The Analysis Policy. Johns _ and Salman Bank Development to Zaidi, Guidelines for Constnicimg Consumption Aggregates for Welfare Analysis, World Publications, 2002. Dell, Melissa, "The Persistent Effects of Peru's Mining Mita," m.anuscript, MIT Department of Eco- nomics: http://econ-www.mit.edu/grad/mdell/papers, 2008. _ , Benjamin Jones, and Benjamin Olken, "Temperature and Sectional and Panel Estimates," Income: Reconciling New Cross- American Econom.ic Association Papers and Proceedings, 2009, 99 (2). Donaldson, Dave, "Railroads of the Raj: Estimating the Impact of Transportation Infrastructure," manuscript, London Schol of Economics: http://personal.lse.ac.uk/DONALDlS/, 2008. Duflo, Esther and Rohini Pande. "Dams," The Quarterly Journal of Econom.ics, 2007, 122 (2), 601-646. Gasparini, Leonardo, "Different Lives: Inequality in Latin America and the Caribbean," et al.. ed.. Inequality in Latin America: Breaking with History, IBRD, 2003, Gibson, Edward L.. "Boundary Control," World Grossman, Gene M. and Elhanan Helpman, in De Ferranti Politics, 2005, 58. 101-132. Innovation and Growth in the Global Economy, Press, 1991. 23 MIT Robert and Charles Jones, "Why Do Some Countries Produce So Much More Output Per Worker Than Others?," Quarterly Journal of Economics, 1999, 114 (1)5 83-116. Hall, Heston, Alan, Robert Summers, and Bettina Aten, "Penn World Table Version 6.2," Center for International Comparisons of Production. Income and Prices at the University of Pennsylvania, 2006. Hijmans, Robert et al., "Very high resolution interpolated climate surfaces for global land area," International Journal of Climatology, 2005, 25, 1965-1978. Kanbur, S.M. Ravi and Anthony Venables, Press, USA, Spatial inequality and development, Oxford University 2005. Klenow, Peter and Andres Rodriguez-Clare, "The Neoclassical Revival in Growth Economics: Has It Gone Too Far?," NBER Macroeconomics Annual 1997, 1997, pp. 73-103. Londono, Juan Luis and Miguel Szekely, Applied Econom.ics, 2000, 3 (1), "Persistent Poverty and Excess Inequality," Journal of 93-134. Mankiw, N.Gregory, David Romer, and David N. Weil, "A Contribution to the Empirics of Economic Growlh," Quarterly Journal of Economics, 1992, 107 Naritomi, Joana, Rodrigo R. Scares, and Juliano J. [2], 407-437. Assungao, "Rent Seeking and of'De Facto'Institutions: Development and Colonial Heritage Within Brazil," NBER the Unveiling Working Paper, 2007. NASA and NGIA, Shuttle Radar Topography Mission 30 Arc Second Finished Data, National Aero- nautics and Space Administration and the National Geospatial-Intelligence Agency, 2000. O'Donnell, Guillermo, "On the state, democratization and some conceptual problems: A Latin American view with glances at some postcommunist countries," World Development, 1993, 21 (8), 1355-1369. Romer, Paul M., "Endogenous Technological Change," Journal of Political Economy, 1990, 98 (S5), 71-102. Tabellini, Guide, "Culture and Growth :::,,- ;._, in the Regions of Europe," manuscript, 2006. 24 •, W) cu ^ r^ o GO — ^O 'S > •^ ^ ro lO cvi o LO lO 00 LO lO 00 00 C3 CO lO d S LO d 1 w ca g az) o o u bO H su -o cS QJ 00 lO o o ^ +J g o 00 00 o oo LO CM CD d 3 o '5 QJ a, ^^ ^ ^ >^. u CM CN ^ 5 > > Oo 'a 3 CM r^ '» d CM d >-. uh ^ ^ o ^ CJ o LO lO GO C^J C-) 3 CI, o en CN o 1^ LO CM d CO 2 s o "E d cu r^ LO CM o CM LO CO CM o — , CO o ^ 3 o g ti C-l I^ 1 _ o o CD < o ^ O CM rr^ .y <" H ^ 3 a! ^ i) ""-^^ ^ CJ "5 -U a) 2 a o CO X) -* CO LO CM 00 LO o d 00 o Oi LO O 00 00 CM C~) o o _p d LO r- CO o o o S .^ ^ U( "p^ s ^ -^ ^ I<* CO lO o o ^ t- o O -- s c; S qS =3 — ! ^^^ o 00 O o o d LO 00 o d >, ^ 3 o CN ^~^ cr; C-. oo CM C^4 00 CM CM o C-1 CO CO CM CM CM o n '-3 p: d q r-l LO CO CM CO CM d d O o CO LO f^ CO CM CO o CM d 00 b- d o d a o en ts ^ o CM ^ j3 3 -^ ni S u S >. _2 o a m OJ i- g J2 c-j M ;; cr 2 S o c^ d 3 <5 .-- LO '^ >v C -D C of -3 — L- _^ —o tp c; p p O CM CO CM < d o f^ d o UM , ui cu .-r^' c .•€ y: 3 o c .2f s Cj > QJ "5. d o o c >- 3 CO -w 72 CD ' ' ffS < c 2 Pi H 1 < B" a 5^ —o < o CJ 3 5 l2 < O — a ~ ,-^ C/5 ct 03 0) ci3 1 QJ '3 "bib 01 2 bC >, 1-H a ^ ^ 2 5 Q O <1J c CO -J hC ^ !-:> (^ o ^ •o 0) ci — ^ 'O -* 10 ro , 10 ^ 02 02 N00 00000 •^ 3 Csi Ol 00 CO -* ,-H ^ cs -* CM o O t^ CO CO CO -* ^ d * d t- CO CO "^ 10 d d d •^ LO -* CM -* d d CM C 2 ii § S >— hO s ^ ^ 3 c^ ^ ~^^ -^ ^ *-:' m ^ —1 00 1.0 O' CS •^ 000 00000 is ^— CN r^ -Nf CO LO I-- 10 o o o o o o o To g — QJ ^ ^ ^.' O c^ cu > -^ o o o CO CO "t— =J CO CO o d CO o ^ a r^ d c ^ 3 p ^ "O 3 3 *J bC c .^^ aj j; CO -^ CM C^l l-O o -rf o CM CM •* CO "; d 3 5; ? ^ ^ "u 00 — — 1 d 3 > O 10 O CN O o q CM o q o cri X' d q d '^ "^ ^H t^ d CO CM LO CO d d ;§ •^ — --H 00000 5 CO o o o CO LO CO 00 ,0 ^w- r>' ^ .= S o -S j5 £ X cj oc ;__ e to ^bC ^ — ^ -.S cC ^ V y > .> ^ < >-. Uf LO -i^ CO CO i-~ t~- CO LO 1—i S '0 ~^ H -^^ "-t^ K ^ c:; CM 0.708 0.629 LO LO CO CO d '^ CO 1 — q -f CM -r LO LO CO LO LO 00 cr: LO CM d d c CO CN 10 cd ?: .4-3 -o ^ d M 3 8 3 5 g 0) •^ o ? q d 11 ^ M ';j j^ II Si )S E o CO CO D3 -. > k LO LO 00 CM -H -—^ 1 r^ LO LO d d d d d d u 2 — — CO CO LO CO ^3 — 1 d d d d d d d d en ^ ^ c^ -71 N Q> — ^ >. ^ CO ;> ...^ CM CO CO c ^ CO .— .tH "T*^ CO CM — t^ CN — — — r~ (/i r-i 5 iB ^ , d ^^ CO 3 0? S CT 55 j5 ^s N ^ C/1 D 2 ,--, ±, (5 s Oh D > ^ '3d m Pi S ^ a^ tf 4-; ^ X) S S- "cS ^ < & tj cr CQ Oj Is y rr; fab GJ K , , u < ^ < fH ;:d C OJ s^ 2 -^ C7 Q> ^ - .S "11 otn .^ , ^ 13 -^ r-H Co "s CC bb ^ < i) Cv J: i~> CQ "3 ^ >> — , CO u LO CO ^ c ? 1 s CJ <; "? -T^ o JJ -c .2 S X c c o 2 " O 5^ -H P o H o g^ 2° 'CO G O '~^ U i" £i . — '"' > _QJ a; o c s ^ q J^ '3 C^ g -H o o _ — — ^ — oo ic £_ . c o x: o s J^ "ti ^o § jt Cm .2 _y '5 CO c? £- o Oj •u 1^ o aj ' "^ |£ *"; CO CN c; S Q o "^ to S =S CN o P o P o ^ J "ct ^ CQ OJ *J to 0/ O X C 3 O o s ou ^ ^ QJ c^ CI, ci c; to o I00 o ^6 oo o LO it! 7j ro >r. a o io CO O 2 c/: ^ o o c-0 U o -* t1^ K s ^ o 0) rt S s ~ ^ u c CN to X o C2 o in CO ~- CO o i-. o O TJ* o d "3 o -% CM CO ^ CC; to LO lO to oo oo I00 to o o S3 3 3 3 U '^ -: L-O 0) in '^ -S 3 s 3 >-« oo 4^ i-< 0) 1 "2 is s 5 HI X 3 o d oi ^ *„ ^ 5 t^ 'X — ^ ^ to , v: -o ~c -g 'o >> -J 3 C 3 ^ —1^5 0^^ hp - o ^ to ES 15 en > oo j:; - o b > CO "cS en CN c3 o CU 2 3 o 00 ce c S *G O o a. 1 to o (J 3 "o • u E-s if' 00 ::; 3 -a "3 o c o OJ -a y-r c rt lij 3 S CO 1 .™ ^ c LO bp f^. M i c: o oo X ^ CM CM OO ^ to oo OJ o 5 I X I O *j -^3 c -5 S LO 3 3 S -c 3 X' '3 S P 13 o — Cj 1 c5 'x ^ eS CC s (Ih a S; 3 O) SI a; I Ji cH CO C; < a. D. 5 cc$ ^ "O r3 OJ -M .-ti 3 o c- - CO CO o cd .=^ ^ cs o „H OI O d? 5 c .2 >1 -^ x: — 35 -a 6 CO o ^ bO O .S a> c^ .^ a. ^ 0) c o .c « 5 en 5 ^ p a .2 ai Z c) Q QJ 3 >> ^ CJ -c td Js 2 ' 0; 2; a 2? QJ 3 n o o bJJ a> cc; g-^ OJ :e ^ *^ s ^s 3 -0 CJ 3 3 3> C a § 'S o ^ 4 CCS CO o CO 3> •— a u a IJ -0 a fco § cs a ce b 3 J3 3 O bc g" -c ce > Ci 3 ^ "O 5 UJ q; 7: c ^ cfi i; "O CO CD ^ ir. a CD ce u W > a bO M-l CJ — en 0) Cd Je « CO „, bO ce CD 9;' CJ nj je CO CD CO -a T3 ccj cfl >-) a- z 5, -^ £ ce U> q; "^ U o; K " 3 ^ Cj a -^ 'u CO «J C/3 hSj C c 14 2> 1 1; T3 CO cu LO — -^ CO S en C/J cn = "(5 c^ SJ o -r "O QJ cu g — ^ LO 2 X o s u o 00 T, o bb r- o 1 *i CO CJ 3 CJ « ^ "3 CO CO CJ a M CO rt ."^ rM .^ S CM O K^ CO o ^ 0- K o -a c ci CD ho o o. IP -0 i:! O '5 ji -0 0) 3 CO Ci CJ cS S 0) o OJ •— cC a a. fc CN ^ o i C M re c CO -a "2 "3 cS c g O G cS D:^ -a a w 3 > 5' en .^ CO cu 7^ a> CO E 5 *y5 CO 5 bO a ^ 1 ^ D. ri cp a; ^c« CO OJ c^ !fi cu b£ X! -o >> c; cti 3 -O .2> 3 a > U .5 'u CO Z 1 i3 OJ CO c; 0) cfl '"' W 1 T3 ]G _o 1 ^ '^ cc: 0) Z -a W (P ce 'T^ J5 .^ 3 > a -S *^ CO CD ci^ '0 .i. a S OJ — ^ ce s. ce L- 2. '^ 1 P Ml 3 o S-2 > -0 CJ g a CD CO cu bb cS CO C CJ CJ ce nj I ^ >-. OJ -c "p o -a a Dq 3 CJ 3 C c II a cS jti X CO 0) y: 5 _!, ~ X 3 Z ^ ji ce 4-3 CJJ CO CJ X CJ CJ L- ^ -^ "m CJ CD CO s CJ :s CO Je a 3 a ^1 ce ^1> z U 3 ce ce U a a 1a bb r^ !5 ^ 1 — s ^ £? s? •-D lO Tj o 33 o o U o o O 'Ij 'Ij CJ JS o CO H 2 O IX tu _S & — E c. i^ OJ a ^^ o Oh a CO ^ ^' CO CO 1.C •m c -^ — OJ c8 a. g r^ u ^^ "O ce ;5 =t l/l "^^ -0 1— 3 ce l> S en z ~ 3 en i en o 3 ^ yj ce re *j CD cu 3 — u ^ " 3 ccJ -. CJ CO C £ a > a fe Uj CJ c3 cfl ^ C >-' "a 3 u 3 4J :§ 1 ~ 3 c ca^ a s a CO Cj 1 '/) .2 ^ — •r ce '13 -^ .^ a^ CS 3 a -*j Ci 3 'E Ci. p s (£ ^M crt CO Cj l- ^ *j t3 '+3 s a> 'n 3 CJ CO ce CO -2 OJ ce s d i a X a bO S c^ _S^ ce CO CJ § 3 S z ctf > 3 ss CD t; c ce -J t: y. ^ yj o ^ _i ^' 0/ OJ Cj CO '" — bb 0) s Jn K 1? '/I O , bC -c ;S 3 CJ 73 u ri -a :S O QJ s ^ Q, ~ CO - _^ CO Jl CD k D- r-i w LO CO ^ -^ S S OJ CJ 3 ^ "E p 3 a 2 ce bb C* CS S >-i 2 '5 rt o U "o Da (C m (J ? O CJ 2 (-1 S 'O cC "3 Q o u H "re > S 13 w •0 Cfi ifl c CD re D s re CO T3 X 5 c >•, ce w C u cc re 3 S a, fc rt C CS "3 3 3 N U) CD re 3 3 D S > CD "E. CO Table A2: Summary Statistics Income per Work:er Bolivia Brazil No. Males No. Mean Mun./ Obs. 18-50 Mun./Reg. Reg. Pop. Munic. Pop. (1) (2) (3) (4) (5) (6) 8,166 4,227 106 108,897 yes 8,152,620 3,481,697 1,920,149 1,519 94,823 yes 175.552,771 30,689,040 R,ef. to Country 441,740 196,208 11 2,695,307 no Chile 14,879 6,952 2 7,551,517 no 15,153,797 Colombia 18,479 8.270 9 4.665,758 no 39,685,055 Costa Rica 5,699 6 no 3,710,558 12,920,092 Canada Ecuador 22,275 10,581 20 628,822 no El Salvador 22,937 10.796 64 64906.51 yes 6,122,515 Guatemala Hondm-as Mexico 11,440 5,707 226 41,901 yes 12,820,296 13,160 5,978 98 44,973 yes 6.200,898 2,660,016 1,562,092 2,442 41,390 yes 100,087.900 Panama 94,645 55053 30 40,776 yes 2,836,298 Paraguay Peru 6,867 3,441 175 26,820 yes 5,585,828 22,207 11.333 610 30.619 yes 27.012,899 7,401,156 3,272,003 2,071 126,211 yes 284,153,700 8,082 3,707 19 141,812 no 3.334,074 677,524 380,797 219 110,118 yes 23,542,649 14.910,969 7,457,300 7,627 United States Uruguay Venezuela Total See Appendix Table Al for variable sources. 799,871,887 I i 1 -^rooosocootooccot^ooo-^csi oco^^^-cNcoocr^oc-^^oc>Dc^cn ( LO CO 00 oo CO UO -+ •^ en .— CO LO CO "* CO < OOOOCJOOOOOOOOO ( I ( 1 LO LO o o o o ! LO 00 00 00 LO 'Tf LO co IT) —< oo LO LO t-- , 11 1 -rr to or tu H w OJ UJ •^ "a +J is m r IV to ^^_ o ^" O LO -* o o -f -^ to oo CD CD O CD c q d d d d d d ^ o o o o -* LO Oi •:}" LO 00 XI t~ OO CM •* CM lO 00 CM --I O O ooooooooooooo --I -s J ^ rt LO to oo o o o o o o to CO CJ5 -H =s< C-l o o o o .< e-i p-i o O O o CS O ^ O CM CM O CM ci -. >, O CC O O o tt o 00 lO CO o o ci o c:^ L.O 00 o O c o CO tt 00 ci -i< O c^ o — 1^ o 00 ^ CM o m 00 CO t— o o o o oo oo •r-^ l6 O o o CM CM CM CO 00 CO LO CO o o o o c5 o 00 LO to .— LO CM LO -r 00 o ^ to CN r— d d d cN a> Tf LO CO CO o o 00 lO CM CM d d >, 3 O UO LO 00 00 Jv-; C'l CNl CM d d ^ o ^H 1^ — CO LO LO CO CM CN CO t^ C--1 CO CO CO -* CM CO a) LO LO 00 05 CD LO 00 O^ CD CO 1 [^ CO CO 00 o CD bJj s CO o o ^ O OO o CO LO t^ o o o o o t--O0'-H—lCM05C5-*r-IOOLOOOl-0 1^ r* CM r-- CM c: o o -^ o 00 CM L.O CM CO CO LO LO o dddddddddddddd 'Tf c:^ o c^ LO o o o M K a) K COC2OO50i5O'-HCCr-l00 1— cd tICMO^COOOOOCOCO lOCM—(--hOOC^JO CM O 00 o <D ^ o o CM LO 00 LO CO — o o o CO o o co o o o o o o o o o CO o o CD C^l < ddd'dicicicicidici o o o O 0) o o o o ^ o 1 ::5 o o CO r-i— .-* -^ -3 COOOOCMlOO'— Olt^OOCTiCOr-tLOO r'. i3 I O O V O lO t- CO O ^ ^ o O to O CM CN t^ oo LO 00 000C200000000 u ?) I> t- LO -tT n) r^ 2 -2 to -U C75 C-l O CM r-( O CO CO 05 CO >— 1>- CO GO CO d O •Tf Tt< O CO CM CO CM CO LO n^ CO .-H ,-H CO CM CM d -_J_ t-J^ _ I __ ^ __ ^, O O td CO ^ CO r-W ' ' o LO en 00 c^ CM t^ CM --H __ 'CO — C^l -* _ CM to CO a> lO ^ o CM '^ C5 oo o x; ^ CO CO ^ O t^ CO O o Q O W crT ^ O — ,X" CO CO — -^ to LO ^-^T—4t-HT—l^-^i—I^Hi—I^HtCi— o CD 1 c:r2 dj .— CO CO 00 LO — CO 00 to to to t-- LO f^ CO to o M d d ^ . d O d d d o o d tLO to LO C7i CM .-H •* OJ CO LO C^) CN p:^ ca tco lO LO d d CO 1-^ CM CO CM d d f- +J o o 00 lO l^ CO _^ I o CM^ o. d" 1--" cvf C-1 I— LO f CO_ ^ CM — , o LO -* CO CM o CO 00 ^; CO CM T—i C5 oo d CO LO CM CO CO CN CM CM lO r^ o CO 00 CO 1O-l d d ^ cJ CO CM "* 00 o' d' — 1 CM ^ a ^ —^ CAl c/} r1 C o c 'S ^ 3 'z: g; 3 00 o o > s OJ S o K s tuO c3 pop. Li Cu Cu p- a 'S ri !=^ IS R 03 CJ O c en , 3 X O o O O a D K < ctual 5 > < a z. 3, Di 1 0) < "7^ _ CO 3 cri :d ef. CJ "3 3 o to niguay — "£b f~\ pop U.S. 3 CD nj 0) qual OJ cS 3 a tuO (e< o -c "§> 'E 1 5^ r— .4^ . tJ H. o < a z Cd -1 .— ' 1i cc; 1 ct --; & 0) t o t.> ;.i t-( /^ —1 CLh Cu a a Q Q O O - • I ( 1 t-- "rr -^ N- ^ M lO -^ -3" O tT O 1 — a^ oo rO (M 00 C^ — lO-^-^ooor^ O -^ lO t^ tO^GO lO LO IC : oo I CO lO o o o o — < o C> lO ooio T lO o o GJ S CJ H ^ 0) !li ci c '^ CO f- O O O (M 00 OCTj-^CN'-^O-hOOOCOCSIiOOO ffi o •* -^ " t-~ ooooooooodooo 5 > •-• 005QSI— ICSICT:!^—<^CN-HLOO (NlMCNOO-^l^OiCOO 05 ro ^ lO o oooiocooor^cot^LO O ^ a^ (M o (M * ^ — I lO O t-L-Ot-tcMt^rocMcri O'^O^COCOCOO to 00 LO o O oooooooooooo o -* CO UO L.O Q O O C 0! I .2 CM oo-^^lo;o-hcn 000C2000C; oooooooo o '-' O -H O d d LO CO LO d —O r~ r>n CM C2 o CJ t^ LO =5 CD oo •* CO h£ C o U bi3 'Ol^OOCiOCOOCO'O -:rO— OLOt^tOCNOOOCOOOO-^^ LO lO ^ Tf O ^ CN O Tf LO LO OOOOOCZJOOOOOOO ^ Cl !>) t-- LO' c '^ toooo5Csirot^cc>LOco-^050a> ? doooodoc csioj-^^coCTCsioaicocMi-o rtOOiM^O^CMCOOO— o o ^^ OJ 3 CO r^ 00 CO o o CM COO-^'^OCMCM'Tf LO LO CO LO 05 CD c^i LO LO CO' O dddcicDdicici t-^ LO ;o r~ -^ CM CO O o O o o c LO o O o o o di d> d d> di <M CO — t-H i-H - • d c; o LO LO rooo oo OC d d CO cs CD oo o o -zi o q d c a ~^ CT' o — oo O 00 O O CJ CM t- O t- O CO -lO 'i" o CO 3C' LO -5- -r lO LO CO OO lO o o o o o o o O o 2 -3 CM CM —i -^ o ^ l~ CM C; j^ ^ CM r- Tt 1— LO CM -^ -d LO d d t^ 00 -r T' —: —ICM''^--r.-H^ — COCM CO^,— l-~ o o o CO LO — O o o C3 -9< 00 CO CM CM rsi "5 LO lO LO -^ C^ C^l LO — 1 o o i_; -r^^ < r-- o ' r^ lO era o o — 1 05 CM O O ^ O o CO CM 05 CD — CO OC oo lO r^ '^ — O 00 CO oo C^ t- CN LO CN G5 CO 00 CO -^ CM CO CX3 cr^ 04 I^ — 00 lo LO LO lo ct. , c>i c~~i ^ o -^ oo lO o c^g C--1 1 ^H LO CO CD -^ r^ ^ C5 t- LO O --H CM C-^l CM -3 ^ O -r LO 2 o 05 1co O o o CM — ' g U o -— 4-^ > p a cS 3 o CO O 5 p~i o ^ _o 1-1 1^ Oh Ci- C^ 5 5 > S S o 2; a c X - c S c- -. '5) _c§ >> o en s 6 "5 "£ 5i CO o CO CO cf: oo LO CM r-H 2 ;:^ K a; O C' o O o a bc 5 ^ o _ a .i -e :r o a < < I CL a) c b£) o O DD 05 ^ ^ !M t^ OJ iC LO O500O5O5COCOO5C0 COOCOOOCOlOCJl-^ C05 CMCO IOCDLOCM05COI---0 CM-^ TT'^CMCOCMCSICOCO COCN bO t^ ^ CN ooooooooooo 3 _ CO t^ (M ld r~ ?<i lO 00 m oo o) -H CO CO o r~ COCOOOOC5CDO ;"-(oo— oocooop— ooooooooooo ^ CO -Tf c~j dcDod'ci'didid) 00 do CO r~- a ^ 10 — ri 0) Oi ;:; O to C5 CN e5 OO t^ OO CN -g* -^ C~0 C^l I-- -^ O CN O? !CN T-i •* t- 00 —(CSLOCsrocsicocsitNroco 'cr ^ ^ CO 1-- ooooooooooo 8 -Nf -- -^ — LO ^H Of) , 0 LO CO an CO tCM g< CO 000000 'g* CM o o o o t: j= n U __j OJ % 'fl CO CM O CM O d d T-H a CM CO 00 Tf CO ro O •^ LO o o d d CJ5 -3 3 X ^^ c 0) T3 -^ cS :3 > -c CU .3^ M CN 05 OO O 11^ CN o 00 O CM r— c o o o d o o o o o o ro in LO -^ t^ CS 00 00 CN CM Oi I 05CMOCMC35C0C005 CM C^ [^ 00 CM -*COCMCOCMCMrOCO '00 r-l 1.0 CO 00 CO CO CM C S -3 O d>c5d!cid>d>dc5 J2' cr Ml a T' ^ Cfi '^ > 60 r* -" CJ S g 2i ,a3 T-! 05 CD 05 CM O COOOCM— **^COCO oqocooooo ^t^rCOCMl-tiCMOO-^mO t--COt^C3 0C-^O^CC5'-H OOCMOOO—1,-H^O^ h- l-O r-H d> di (d d: d> 000 CO CO a) O CTi !-- o o d d OJ C) u; ^ 3 -1-3 f« r<i CM CM 00 TO lO 05 Tf CM CO cr, r-5 — 1 — to 10 00 C35 1^ C5 C^l CTl 00 CO CM CM CM C--1 CO t- CO CM CO en CO -t CO CO -V 1 l^ cc CM — 05 CM -* CO 0-) 00 tCO CO CO CO 00 02 LO 1 00000000 oooo<oooooo U C^J CM I^ o o o d d •>* T-( o CM o o o ^ ^ o o d d ^ ^*- " J3 ^ *3 " y H .> -rt^CMCDOOOOOOCOO-* r-i-rj<OOCMC7)OOC35LOCN100 ^ t— O LO CO -f 00 r^ o o c^i —r L.O cri lo f~ -Tf t:~- :>o C75 CO c^ l-~ c;^ o ^ d CMCOOOOJ.—iCNrOb-t-CM LOLodoodiodt^Looooo §ci '-r 10 or; l-~ --f '• 00 — 1 CM 'r- t~- L< ej CO CO CO 00 CO 05 LO CO 05 t--^ uo t-^ •; — 2 ^ d c:^ o CO LO 05 '-< CO 00 -r CO CO CM CO C^ CM CO Cj r-icMC^CMCMCMCMCMCMCMCM o -p r^ •r+^ LO en i^ CO 00 LO LO 00 t^ '^ ddddddddddd '^^ cr> 05 tV 06 a rj — O 05 00 CS CO CM i-O lO CO oo" CO CiO cm' lo co' CS^ -* -:f__ — u 3 S O ^ > ^ CQ Ui "§ o bO 'x o o o K "S r> >-•. ^ CO ^ Qi '.>! 2 ;^ 3 2 ct a; b£ rr. a. cd -1^ r^ Dh -^ <£; c*^ "n * i— cq a ^ , r*^ ^ ^ ^ 1 CJ ^ ^ 5 oJO 3 bC c; u, CD tJ D on U ii= J .2 -I tti . ^* g' 3 . c ^0 cS 3 o M) 0, CO "" *" .« K S '5b > -c :: o bC0) Oi ^ a; hn Pi rt Q a; <; u a r. 5 " ^ o d ^'' 0) d CO -tr 0> 00 CM 00 CO 05 (^ CO t» o_^ -r CD ^^ 10 1^' ^' LO i~-" C:5 a" •^ c M CO COO^lO^CMOOCCN lO 05 -^ t^ 1^ O LO 1> -^ s n _a; c < cd Pi — ro 5 "o CO ' . — ' O cj 0^ 1- O bO o -d 00 O CN O t^ CO 00 o o o o o lO '^ CM ic -* LO cr; -)-J ^ ^ :£) CSI lO CO 05 CO 05 o "^ o l-O o o o CO o o 00 lO * (M ro o ^ o -^ -^ ^ bO CM t^ O '^ O CO 00 ^ (M <M O O O O O O O O O O CJ ^ +^ o CQ cd C^ 2 CM d ^ ^ O O O O O O O >^ LO CO CS (N lD '^ 00 CSl '^ 0<1 CO CD (-1 — 4-2 CO 4_; c (— ^ 6 > -4-= OJ ti 00 o oi »— m o o a 5 S' 00 00 oo o o o o "2 i5 bd O r^ <w d CSl r^ -i-i OG CM O O CO ^H O O O --^ r-H C^l ^^^ r:j ^ t-- (M ,-H T-^ T-^ CD ^ tN O 00 CO LO O d >• "S g .." <i S S oD c; ^ 00 cs cs rr; -*-J Qj m -i-j a. r^ ^ o t-- -^ o o o o LO o o 00 o o o s Si 8 o _N l| >i -3 o CM 00 CO 00 CM O CN — — o CN o o o o o r-< ^ o s t- O O o o o o o o o o o p LO (M >> o — ^ O d — CM o lO C5 CM 1 I CM CM CM t-H 1 m o 1 CO o X ~ (V .> J= c t: Xo S 3 bC o b- ^^ LO oo LO CO oo CO CO CO c^ f~- ^ ai K^ c — O o o o o o o -ui P* ^ t-- LO CM CM LO o o o o o CM LO bCM CO o o o o o d CO — 1 s bC a S; 'O oo o CO LO LO CO O r^ ^ O O O o CD d o d LO CC T— .fcj> O C3 c s C-^1 o o d cr> o ^ o >> +J r; -ii i^ O CM t^ -^ TT 00 t^ CM ^ oo CO CJ -f 'TT 1^ t^ -^ CTi d d d d d 00 ^ CO d o LO CM I- O , >, -;J q; o CQ CM ^ t- i t ?: .t^ ^ c CO o CM o ^ < c o a '^ -tJ 3 S ffj CQ S Ch -^ a^ CO S g c D > cr — ^ tf CD -' o Ul D o Z -n ^ cc ^ a'i C^ < O -3 r; O OJ < CO 1' o <r ]§ a 2 CD OJ CO a Table A7: Proximity to Paved Roads Argentina Mean SD Dist. to Dist. to Road (1) MLD Road 90/50 Ratio Between Country (2) (3) (4) Index Within Country Theil Index Between Country Within Country (7) 8.5 9.7 3.2 0.436 0.413 Bolivia 73.5 83.0 6.3 0.609 0.513 Brazil 29.3 64.1 7.9 0.956 1.049 Chile 25.1 21.1 4.1 0.353 0.306 Colombia Costa Rica Ecuador 19.3 31.7 3.7 0.692 0.681 5,7 6.0 4.0 0.622 0.470 23.8 35.7 9.8 0.641 0.703 El Salvador 4.3 2.4 1.9 0.147 0.139 Guatemala Honduras Mexico 13.3 24.3 6.8 0.824 0.860 11.6 18.0 3.7 0.418 0.535 9.5 9.9 3.2 0.382 0.379 Nicaragua 21.8 38.3 8.6 0.915 0.886 Panama 12.4 23.2 3.0 0.643 0.756 Paraguay Peru United States 34.7 49.6 6.0 1.001 0.752 19.5 33.9 9.1 1.180 0.889 1.5 3.3 3.4 0.914 0.795 Uruguay 10.7 10.6 4.5 0.478 0.416 Venezuela 11.1 26.2 2.7 0.508 0.747 All (actual) 13.2 36.5 6.4 0.621 0.774 0.439 0.815 All (equal) 18.6 37.8 5.8 0.311 0.649 0.286 0.656 No 19.7 38.6 .6 0.240 0.634 0.249 0.655 U.S. (equal) See Appendix Table Al and the text for source.s. Column (3) pr&sents the ratio of the 9()th percentile of the proximity to paved roads distribution to the .50th percentile. Columns (4) through (7) decompose inequality, using the Mean Log Deviation index and the Theil index, respectively. "Actual" refers to weighting by actual popnjUvtion, whereas "equal" normalizes each country's population to be of equal size. The 34 final row omits the United States from the sample. o £ < o g; a o o s < S o O C 03 O U >< o 1/1 S O u o (fa < bJO • i-H Figure A3: Labor incomes in South America Mean Labor Income (PPP <4,000 4 000-7.000 7 000 • 1 0,000 10,000- 15,000 >1 5,000 $)