WPS6187


Policy Research Working Paper                        6187




     The Persistence of (Subnational) Fortune
          Geography, Agglomeration, and Institutions
                      in the New World

                            William F. Maloney
                           Felipe Valencia Caicedo




The World Bank
Development Research Group
Macroeconomics and Growth Team
September 2012
Policy Research Working Paper 6187


  Abstract
  Using subnational historical data, this paper establishes                          data and historical evidence suggest this is due partly to
  the within country persistence of economic activity                                locational fundamentals, but also to classic agglomeration
  in the New World over the last half millennium. The                                effects: colonialists established settlements near existing
  paper constructs a data set incorporating measures                                 native populations for reasons of labor, trade, knowledge
  of pre-colonial population density, new measures of                                and defense. Further, high density (historically
  present regional per capita income and population, and                             prosperous) areas also tend to have higher incomes
  a comprehensive set of locational fundamentals. These                              today, and largely due to agglomeration effects: fortune
  fundamentals are shown to have explanatory power:                                  persists for the United States and most of Latin America.
  native populations throughout the hemisphere were                                  Finally extractive institutions, in this case, slavery, reduce
  found in more livable and productive places. It is then                            persistence even if they do not overwhelm other forces in
  shown that high pre-colonial density areas tend to be                              its favor.
  dense today: population agglomerations persist. The




  This paper is a product of the Macroeconomics and Growth Team, Development Research Group. It is part of a larger
  effort by the World Bank to provide open access to its research and make a contribution to development policy discussions
  around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author
  may be contacted at wmaloney@worldbank.org.




          The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
          issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
          names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
          of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
          its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                        Produced by the Research Support Team
  The Persistence of (Subnational) Fortune: Geography,
   Agglomeration, and Institutions in the New World∗
                 William F. Maloney†                 Felipe Valencia Caicedo          ‡


                                        August 29, 2012



JEL: J1, N9, R1, O1, O49

Keywords: Persistence, Subnational Growth, Geography, Agglomeration, Institutions




  ∗
      Corresponding author Email: wmaloney@worldbank.org. We thank Daron Acemoglu, Laura Chioda,
Antonio Ciccone, Ernesto Dal Bo, Leo Feler, Claudio Ferraz, Oded Galor, Steve Haber, Lakshmi Iyer, Aart
Kraay, James Robinson, Luis Serven, Hans-Joachim Voth, David Weil, Noam Yuchtman and participants
at the 2012 LACEA-PEG conference for helpful discussions. We are especially grateful to Miriam Bruhn
and Francisco Gallego, and to Navin Ramankutty for their kind sharing of data, and Henry Jewell for help
processing the HydroSTAT data. Special thanks to Mauricio Sarrias for inspired research support. This
paper was partly supported by the regional studies budget of the World Bank Office of the Chief Economist
for Latin America and the Knowledge for Change Program (KCP).
    †
      Development Economics Research Group, The World Bank, and Universidad de los Andes, Bogot´      a,
Colombia.
    ‡
      Universitat Pompeu Fabra.
1        Introduction
              an was home to one of the largest concentrations of indigenous peoples in the
    Tenochtitl´
New World when it was conquered by the Spaniards �ve centuries ago, and constituted an
                                                                  an Cort´
urban agglomeration rivaling those of Europe. In the words of Hern´      es (1522):
         This great city of Tenochtitl´  an is as big as Seville or C´  ordoba...It has many
         plazas where commerce abounds, one of which is twice as large as the city of
         Salamanca...and where there are usually more than 60,000 souls buying and selling
         every type of merchandise from every land... There are as many as forty towers,
         all of which are so high that in the case of the largest there are �fty steps leading
         up to the main part of it and the most important of these towers is higher than
         that of the cathedral of Seville. The quality of their construction, both in masonry
         and woodwork, is unsurpassed anywhere.1


 exico City, erected on the ruins of Tenochtitl´
M´                                             an, remains one of the largest and most
prosperous cities in Latin America. This paper uses new subnational data from 18 countries
in the Western Hemisphere to examine the degree to which such persistence is generally the
case: Do rich (high pre-colonial population density) areas before the arrival of Columbus
tend to be populous and rich today?


    Most similar to the present work, Davis and Weinstein (2002) �nd persistence in Japanese
population concentrations over very long historical spells, and despite massive wartime devas-
tation.2 Other recent works suggest persistence in economic activity over thousands (Comin
et al., 2010) or tens of thousands of years (Ashraf and Galor, 2012). Such persistence is
consistent �rst, with the importance of locational fundamentals such as safe harbors, climates
suitable to agriculture, rivers, or concentrations of natural resources that, even if not used for
exactly the same purposes, nonetheless retain value over time (Ellison and Glaeser (1999),
Rappaport and Sachs (2003), Fujita and Mori (1996), Gallup et al. (1998), and Easterly and
Levine (2003)). It may also suggest the importance of agglomeration effects, perhaps arising
from increasing returns to scale (see Krugman, 1991, 1993) or Marshallian externalities

    1
                          an, Segunda Carta de Relaci´
        La Gran Tenochtitl´                          on (1522). Authors’ translation.
    2
    In a tragically similar case to that of Hiroshima and Nagasaki, Miguel and Roland (2011) �nd that
heavily bombed areas of Vietnam also recovered almost fully relative to non-bombed areas.
                                                      1
arising from human capital or technological externalities (Krugman (1992), Comin et al.
(2010), Glaeser et al. (1992), Bleakley and Lin (2012)) which may lead to path dependence
and persistence across time even after the initial attraction of a site has faded in importance.3


   However, working against persistence in the context of colonized areas, Acemoglu et al.
(2002) argue for what they term a “reversal of fortune:� areas colonized that had large
populations of exploitable indigenous populations developed extractive institutions that
were, particularly during the second Industrial Revolution, growth impeding. Following
Malthus in associating high pre-colonial population density with more productive and
prosperous areas in pre-industrial periods (see also Becker et al. (1999), Galor and Weil
(1999), Lucas (2004)), they �nd a negative correlation between pre-colonial population
density and present day incomes. This �nding has been influential in moving institutions
to center stage in the growth debate, and suggests that such forces can more than fully
offset agglomeration and locational forces in determining the geographical distribution of
prosperity.


   This paper revisits the persistence question at the subnational (state, department, region)
level for the Western Hemisphere. We focus on the Americas because of the availability
of anthropological and archaeological estimates of indigenous population densities before
Columbus at a geographically disaggregated level, the near universal colonization by one
or more European powers, and the diversity of subsequent growth experiences. We match
the pre-colonial population estimates to new data on present population density and per
capita income generated from household surveys and poverty maps. We then incorporate a
comprehensive set of geographic controls, including new measures of agricultural suitability
and river density, which we show to have explanatory power as locational fundamentals
determining pre-colonial settlement patterns.          Data at this �ner level of geographical
aggregation allow us to take a more granular look at the role of locational, agglomeration

   3
    For a discussion of the importance of these effects for the ongoing evolution of economic geography
among developing countries see World Bank (2008).
                                                  2
and institutional forces behind the distribution of economic activity. In particular, using
subnational data with country �xed effects mitigates identi�cation problems caused by
unobserved country or region speci�c factors arising from particular cultural or historical
inheritances, and national policies, albeit, now asking the question at a level of aggregation
where the relative influence of forces for and against persistence may differ.


    Our empirical results suggest that, within countries, the forces for persistence dominate.
Population density today is strongly and robustly correlated with pre-colonial population
density. Current per capita income, while somewhat more sensitive to the functional form
of the estimation, is as well. Combining these results with historical evidence suggests
that both locational fundamentals and agglomeration externalities plausibly explain why
such persistence should occur despite the violent interaction of cultures of entirely distinct
cultural, economic, institutional and technological characteristics. We do also �nd evidence
that the institutional dynamic forwarded by Acemoglu et al. (2002) at the country level is
at play. Using subnational data on slavery from the US, Brazil and Colombia, we show that
such extractive institutions appear to reduce persistence, although they do not dominate
the forces in its favor. The only two examples of countries with a signi�cant negative
correlation of present and past prosperity, Argentina and Chile, and the notable individual
subnational unit reversals occurring in a overall country context of persistence, seem driven
by idiosyncratic geographical and historical factors.




2     Data
    The use of subnational data to explore differential performance along various dimensions
is now well established. As noted above, Davis and Weinstein (2002) use regional level
data for Japan to document the remarkable persistence of population densities, highlighting
the importance of both locational fundamentals and increasing returns to scale. Mitchener
and McLean (2003) exploit modes of production and geographical isolation leading to

                                              3
differential de facto institutions as explaining differential growth rates across US states.
Banerjee and Iyer (2005) exploit the variation in colonial property rights institutions
across India to explain relative performance in agricultural investments, productivity and
human development outcomes. Within Colombia, present regional development outcomes
are shown to be affected by colonial institutions such as slavery, Encomienda, and early
                    ıa-Jimeno, 2005); and slavery (Bonet and Roca, 2006). Naritomi et al.
state capacity (Garc´
(2007) analyze how variations in colonial de facto institutions in Brazil led to different
public good provision outcomes in modern times.                Acemoglu and Dell (2010) examine
differences in productivity across Latin American subregions and postulate that the large
differences in institutions and enforcement of property rights, entry barriers and freeness
and fairness of elections for varying levels of government are important. Dell (2011) uses
                            u and Bolivia to demonstrate the long term impact of the Mita
district level data from Per´
on development through the channels of land tenure and long term public goods provision.4
In a kindred paper using subnational data across the hemisphere, Bruhn and Gallego (2011)
argue that differences in the types of regional colonial activities, whether engendering
extractive or inclusive institutions, lead to lower or higher incomes, respectively. Most
recently, Gennaioli et al. (2011) use subnational data from 110 countries to argue for the
overriding importance of human capital in accounting for regional differences in development.



2.1     Population and Income Data

We compile subnational data on pre-colonial population densities, contemporary population
densities and household incomes for the 18 countries in the hemisphere listed in Table 1,
which summarizes the data by country.


   Present Subnational Population Density : This measures present population per square
kilometer in each subnational unit and is drawn from a highly disaggregated spatial data set

   4
     Regional differences in institutional arrangements have also been documented in the case of slavery in
the US and Brazil (Degler, 1970), and sharecropping and women’s rights in Colombia (Safford and Palacios,
1998; Bushnell, 1993).
                                                    4
on population, income and poverty constructed on the basis of national census data by the
World Bank (2008) for the World Development Report on Reshaping Economic Geography.
Population is aggregated from the census by the present subnational unit and the density is
then calculated.5


   Subnational Income per Capita : Income in 2005 PPP US dollars is drawn from the
same spatial data set.“Poverty maps� are generated which combine household level data
sets with limited or non-representative coverage with census data to generate income maps
for much of the hemisphere (see Elbers et al., 2003). These poverty maps address the
                                    exico household income surveys are not representative
problem that in some cases such as M´
at the “state� level.6 Household income data are preferable to national accounts data as
a measure of regional prosperity. In the case of natural resource rich regions, income may
or may not accrue to the locality where it is generated and hence may provide a distorted
measure of level of development. As an example, the revenues from oil pumped in Tabasco
               exico, are shared throughout the country, although they are sometimes
and Campeche, M´
(but not always) attributed entirely to the source state in the national accounts (see Aroca
et al., 2005). This is a broader issue that emerges wherever resource enclaves are important.
For instance, from a national accounts point of view, the richest subnational units in
                                  u, respectively, are Tierra del Fuego (oil), Casanare
Argentina, Colombia, Chile and Per´
(oil), Antofagasta (copper), and Moquegua (copper), all of which, with the exception of the
last, are average or below average in our household survey measured income. Further, the
geographical inhospitability of these locales ensured and continues to ensure relatively little

   5
     Censuses: US, El Salvador 2000: Brazil, Panama; 2001: Bolivia, Ecuador; 2002: Chile, Guatemala,
Paraguay; 2005: M´ exico, Nicaragua, Per´u, 2006: Uruguay. All other countries: �gures correspond to survey
data estimates at the regional level or small-area estimates based on survey and census data.
    6
      We thank Gabriel Demombynes for providing the data. See original study for methodological details. We
expect that while somewhat more complete, our data is similar to the census based data used by Acemoglu
and Dell (2010). For Argentina, Colombia and Venezuela, the spatial data base reports the unsatis�ed basic
needs index rather than income. We project subnational GDP (production) series on this index to scale it
to household income. GDP source: Argentina (CEPAL, Consejo Federal de Inversiones, Colombia (DANE)),
Venezuela (Instituto Nacional de Estad´ıstica). We expand the sample to include Canada and the United
States using the (2005) censuses. The resulting estimates of mean per capita income have been rescaled so
that the population-weighted average matches 2007 GDP per capita at 2005 US dollars (PPP adjusted).
                                                    5
human habitation: Antofagasta is in the driest desert in the world and Tierra del Fuego
is the closest point in the hemisphere to Antarctica. This combination can give rise to a
negative, although relatively uninteresting, correlation of pre-colonial population density
with present income. That said, such correlations still emerge even in our income data
due to the selection of the population in these areas: The very small population related to
extraction of natural resources has relatively high levels of human capital and remuneration
and hence, we may still �nd that areas which the indigenous avoided are now relatively
well-off in per capita terms.


   Pre-colonial Population Density : This measures the estimated number of indigenous
people per square kilometer just before colonization. These data draw on a long tradition
of academic research dating from the turn of the last century, much of it fuel for the debate
over whether the colonial powers encountered a “pristine wilderness� or, alternatively, a
world densely inhabited by indigenous peoples subsequently devastated by disease and
conquest (Denevan, 1992b).     The estimates contributed by the authors in The Native
Population in the Americas in 1492 (Denevan, 1992a) are among the most comprehensive
and re�ned to date and they form the core of the data. The details on the construction of
the pre-colonial density measures and their mapping to modern subnational units can be
found in Bruhn and Gallego (2011). We expand the sample further using analogous data
on Canada from Ubelaker (1988), and Nicaragua from Newson (1982). Though the project
of estimating populations half a millennium past is necessarily speculative, the estimates
synthesize the most recent available geographical, anthropological, and archeological
�ndings. In particular, they draw on documentary evidence such as reports by Europeans,
actual counts from church and tax records, as well as contemporary and recollected native
estimates and counts. Depending on the country, projections across similar geographic
areas, regional depopulation ratios, age-sex pyramids, and counts from subsamples of the
population (such as warriors, adult males, tribute payers) are used, as well as backward
projections from the time of contact with Europeans. These are corroborated by evidence

                                             6
including archaeological �ndings, skeletal counts, social structure, food production, intensive
agricultural relics, carrying capacity, and environmental modi�cation. Importantly, neither
modern GDP, climate models nor current population measures are used in the construction
of these estimates. Figure 1 maps these pre-colonial population densities. While some
related studies have focused on cities as the unit of observation, such data are not available
at our frequency for the pre-colonial period and we work with regional densities. However,
as Davis and Weinstein (2002) note, for numerous reasons in particular related to de�ning a
city over time, estimated regional population densities are arguably preferable.



2.2     Locational Fundamentals

To establish the importance of locational fundamentals, we match the population and
income subnational data to a comprehensive set of geographical controls. Accounts of 18th
century explorers, and anthropological studies con�rm the importance to Indian settlements
of both arable land and waterways for food and transport, characteristics also attractive to
subsequent European settlers and potentially current inhabitants.7 We incorporate two new
measures to capture agricultural suitability and river density.


    Suitability for Agriculture : Since agriculture was critical to early settlement, we employ
a new measure of agricultural suitability as developed by Ramankutty et al. (2002). This
measure uses a combination of three different data sets. First, it calibrates the satellite-based
International Geosphere-Biosphere Programme Data and Information System (IGBP-DIS)
1 km land-cover classi�cation data set against a worldwide collection of agricultural census
data to capture cultivable land. To derive climatic parameters that may restrict the use
of this soil, a second data set captures the mean-monthly climate conditions including


    7
     Denevan (1992b) discusses the extensive evidence on the importance of agriculture throughout the
hemisphere in precolonial times. De Vorsey Jr (1986) cites the 18th century explorer William Bartram as
noting that “An Indian town is generally so situated, as to be convenient for procuring game, secure from
sudden invasion, a large district of excellent arable land adjoining, or in its vicinity, if possible on an isthmus
betwixt two waters, or where the doubling of a river forms a peninsula; such a situation generally comprises
a sufficient body of excellent land for planting corn, potatoes, squash, pumpkins, citrus, melons, etc.� p. 13.
                                                        7
temperature, precipitation and potential sunshine hours. Finally, it draws on the IGBP-DIS
global soil data sets that contains soil properties such as soil carbon density, nitrogen content,
pH, and water holding capacity. Combining these through a model of land suitability,
Ramankutty et al. (2002) generate an index of the probability that a particular grid cell will
be cultivated. We employ a spatial average of this measure over subnational units.


    Waterways and Coasts : For measures of the ubiquity of settlement-suitable waterways,
we employ the recently developed HydroSHEDS data that provide globally consistent
hydrographic information at high resolution as collected during a Space Shuttle flight for
NASA’s Shuttle Radar Topography Mission (SRTM). HydroSHEDS generates a mapping of
river systems from which we develop a measure of the number of potentially suitable river
sites based on the density of rivers in each geographic unit.8


    Clearly, populations could also be sustained by marine-based economies where farmland
and rivers were of less importance.             Hence proximity to the coast for saltwater trade,
transport, �shing potential and amenities potentially persists in importance, much as it was
subsequently for European settlement, and to capture this we employ a measure of whether
or not the region is landlocked.


    We also employ several other controls that capture suitability for human settlement as
collected by Bruhn and Gallego (2011): average temperature in degrees Celsius, altitude of
the capital city of the state in kilometers, and annual rainfall in meters. Some of these clearly
overlap the agricultural suitability measure and hence need to be interpreted as capturing
effects beyond those on agriculture. Table 2 summarizes the data.9
   8
     HydroSHEDS stands for Hydrological data and maps based on SHuttle Elevation Derivatives at multiple
Scales. The HydroSHEDS project was developed by the World Wildlife Fund and U.S. Geological Survey
among other organizations.Densities were calculated using zonal statistics in ARC-GIS map. Though
HydroSHEDS depicts the flow of cells into a given river system, beyond a certain size we do not take
into account the flow of the river per se for two reasons. First, settlements are not likely to be proportional
to the size of a river, again, beyond a certain threshold. Second, due to the geographical projection, the cells
do not map precisely one to one to actual flows.
   9
       Ashraf and Galor (2011b) have further suggested as determinants of population time elapsed since the
                                                       8
2.3      Institutions

Slavery: As a direct measure of extractive institutions, we exploit the data on slavery,
measured as the percentage of enslaved and “free colored people,� in the three countries for
which they are bnational historic censuses. For Brazil, we used the 1872 Census and for
Colombia the 1851 Census.10 For the United States, we used the 1860 Census as well as the
data compiled in Nunn (2008). To capture the broader influence of slavery, both in the year
of the census and in previous years, we include both slaves and the general black population
which would include now-freed slaves.




3        Empirical Results

3.1      Locational Fundamentals and Pre-Colonial Densities

Figure 1 and Table 1 present a map and summary statistics of pre-colonial population densi-
ties. What is immediately clear is the great heterogeneity of pre-colonial population densities
both within and between countries, as well as the substantial overlap of distributions across
countries. The Latin American countries span densities averaging from around 0.4 person
                                                                                     u and
per square kilometer for Argentina to 1.7 for Venezuela, 2.5 for Brazil to 17 for Per´
        exico. Further, Table 1 con�rms a large range of variances of initial density within
32 for M´
          exico and Per´
country. M´            u are not only dense on average, but have much larger variances
than, for instance, the US. However, overall, the US and Canada �t comfortably in the Latin
American distribution. With a mean population density of .39, the US is above Uruguay
and is roughly the same as Argentina. Canada, at 1.22 is above Argentina, Bolivia, and
Uruguay and is just below Paraguay and not so far from Venezuela. Looking at both mean
and variance, the US and Argentina are effectively identical: (.39, 1.34) vs (.44, 1.45).


Neolithic revolution, distance from the regional technological frontier, and absolute latitude. As their data
is at country level, these effects would be absorbed by the �xed effects and we do not include them.
    10
     We thank Jaime Bonet and Adolfo Meisel Roca for providing their colonial data for Colombia and for
                              on et al. (1994)’s compendium of colonial statistics.
pointing us towards Tovar-Pinz´
                                                     9
   As a �rst check on the relevance of our locational fundamentals proxies, Tables 3 and 4
report the results of running


          f (DP recol,ij ) = α + γLOCAT ION ALij + γ2 LOCAT ION AL2
                                                                  ij + µi +       ij       (1)

where DP recol,ij , is pre-colonial density, and f (.) allows flexibility in functional form of
the dependent variable. As will be discussed further below, the persistence �ndings are
somewhat sensitive to functional form and hence as robustness checks we run all our
speci�cations with pre-colonial density both in levels and in logs. LOCAT ION AL is a
the vector of subnational locational fundamentals and µi is a country speci�c �xed effect.
Though we report a speci�cation where fundamentals enter linearly, we also report one with
quadratic terms to acknowledge that human utility is not likely to be linear in locational
fundamentals, most obviously for temperature and rainfall. With altitude, also, gaining
some height may limit the likelihood of disease or invasion, but the bene�ts potentially
decline again after a point.


   Tables 3 and 4 respectively present estimates for each of the level and log formulations of
the dependent variable, pre-colonial population, both without and with country �xed effects
(FE). We report the latter despite the fact that the territorial boundaries and corresponding
national governments, institutions, and other characteristics clearly were not established
at the time for two reasons. First, the generation of the pre-colonial populations was done
with the present national boundaries de�ning the unit of analysis and by different authors
and hence there may be subtle differences at that level. Second, in subsequent regressions
we will care very much about abstracting from country wide effects and hence the analogous
speci�cation is desirable for reference.


   For robustness purposes, we also report the corresponding results using the MS (or M-S)
estimator (Maronna and Yohai, 2000). Our data combine countries with very different levels

                                             10
and variance of initial population densities as well as often very dramatic spreads within
            exico, for instance, has many states with modest densities, but then Morelos
countries. M´
     exico City are �ve to ten times as large. The MS estimator is designed to handle
and M´
potential good and bad leverage points (explanatory values with extreme values) as well as
conventional vertical outliers in a structured way that has been shown superior to other
alternatives like quantile or robust regression.11 The MS FE are our preferred estimates and
we base our calculations of maximum and minimum influence on them.


    Locational fundamentals appear important to where pre-colonial native populations were
concentrated, explaining up to 70 percent of the observed variance in the case of the log
OLS FE. R2 are not calculated for the MS estimator. Since the variance of the dependent
variable is attenuated by the log transformation, both the higher explanatory power in the
log forms, and the larger standard errors (generally less signi�cance) in the levels form are to
be expected. In general, the MS estimators generate more signi�cant coefficients consistent
with standard errors that are estimated taking into account outliers and we focus mostly
on these. That said, while there are important differences, the stories, particularly in log
formulations are not radically different between the OLS and MS estimations.


    Suitability for agriculture enters signi�cantly and positively in all but one log form
speci�cation (columns 2-8) and all levels form MS speci�cations (columns 5-6), albeit
with diminishing value: Pre-colonial native populations were attracted to good farming


   11
      The estimator is a combination of M and S estimates. In the case where some variables are categorical
(0-1) and some are continuous and random which may contain leverage points, as is the case here, M
estimates are not robust and S estimates are computationally intensive. The MS estimator combines both
and though less well-known than, for instance, quantile or robust regression for managing potential outliers,
it has several advantages. First, it is more robust to bad leverage points. Second, it is likely to provide
more efficient estimates of the standard errors than the bootstrapped quantile estimates since it adjusts for
outliers. Finally, it attains the maximum breakdown point, being robust to up to half of the observations
being contaminated. In practice, a sizable share of our observations in all pooled speci�cations are identi�ed
as outliers and hence a high breakdown point is desirable. The standard “robust� estimator in STATA is a
class of M estimator, however it is not robust, in particular, to masked outliers. That is, when calculating
whether a observation has a standardized distance above a critical value, it uses the variance calculated using
outliers. An upward biased estimate of the standard deviation may therefore allow a true outlier to remain
in the sample. The MS estimator obviates this problem.
                                                      11
conditions, but not necessarily the best we now know to be available. In both the log and
level formulations, the FE estimates are quite close in value as well. Overall, this is a
robust �nding. The negative quadratic term reflects the low population density found both
in the US Midwest and the high suitability areas of both Argentina and Brazil which, for
whatever informational or historical reasons, were not heavily settled. In the log form MS
FE speci�cation (column 8), the maximum appears at a probability of cultivation of .68
which corresponds broadly to Missouri, Misiones (Argentina) and Caldas (Colombia).


   The relatively high densities found in often arid relatively unsuitable areas along the
coasts (see Figure 1) may, again, reflect the marine, rather than agricultural basis of the
local economy.    Consistent with Rappaport and Sachs (2003), being landlocked enters
negatively and signi�cantly in the levels form MS speci�cations (columns 5 and 6), and the
log form FE and MS FE speci�cations (columns 3 and 7), although in this case it is rendered
insigni�cant by the inclusion of quadratic terms of the other variables. While there is some
concern that the coastal dummy both here and elsewhere may be picking up non-linearities
in other locational fundamentals, there is therefore some evidence that proximity to a coast
is associated with higher population density.


   Our measure of river density enters signi�cantly, although with somewhat inconsistent
influence. It enters negatively in the levels form linear OLS and MS FE speci�cations
(columns 1 and 7) and the log form linear MS speci�cation (column 5). In the log form
quadratic MS speci�cation (column 6), it enters negatively and signi�cantly with a positive
coefficient on the quadratic (column 6) but then enters with the reverse polarities in the
MS FE (column 8). In practice, the curvature is mild but arguably the MS FE is preferred
given the correction for possibly correlated �xed effects. Looking at Figure 1, it is likely that
the negative/concave tendencies are partly driven by the Amazon which, despite having the
richest network of waterways in the hemisphere, has very low indigenous density and which
                                                                                       u
includes a substantial number of subnational units in Venezuela, Colombia, Ecuador, Per´

                                                12
and especially Brazil. A similar pattern is found in the US where, despite relatively (for the
US) high population concentrations in the Mississippi watershed, particularly the present
day states of Mississippi and Louisiana, even higher concentration are found in Connecticut,
California, Massachusetts,and Rhode Island which show lesser river densities. The maximal
value for the log MS FE speci�cation occurs around the mean value for the sample at 3.3,
roughly the value in California or New Hampshire.


   Average temperature emerges positively in the levels form linear OLS, MS and MS FE
estimates (columns 1,5,7), and in the log form linear OLS and MS (columns 1 and 5). In
both the log form MS and MS FE quadratic speci�cations (columns 6 and 8), temperature
enters positively with a negative quadratic term. The levels form MS quadratic (column
6) somewhat contradictorily shows a negative free standing and positive quadratic term,
but when �xed effects are introduced, (column 8) the variable becomes insigni�cant. On
balance, the results support the idea that humans dislike cold but, after a point, would
prefer not to be any warmer. The optimal temperature in the log form MS FE speci�cation
                                                          exico City (M´
plausibly appears to be around that found in Virginia or M´            exico).


   Altitude enters positively and signi�cantly in the levels form MS linear and quadratic
estimates (columns 5 and 6) and log form OLS linear and quadratic, OLS FE, MS, and
MS FE (columns 1,2,3,5, and 7). The log form quadratic speci�cations introduces a slight
convexity that reflects the higher densities at sea level and then, especially, at the highest
                                                                     ı (Bolivia), as well as
altitudes where we �nd substantial Inca populations: La Paz and Potos´
                         u). Overall, within the limits of our sample, altitude is conducive
Huancavalia and Puno (Per´
to human settlement.


   The results for rainfall broadly support, again, a concave relationship with population
density: rain is desirable up to a point.      In the log form MS quadratic speci�cation
(column 6), it emerges strongly with a positive free standing and negative quadratic term,

                                             13
a result echoed somewhat more weakly by both the levels and log form MS FE quadratic
speci�cations (columns 8). It also enters negatively in the log form linear MS FE estimates.
Recalling that the agricultural suitability term has already factored in rainfall, this �nding
is telling us that desert is intrinsically undesirable, but that rainfall also has diminishing
value. New Hampshire appears to have the optimal level of rainfall.


   In sum, we �nd quite strong correlations between many of our locational fundamentals
proxies and pre-colonial population density suggesting that, unsurprisingly, indigenous
populations were concerned with agriculture, �shing, transport, being warm enough but not
too warm, perhaps avoiding diseases, and not being too dry or too wet. Of equal import, the
exercise suggests that these are credible controls to be used in the subsequent regressions on
persistence.



3.2    Persistence: Overview and Speci�cation

We next explore the correlation of pre-colonial population densities with present population
densities and with present per capita income. Again, the summary statistics for all three
variables are found in Table 1. As a �rst look at the data, Figures 2 and 3 offer a striking fact.
For the US, both today’s state level population density and income per capita are positively
correlated with the density of the indigenous population before the arrival of Columbus:
economic activity appears to persist. In the next two sections, we document these correlations
more rigorously for a broad set of countries of differing pre-colonial densities and present per
capita incomes. For each dependent variable, we begin estimating country by country


                         f (D2005,ij ; Y2005,ij ) = αi + βi g (Dprecol,ij ) +   ij             (2)

where D2005,ij , population density of subunit i of country j, and Y2005,ij , present per capita
income, are sequentially the dependent variables, Dprecol,ij is pre-colonial density, and f (.) and
g (.) again allow flexibility in functional form. In addition to the log and levels speci�cations,

                                                   14
we also report the rank correlation coefficient (effectively another transformation) which
makes our results directly comparable to those of Davis and Weinstein (2002).12 Finally, we
then pool the data and estimate the within parameter β :


            f (D2005,ij ; Y2005,ij ) = α + βg (Dprecol,ij ) + γLOCAT ION ALij + µi +    ij        (3)



3.3     Current Population

3.3.1   Evidence for persistence in population

Table 5 reports β from the country by country regressions of present on pre-colonial
populations density. The estimations are in OLS due to limited observations and hence
are subject to the outlier concerns discussed above. However, they appear to con�rm that
the positive relationship found for the US is signi�cant in both the log-log and level-level
speci�cations. In fact, they suggests it as an overall stylized fact for the hemisphere. In
the log-log speci�cation, 15 of the 18 countries show a signi�cant and positive elasticity.
Canada is the only one to show a signi�cant negative coefficient, largely driven by the Arctic
Northwest Territories, Yukon and Nunavut which have relatively lower population densities
today. For Colombia, El Salvador and Guatemala, the elasticity is above 1 suggesting a
concentration of population across time. In the level-level speci�cation 11 of 18 countries
show signi�cantly positive relationships with, again, only Canada showing a signi�cant and
negative coefficient. 14 of the 18 countries show signi�cant and positive rank correlations,
12 show correlations that exceed .5, and Chile, Guatemala, Nicaragua and Venezuela all
exceed .75. Overall, the magnitudes are broadly similar to the .71 found by Davis and
Weinstein (2002) for Japan over the period CE 725 to 1998. Consistent with the lower
coefficient in the log-log speci�cation, the US is among the lowest of those showing a positive
and signi�cant correlation at .37. And again, Canada is the only negative and signi�cant
entrant. In general, the Latin American countries show far higher degrees of population

  12
     Davis and Weinstein (2002) report the raw correlation coefficient rather than OLS coefficients. We do
not do this for reasons of comparability to the multivariate regressions that follow.
                                                 15
persistence than the US or Canada. This may partially reflect the dramatic differences in
immigration experiences at the country level between the US and Canada on one hand, and
Latin America on the other which, with the exception of Argentina, was relatively closed.13


    Table 6 pools the countries. Here, �xed effects are potentially of greater importance
than in the last section because of the desire to control for country level historical effects
or policies that would affect the between dimension, and we focus primarily on the FE and
MS FE estimates. That said, the OLS regression estimates (column 1) are positive and
signi�cant in both functional forms and the between estimator is positive and signi�cant in
the log-log and insigni�cantly positive in the level-level form (column 2). The ‘�xed effect
estimator (column 3) generates a strongly signi�cant coefficient in both the level-level and
log-log speci�cation, the latter yielding an elasticity of .4. The results remain very similar
when the sample is reduced to reflect the unavailability of locational fundamentals for
some countries (columns 4) with the elasticity rising to .5. The MS FE estimates (column
6) also remain strongly positive and signi�cant. Strikingly, despite an important drop in
the magnitude of the coefficients in the the MS level-level speci�cation, every �xed effect
estimation in each of the two functional forms yields a coefficient on pre-colonial density
that is strongly signi�cant and positive, with the elasticity of the log-log speci�cation on the
order of .5. Population density shows strong persistence across time.


    We include the fundamentals in quadratic form to absorb up as much fundamental
influence as possible and this lowers the magnitude of the persistence coefficient somewhat
but leaves it strongly signi�cant: in levels, the OLS FE estimates do not change (column


    13
       From the point of view of establishing the particular channels postulated by the reversal of fortune
literature, it may be argued that capital cities have a sui generis dynamic and should be excluded. From
a general point of view of understanding agglomeration effects and persistence, this is less clear- whatever
the impetus that established these cities, the existing megalopolises in Latin America are not supported in
the main by government activities at present. Precisely the emergence of such “Urban Giants� has been
studied by Ades and Glaeser (1995), while Krugman and Elizondo (1996) have focused on M´          exico City.
In the end, even dropping these overall strengthens the persistence results. The levels levels regression for
Argentina, Brazil, and Guatemala become signi�cantly positive and nothing becomes less so. We thank
Daron Acegmoglu for bringing this point to our attention.
                                                     16
5), but they substantially do in the MS case (column 7) where they fall from 3 to .6
reflecting the treatment of outliers. In the log form, adding the locational fundamentals
drops the coefficient substantially less, from .5 to .3 in the OLS FE (column 5) and from
.5 to .4 in the MS (column 7). Part, but not the majority of our �nding of persistence
is thus due to persistence in fundamentals. These fundamentals enter broadly similarly
to the way they did in the previous exercise, although with some important differences.
Agricultural suitability is not signi�cant in any speci�cation con�rming that it is not the
driver of population agglomeration in the modern world.          Rivers emerge with roughly
the same degree of signi�cance as previously with a negative quadratic term in the
levels form MS speci�cation and with both coefficients signi�cant in the log form MS
FE. Landlocked is negative and signi�cant only in the levels form MS FE speci�cation.
Temperature enters convexly, again, although generally more signi�cantly. Altitude is gen-
erally insigni�cant except for entering negatively in the MS log speci�cation. Rainfall again
appears signi�cant in both terms in the OLS FE level, MS FE level and MS log speci�cations.


   In sum, many of the locational fundamentals emerge statistically signi�cant in deter-
mining current populations and with similar sign as they did in explaining pre-colonial
population density although, taking all the variables together, the explanatory power is
not as high as it appears to be then. The maximum R2 is .43 in the log FE speci�cation
compared to .7 previously. Critically, however, despite a quite complete set of locational
controls, the pre-colonial densities themselves appear to be robustly signi�cant. In the
log-log, speci�cation, for instance, the pre-colonial coefficient falls only 20%. Much of the
persistence appears to exist for reasons related to the existence of the populations themselves.



3.3.2   What drives population persistence?

What might these reasons be?         In Latin America, native populations were indeed a
source of tribute and labor and hence it is not surprising that Spanish cities would be
built near existing population centers, whatever factors drove their initial settlement.

                                              17
However, in the US, pre-colonial native populations were relatively small, topping out
at around 2 people per square kilometer, and they were generally, with the exception of
South Carolina (Breen, 1984), not exploited for tribute or labor by French, Anglo and
Dutch colonizers. This suggests that while the argument that the Spanish and Portuguese
located near indigenous populations for purposes of tribute or forced labor through the
Encomienda or Mita is compelling, it is not the only mechanism through which pre-colonial
agglomerations were perpetuated. To begin, throughout the New World explorers depended
on native cartography and knowledge to map the relevant geographical and demographic
sites (De Vorsey, 1978).         New settlement was likely not to be random, but influenced
by the previous “known world�. The Spanish further needed the knowledge and skills
                                           es employed the stone masons and architects
accumulated by the native populations. Cort´
                                                              an to remodel Moctezuma’s
of the pyramids, canals and aqueducts of vanquished Tenochtitl´
palace into his own, and to raise the most important city in the New World from the ruins
of the Aztec capital. The large population of craftsmen and artisans was of world caliber
(Parkes, 1969). The conquistadors more fundamentally needed those with a knowledge
of plant life, agronomy, and hunting to feed their new towns. Hence, just by virtue of
                                                                   an was attractive beyond
already supporting a civilization in all its dimensions, Tenochtitl´
the brute labor force it offered and in spite of its actually lackluster locational fundamentals.14


    In other regions under Spanish colonization, the native populations were valued for
otherworldly and strategic reasons. The missions set up along the Alta California (now
US) coast-San Diego, Los Angeles, Santa Barbara, San Jose and San Francisco-were


   14
                an’s location was allegedly determined by the god Huitzilopochtli to be established on a small,
      Tenochtitl´
swampy, island whose chief attraction appears to be that it was uncoveted by the neighboring tribes and was
defensible. Parkes (1969) notes that the Mexica (Aztecs) were the last tribe of seven to enter the valley and
wandered as outcasts, selling their services as mercenaries to the dominant tribes, and eating reptiles and
pond scum to survive. They had the worst pickings of a not entirely favorable locale. The valley of M´   exico,
and in particular Tenochtitl´ an, had unreliable weather, with a short growing season and frequent drought.
Famine was not uncommon. The lake was subject to storms and a major flood in 1499 caused the loss of
much of Tenochtitl´  an (Thomas, 1993). Simpson (1962) notes that “With the silting up of the lakes and
consequent flooding, the city was frequently inundated with its own �lth and became a pest hole. Epidemics
were a scourge for centuries and were not brought under control until the opening of the Tequixquiac drainage
tunnel in 1900� (page 164).
                                                      18
established beside major native population centers (as in the Southwest) to recruit souls to
Christianity, but also to create colonial subjects to occupy territories perceived under threat
of English and Russian encroachment (Taylor, 2001). In these cases, it was the population
agglomeration itself, rather than the locational fundamentals per se, that were the attraction
and exploitation was not the primary motivation.


   In non-Spanish North America, the competing colonial powers also established many
cities including Albany (Dutch), Augusta (British), New York (Dutch), Philadelphia
(British), Pittsburgh (French and British), St.       Louis (French) on or next to native
population settlements. Partly, the colonists, like the native populations, valued the areas
of rich alluvial lands along the major river systems that served as the primary mode of
transportation and communication, or the strategic locations. Bleakley and Lin (2012)
argue that portage sites around rapids or falls gave rise to agglomerations in commerce and
manufacturing that persist today, suggesting path dependence and increasing returns to
scale. However, the native populations were critical attractions in themselves as well, again,
largely for informational, commercial, and strategic reasons. As Taylor (2001) notes, “On
their contested frontiers, each empire desperately needed Indians as trading partners, guides,
religious converts, and military allies. Indian relations were central to the development of
every colonial region�(p. 49). From Canada to Louisiana, trade and defense led the French
to establish their trading posts as nodes of trade and negotiation for securing alliances
and food.   In the North, French and Dutch Fur traders exploited existing networks of
native tribes as suppliers of pelts. Quebec, for example, was located in an area where
the local natives were skilled hunters and the nearby and numerous Huron nation served
as provisioners and trade middlemen. Similarly, on Vancouver Island and throughout the
Paci�c Northwest, the British traded extensively with natives in sea otter pelts. Linking
geographical fundamentals and pre-colonial populations, Bleakley and Lin (2012) note that
the portage site on the Savannah River (now Augusta) was an important collection point for
pelts brought by the native hunters. Pre-colonial Indian population concentrations offered

                                              19
bene�ts to colonizers along many dimensions, and those of trade in goods and information
are classic positive externalities associated with agglomerations.


   Conversely, for some indigenous agglomerations, contact with European culture and tech-
nology may have perpetuated their dominance after an initial period of trauma, particularly
given the proximity to the Industrial Revolution. Comin et al. (2010) for instance, document
an association between technology in 1500 AD and present income, roughly our period.
Ashraf and Galor (2011a) argue that at the moment of transition between technological
regimes, more cultural diffusion facilitates innovation and the adoption of new technologies.
As one suggestive example, Steckel and Prince (2001) argue that one reason that the US
plains Native Americans were the tallest people in the world in the mid-19th century was
the buffalo and game made more accessible with the introduction of horses, metal tools, and
guns by Europeans (see also Coatsworth (2008). Our documented patterns of persistence
may therefore be partly driven by the degree to which the conquest transferred the old world
technological endowment. The population centers of the earlier Maya, Anasazi, and Toltec
civilizations have vanished. Perhaps partly because of their contact with the Spanish, the
Aztec population center persists.


   Taken together, locational fundamentals, agglomeration externalities, and technological
transfer may plausibly contribute to an explanation of why pre-colonial densities mapped to
early colonial densities which, in turn, have persisted to this day.




3.4     Current Income

3.4.1   Evidence for persistence of income

The previous section con�rms for the Americas Davis and Weinstein’s (2002) �nding that
population density is persistent over very long periods of time. A large literature argues
from Malthus that high population densities in pre-industrial periods (although not now)

                                              20
signal higher productivity and prosperity (see, for example, Becker et al. (1999), Galor and
Weil (1999), Acemoglu et al. (2002), Lucas (2004)).15 Acemoglu et al. (2002) precisely
argue at the country level, that high prosperity areas in pre-colonial times, measured by
population density, became low prosperity areas today as measured by GDP per capita. We
analogously investigate this relationship at the subnational level. Because the relationship of
pre-colonial populations density is more heterogeous and complex than that with population,
we reproduce �gures for eight informative countries in Figure 4 which again, with two
exceptions, suggest persistence. Table 8 suggests that, in fact, Colombia, El Salvador,
Nicaragua and the US show positive and signi�cant coefficients in either speci�cation, while
                                                               exico is mixed depending
Argentina and Chile show a signi�cantly negative correlation. M´
                           u positive although not signi�cant.
on the speci�cation and Per´


   The pooled regressions (Tables 9 and 10) are more sensitive to speci�cation than those
for current population density but, taken in total, support persistence. Since it is more
common to include income in logs, we report a log-log and log-level speci�cation. In the
log-level estimates (Table 9), the case is particularly strong. The coefficient on pre-colonial
density shows a negative, but insigni�cant relationship in the OLS and between estimators,
but a strongly signi�cant positive relationship in all of the within estimates. In the log-log
speci�cations (Table 10), the coefficient on pre-colonial density again enters negatively and
signi�cantly for the OLS (column 1) and between regressions (column 2), but no signi�cant
relationship for either persistence or reversals in either of the within OLS FE or the MS free
standing speci�cations (columns 4, 5, and 6). For reasons to be discussed, despite leaving
unattentuated some quite extreme values, we �nd the persistence �ndings from the log-level
speci�cation with the MS corrections for outliers and bad leverage points more defend-
able. However, in neither speci�cation is there evidence for a reversal at the subnational level.



  15
     The relationship between present population and present income may be expected to be less tight than
historically was the case for at least two reasons: as Ashraf and Galor (2011b) and Galor (2011) note, the
traditional Malthusian relationship between population and wealth weakens with technological progress, and
the natural resource endowment effects discussed earlier.
                                                   21
    Including the locational fundamentals complicates the picture of the relative contribution
of geography and agglomeration externalities to these results. For the log-levels speci�cation,
the OLS FE estimates remain unchanged at .09 and become somewhat more statistically
signi�cant. The MS freestanding estimates are lower than the OLS FE case (.06), but
actually experience a statistically signi�cant rise to .1 when the fundamentals are included.
Whether due to the fact that altitude and rainfall now enter with reversed sign relative
to the pre-colonial density regressions, or other factors, locational fundamentals appear to
decrease persistence in this case. The log-level speci�cations strongly support persistence
that is not related to locational fundamentals: high density pre-colonial density areas appear
richer now because they were dense.16 In the log-log speci�cation, however, introducing
locational fundamentals (column 7) now leads to a negative and signi�cant coefficient on
pre-colonial density. In other words, in this case, locational fundamentals appear to be a
force for persistence that offsets a signi�cantly negative agglomeration effect. Although,
overall, there is no evidence for reversals at the subnational level, the contribution of
agglomeration externalities appears opposite in the two speci�cations.


    The differing results with respect to the log-level speci�cation partly arise from the
fact that the two formulations give very different weight to critical observations. In the
regressions with pre-colonial density in levels, very dense regions are heavily weighted.
However, in the log-log speci�cations, high pre-colonial density areas are pulled toward
the mean. At the other end of the spectrum, wealthy areas with very small pre-colonial
populations now take very extreme values and become more influential. For example, the
   apagos Islands in Ecuador constitute the most extreme observation with vast tourist
Gal´
rents paired with the virtual absence of pre-colonial population. The next most influential
                          nez de Campo in Chile are respectively the closest and second
points, Magallanes and Iba˜

  16
     In both log-level and log-log speci�cations, the influence of locational fundamentals appears attenuated
somewhat: agricultural suitability, river density and temperature never appear signi�cantly. In modern
economies, farming areas are no longer the richest areas of the economy; river density is not essential for
�shing or transport; air conditioning allows living in desertic areas such as the US Southwest. Landlocked
enters negatively and signi�cantly and altitude and rainfall enter positively with diminishing effect as in the
pre-colonial regressions.
                                                     22
closest regions to Antarctica and have roughly 150,000 people each today. Arguably, we are
less interested in the fact that tourism can thrive in an environment where natives did not
or that natural resources are often found in uninhabitable places, than in understanding the
impact of increasingly substantial native populations. Hence, we �nd the levels formulations
with the MS correction for outliers and bad leverage points more germane to the question
of persistence although, again, in neither case do we �nd evidence for a reversal. The
weighting effect also appears to give relatively more influence in the log-log speci�cation
to the only two countries, Argentina and Chile, that Figure 4 and Table 8 suggest had
strong negative and signi�cant correlations, yet which have very low pre-colonial densities.
Table 11 shows that, dropping these two countries, there are now no remotely signi�cant
negative coefficients on agglomeration effects in the log-log formulation (columns 5-8) while
the log-level regression estimates (columns 1-4) remain positive and signi�cant.


   Overall, the country by country regressions show several very important clear examples
of persistence, and the pooled regressions, while less robust than in population case, suggest
persistence or at least never reversals. These results raise the question of why, at the
subnational level, we �nd evidence for persistence, while at the national level Acemoglu
et al. (2002) �nd reversals. Several hypotheses come to mind. First, it may be that national
institutions intrinsically dominate those subnational, although, as noted earlier, a sizeable
literature �nds important institutional effects at the subnational level. Further, the recent
literature on the importance of de jure vs de facto power structures (see, among others,
                                  ıa-Jimeno (2005), Naritomi et al. (2007)) allows for
Acemoglu and Robinson (2006), Garc´
greater presence and heterogeneity of influential institutions across subregions than visible
legal arrangements may suggest.


   A complementary explanation is that effects relating to locational fundamentals or
agglomeration effects emerge more strongly at lower levels of aggregation as suggested by
Davis and Weinstein (2002). Indeed, studies that have tried to measure agglomeration

                                             23
economies carefully (Ciccone and Hall (1996), Duranton (2005), Ellison and Glaeser (1997)
and Greenstone et al. (2010)) have all done so using highly disaggregated subnational
data. It is possible that at this level the net effect yields persistence, while at the national
level where local agglomeration and geographical effects are diluted, national institutions
dominate. For example, for all the reasons discussed in the previous section, colonists settled
where there were previous indigenous concentrations and greater agglomerations prospered
relative to lesser. However, the institutions that emerged at the level of the nation may
have reflected the extractive dimension of the agglomerations and have led, overall, to lower
growth of the country.17


    In the next section, we �rst take a more careful historical look at individual country
cases to get a clearer view of how the different forces interacted to yield the patterns we see
at the subnational level. In particular, we examine two clear examples of persistence, the
                                                                exico and Per´
US and Colombia, and then the emblematic colonial experiences, M´            u, which
are less clear, but which we will show to offer important evidence for persistence. Finally,
we examine the roots of the negative relationship found in Argentina and Chile.



3.4.2    Clear examples of persistence: The US and Colombia

Persistence holds strongly in the US whether pre-colonial density enters in log or level form.
California, Massachusetts, and Rhode Island again, show the highest pre-colonial density
and above average incomes. Among the mid-level pre-colonial density states, New Jersey,
Connecticut, Delaware, are also among the richest, and Washington and Oregon are solidly
above average. This mass of points on the two coasts drives the upward sloping relationship
while a diffuse mass of largely southern and mountain states anchors the low pre-colonial
density-low current income nexus.


  17
     Our thanks to Noam Yuchtman for suggesting this interpretation. Relatedly Summerhill (2010) �nds
that in Brazil, a “potentially coercive� colonial institution, the aldeamento, that regulated indigenous
populations is positively correlated with income per capita at the end of the twentieth century. He argues
that there were both extractive and settler (effectively agglomeration effects), and the net effect was positive.
                                                     24
    As noted earlier, higher incomes plausibly �nd their roots both in the initial native
agglomerations and locational fundamentals that attracted both native populations and
Europeans. Both effects continued to be important across the centuries. For example, New
York, Boston and Chicago have all played to their locational particularities, especially in
transport, but they have also built on their strengths in accumulated human capital and
information (see, for example Glaeser (2005)). There is also an argument for poor institu-
tions driving the poorer regions, although perhaps more in line with Acemoglu et al. (2001).
The adverse disease environment and climate of much of the South discouraged settlement
and, in the end, colonization required the importing of African slaves.18 We explore the
influence of slavery on persistence in a more structured manner in section 4. Overall, for a
moderate range of pre-colonial densities, the US suggests the persistence of economic activity.


    Colombia is an important case for understanding the relative import of the different
forces for and against persistence and, in particular, that of extractive institutions. First,
though it is not among the countries with the highest pre-colonial density, it is a classic
example of Spanish conquest with the usual attendant institutions. Hence, while we might
argue that something about Anglo or French colonists led to different colonizer-native
dynamics, this would not be the case in Colombia. Second, it is a perhaps the case which
offers the greatest independence of subnational observations from national conditions and


   18
      The state most likely to capture the colonization-driven inversion dynamic might have been Mississippi
since it incorporated the third largest native civilization in North America, was abused by the Spaniards, and
is now the poorest state. However, the reversal of the state’s fortune from a rich cotton center in the 19th
century is likely more due to the institutional, demographic and education legacy of African slavery than
the long vanished native population. As Taylor (2001) notes, the Spanish conquistador Hernando de Soto
arriving in the fertile Mississippi river valley in 1540-1542 was impressed by size of native populations, the
expansive maize �elds, the power of their chiefs to command large numbers of well trained warriors, even the
pyramids, one of which was the third largest in North America after those of central M´      exico (The pyramid
at Cahokia was near present day St. Louis.) De Soto died on the banks of the Mississippi, frustrated at
�nding no gold, and the Spaniards withdrew to M´     exico City, but not before widespread pillaging and infection
decimated the native population. When the French returned a century later, only the Natchez people near
present day Natchez, Mississippi remained in strength and organization. French encroachments on Natchez
territories in 1729 led to massacres by the French and their Choctaw allies and dispersion and sale into slavery
in the French West Indies of the surviving population. With the passage of another century, Natchez and
Mississippi would emerge very prosperous at the height of the cotton boom.
                                                       25
institutions. The country is highly geographically fragmented and its regions have shown
a �erce autonomy, long resisting centrally imposed rule. As Safford and Palacios (1998)
note, “Provincial government remained effectively independent of the Audiencia [the local
                                               a lacked formal authority of what is now
Spanish seat of control], and Santa Fe de Bogot´
western Colombia�(p. 55). This means that national conditions and institutions were often
relatively less important than those local compared to other countries, and, as noted earlier,
                                                                     ıa-Jimeno (2005); Bonet
a variety of local institutional structures coexisted (see again Garc´
and Roca (2006)) and affected local development outcomes. Several regions employed both
native and African slaves and evolved extractive institutions to manage them, others far less.
Similarly, the Mita, Resguardo, and Encomienda are found to varying degrees in different
departments. Independence saw several (repressed) attempts at regional succession, and the
construction of a strong national state was effectively resisted.19 Yet, despite the relative
strength of local Colombia still shows one of the cleanest examples of persistence in the
sample (Figure 4). Not only the capital, but other areas of high pre-colonial density-Valle
de Cauca, Santander, and Antioquia-have among the highest present day incomes. Hence,
again, local agglomeration and locational forces appear to be dominant.


    One particular reversal within the country is illustrative of the relative strength of
locational fundamentals in particular.            Although understated in the �gures, Cauca de-
                                     an fell from one of the two most important regions
partment and its principal city Popay´
in Colombia–a major provider of early Colombian presidents and possessor of one of
the country’s two mints–to one of the poorer regions. The Spaniards favored it for the
availability of indigenous labor to extract its mineral wealth, and its subsequent use of


   19
      As an example, Bushnell (1993) recounts that the 1863 Constitution created nine United States of
Colombia, but with far more restricted central power than was the case in the North American analogue. For
instance, states issued their own stamps; the national government had responsibility only for “interoceanic�
transport routes (that is, pertaining to the Panama railroad) thereby weakening any integrative national
infrastructure project; and the upper house of the national Congress was called the Senate of Plenipotentiaries
“as if its members were the emissaries of sovereign nations� (p. 122). In fact, the titles of the two principal
English-language histories emphasize the lack of national integration: Colombia, Fragmented Country,
Divided Society by Bushnell and The Making of Modern Colombia: A Nation in Spite of Itself by Safford
and Palacios.
                                                      26
imported African slaves de�ned its culture in fundamental ways. However, the city that it
lost market share to, Cali, in Colombia’s now second richest department, Valle de Cauca,
had an indigenous population density 30% larger and only 10% fewer slaves per capita than
Cauca. In fact, it had the largest number of slaves of any department in Colombia.20 The
period critical to the reversal appears to be 1878 to 1915 with the construction of the Paci�c
Railroad connecting Cali with Buenaventura, Colombia’s largest Paci�c port, and through
                                                                        an remained
the Panama Canal (�nished in 1914) to the rest of the world, while Popay´
relatively isolated (Safford and Palacios 2002). It is likely that the location of the railroad,
while importantly dictated by Cali’s proximity to the Cauca River, is partly due to political
economy considerations. However, a story related to initial populations or slavery does not
appear clearly. It seems more likely that a permanent shock to locational fundamentals
altered the relative attractiveness of the two regions.21


    Of interest is that Cali’s new locational advantage did not also diminish Antioquia and
         a/Cundinamarca agglomeration as industrial centers. In the colonial period, both
the Bogot´
had a locational advantage in terms of climate and soil suitable for agriculture, proximity
to mineral reserves, and disease inhibiting altitude. Yet, none of these are important to
explaining the overwhelming dominance of both areas in the manufacturing and service
                                                     a’s case, several mountain ranges to
sectors, while the need to cross, especially in Bogot´
access world markets is a major drag on competitiveness. This is suggestive that, as as in
the US case, agglomeration effects, in particular, the availability of talent and knowledge-
are critical to the continued dominance of these areas.22 In sum, then, in a country where

  20
     According to the 1843 Census of Colombia, 7.1% of the population was slaves in Cauca and 6.4% in
Valle; in 1851 4.7% and 4.3% respectively. Initial indigenous density was 7.1 and 9.2 respectively.
   21
                                                 ox, Colombia. This affluent port in the Magdalena River saw
     A similar story is the rise and fall of Momp´
its demise when the river shifted course, allowing the development of Magangu´e. Since then, this UNESCO
World Heritage Site has virtually remained stuck in time.
   22
     The Bogot´  a/Cundinamarca agglomeration dominates the country in most modern services and
manufactures. The capital city, Bogot´ a, has revealed comparative advantage (participation of sector in value
added relative to country average greater than 1) in non-food manufacturing (12% of value added), commerce
(14%), �nancial services (10%), real estate services (10%), services to �rms (7%), air transport services (1%),
few of which are tied to locational fundamentals. In these areas and industry in general, it is the largest
single producer in the country. In particular, it accounts for 50% of all �nancial services. Emphatically, it has
                                                       27
local institutions were relatively important, agglomeration and locational fundamentals
appear to dominate.



3.4.3       u and M´
         Per´      exico: Evidence for persistence on balance

 exico and Per´
M´            u are the emblematic examples of the colonization of the New World. They
also show pre-colonial densities, and variances, that are among the highest in our sample.
Hence, positive agglomeration effects should be exaggerated, as, presumably, should those
arising from extractive institutions. Both El Salvador and Nicaragua have comparable den-
sities and show a signi�cant positive coefficient in Table 8 indicating persistence. However,
             exico nor Per´
for neither M´            u is this the case. Closer examination suggests, however, that
they, too, offer support for the importance of the forces of persistence, albeit contaminated
by changes in locational fundamentals.


           u, Figure 4 suggests that Lima, La Libertad, Ica and Piura all correspond to very
    For Per´
high pre-colonial density areas that remain among the better off regions today. However,
Lambayeque province undermines the statistical relationship by showing the highest density
observation but below average current income. In fact, dropping Lambayeque from the
sample causes the log-level coefficient to jump from .39 to a strongly signi�cant 1.1.
Lambayeque’s decline appears largely driven by compounding natural disasters-negative
locational fundamental shocks. In pre-colonial times, the region was a major center of
the Chimor and then Inca cultures. The Spanish colonizers subsequently built a livestock
                                                                             an, taking
industry on appropriated native land and irrigation systems, as in Tenochtitl´


neither comparative advantage or much production in the agriculture (0%) or minerals ( 0%) sectors which
�rst attracted the colonists. As capital city, it also shows a comparative advantage in public administration,
but this is not dominant or unusually large (9% of value added relative to 7% on average for the country as
a whole). The enveloping department of Cundinamarca, maintains a comparative advantage in agricultural
production, but also in both agricultural and non agricultural manufacturing and is the third and fourth
largest national producer respectively. The growth of Antioquia historically was driven by mining and then
by coffee. It maintains a comparative advantage in both, but each accounts for roughly 1% of departmental
value added. It has a comparative advantage in manufacturing (18%) and commerce (11.5%) and is the
second largest producer of manufactures, commerce, and �nancial services after the Bogot´    a/Cundinamarca
agglomeration.
                                                     28
advantage of the infrastructure and knowledge of the previous civilization. From 1650 to
1719, a dynamic sugar based hacienda economy emerged and generated numerous fortunes.
However, after 1720, the economy collapsed into a century long period of stagnation.
While this was partly due to competition from other Peruvian (including local native) and
Caribbean producers, a plague of cane-eating rats in 1701 followed by two devastating
floods in 1720 and 1728 constituted idiosyncratic but very long lived shocks which caused
widespread foreclosures and the bankruptcy of the traditional producing class. Only in
the late colonial period did the regional economy recover somewhat to a now average level
income as the new owners shifted from sugar to livestock and tobacco (Ramirez, 1986).23
Since the shocks driving Lambayeque’s fate seem idiosyncratic and dropping the region
          u to join countries showing persistence with lower mean densities, Per´
causes Per´                                                                     u, should
probably be seen as con�rming persistence across a wide range of initial densities.


     exico appears to combine two distinct sets of growth dynamics that interact to obscure
    M´
any clear relationship. The �rst is the persistence effect. The Mexican Federal District
(city) is the highest density region in our sample and it is one of the richest regions in all of
Latin America. Morelos, the second densest region in our sample, has above average income.
Both suggest persistence in the most native intensive regions of the hemisphere. Tlaxcala,
                              exico ranks among the lower levels of prosperity. However, it
the third most dense area in M´
seems unlikely that we can attribute it to especially extractive institutions since, in exchange
for being the principal allies of the Spaniards and sheltering them in a particularly dire
                                    an, the Tlaxcalans were granted “perpetual exemption
moment in the conquest of Tenochtitl´
from tribute of any sort,� a share of the spoils of conquest, and control of two bordering
provinces, an agreement that was substantially respected for the duration of Spanish rule
(see Marks, 1994, p. 188).24 Among the very highest pre-colonial densities in our sample,

  23
     Lambayeque did differ in its continued heavy reliance on Indian labor as competitor sugar growing areas
shifted more toward African slaves, although it is not clear whether this should have generated more or less
toxic extractive institutions.
  24
     In fact it may have been the opportunities for adventurism in partnership with the Spaniards in other
areas of the New World that diverted energies from the home region. Tlaxcalans aided the Spaniards in
                                                    29
agglomeration effects again appear dominant.


    However, there is a second dynamic. The present high income of the low pre-colonial
                                               on, Baja California Norte, Chihuahua,
density states of Baja California Sur, Nuevo Le´
Sonora, and Coahuila provide a strong countervailing “reversal� that offsets the persistence
effects. The proximity of these states to the increasingly dynamic US border makes it
difficult to disentangle the influence of various types from the North (proximity to markets,
knowledge spillovers), where it was in large part an appendage of the US economy. At the
time of the establishment of the border at the Rio Grande, it was linked by population
flows and contraband; during the civil war, it was a signi�cant Southern export outlet;
and by the turn of the century, it had received substantial US investments in railroads and
mining that gave the impetus to the development of capitalism in the North (Mora-Torres,
2001). For instance, US �rms operated mines in the North for export to US foundries
(e.g. Consolidated Kansas City Smelting in Chihuahua). The three large foundries that
                                                                                       exico,
formed the basis for the future dynamism of the principal industrial city in northern M´
                                                                exico) were primarily
Monterrey, Nuevo Leon (with spillovers to much of the north of M´
oriented toward the US market, and the largest was established by the Guggenheim interests
with US capital (Morado, 2003).25 As Marichal (1997) notes, the emerging industry in these
areas gave impetus to a set of de facto and eventually de jure institutions and pro-industry
regulations which may well have only been able to emerge in an environment where the
regulatory structure had not been driven by extractive considerations. That said, the fact
that a positive correlation (strongly signi�cant in the log levels speci�cation) emerges when
we abstract from the border states causes us to think that the proximity to the US was the
primary driver of the prosperity of the low density North.

                                                                                               exico
dominating conquered tribes moving North. The oldest church in the US, found in Santa Fe, New M´
was constructed by Tlaxcalan artisans.
  25
     As Mora-Torres (2001) notes, these foundries emerged largely as a result of the McKinley tariffs of 1890,
which taxed foreign imports at roughly 50 percent. This threatened both Mexican exports of ore to the US,
as well as the smelters on the US side that processed them. The response was to move the smelters over the
border to the railway center of Monterrey. The result of the accumulated US capital investment was “that
the northern economy became an extension of the U.S. economy and that the North turned into the new
center of Mexican capitalism� p. 9.
                                                     30
3.4.4    Reversals: Argentina and Chile

Above we noted that two low density countries, Argentina and Chile, provide the only
two examples of statistically signi�cant “reversals� (Figure 4) and drove a signi�cant
negative coefficient on agglomeration externalities in the pooled MS log-log speci�cations.
Hence, understanding the cause of their negative relationship is of particular interest. For
Argentina, the evidence supports an idiosyncratic geographical fundamentals story rather
than an institutional one. The richest areas in Figure 2-the Province of Buenos Aires, La
        ordoba, Santa Fe and Entre Rios surround Buenos Aires City-tend, in fact, to be
Pampa, C´
in areas of low pre-colonial population density. The other richer departments, Santa Cruz
and Chubut, are relatively undiversi�ed mineral producers in relatively unattractive climates
and hence show the “resource inversion� discussed earlier. At the other extreme, Corrientes
and Misiones are relatively underdeveloped humid semi-tropical areas that were traditionally
isolated and show the highest pre-colonial density and, hence, potentially extractive institu-
tions. But it must be kept in mind that these densities map in both absolute and relative
magnitude to those of Massachusetts and California within an overall distribution that,
again, is remarkably similar that of the US. Hence, a theory of institution-driven inversion
would need to explain why the endogenously emerging institutions would be so different
in the two countries. In addition, Buenos Aires may well not have been such a paragon of
inclusionary institutions that would account for its unusual growth.It was a major port of
slave desembarcation in the New World and, in the last years of Spanish domination, it was
30% Black (Andrews, 1980).26


    It seems more likely that the present distribution of income arises largely from Buenos
Aires’ status as the principal Atlantic port of the Spanish empire. This was not always


  26
      As a �nal point, Ades and Glaeser (1995) note argue that industry did not play a prominent role in the
rise of Buenos Aires so that a case for it being more suited to the second wave of the Industrial Revolution
seems unlikely. Even by 1914, only 15 percent of the labor force was in manufacturing and the government
displayed “hostility toward manufacturing and innovation� p 221.
                                                    31
the case. Despite the evolution of the surrounding pampas economy, prior to the mid 18th
century Buenos Aires was a backwater, surviving on smuggling contraband silver and slaves.
This was largely due to the repression of natural locational advantage. By Spanish law, the
production of silver and other products of the interior towns were directed over the Andes
to Lima on the Paci�c, where they were loaded on convoys passing through the Isthmus of
Panama and then to Spain. The more logical route-through the Atlantic port of Buenos
Aires, and then directly to Spain-was forbidden. However, largely for geostrategic reasons
arising from the emergence of the North American colonies as a potential Atlantic power,
the policy was reversed in 1776 when Spain established Buenos Aires as the capital of the
new Viceroyalty of Rio de la Plata. Trade was now mandated through Buenos Aires and
forbidden through Lima, leading to an abrupt reorientation of the country’s economy away
from the traditional interior towns, and towards the emerging coastal economy (Scobie,
1964). Hence, by royal �at, locational fundamentals went from being repressed to dominant.


   Chile also shows a signi�cant negative relationship between pre-colonial densities and
present income but one which, again, does not appear driven by the institutional story for
two reasons. First, several observations at the highest end of the country’s relatively low
density (4.7 per square kilometer)-Bio Bio, Maule, O’Higgins, Los Lagos, and Araucania-are
among the poorest. However, these form a contiguous region, with the area below the
Bio Bio River that includes them dominated by the Mapuche Indians and conquered only
very late in the 19th century. That is, extractive institutions would have been set up after
the advent of the second Industrial Revolution. Hence, the institutional case is not as
compelling, perhaps, as one stressing the costs of being out of the global technological loop.
In fact, the eventual conquest had to wait for the Chileans to import recent advances in
weaponry to which the Mapuches did not have access. The capital, Santiago, offers a counter
example: it has the same density and is contiguous to this region, but it was conquered much
earlier and is much more prosperous. Second, the country is one of extremes with extractive
industries in some of the driest and coldest areas of the planet which were not attractive

                                             32
to native populations. This implies a relatively uninteresting correlation of relatively low
pre-colonial densities, and moderately high incomes (for a very few people) today. Excluding
these areas leaves no correlation whatsoever.


    In sum, in both Argentina and Chile the negative correlation of present income with pre-
colonial population density seems largely driven by idiosyncratic geographical and historical
factors. Hence, for both this reason, and because we �nd stronger justi�cation for the log-
level speci�cation, we �nd the evidence overall for persistence more compelling. Further,
while the US and Colombia provide very robust evidence that pre-colonial densities are
                                         exico and Per´
correlated with present income, even in M´            u where the pattern is muddier, we
�nd strong evidence persistence. Part of this may be due to the demonstrated power of
geographic factors. Across all our case studies, the large changes in relative positions, such
           an (Colombia), Lambayeque (Per´
as in Popay´                             u), Buenos Aires, (Argentina), or the North of
 exico appear largely driven by shifts in locational fundamentals. However, the persistent
M´
                                                                           a in Colombia,
prosperity of California and New England, in the US, or Antioquia, or Bogot´
despite massive structural transformations away from natural resource based production
toward more sophisticated manufacturing and services does suggest that the forces arising
from concentrations of knowledge, trade or labor are also important.


3.5     The Institutional Channel: Slavery in Brazil, Colombia, and
        the US

    Though the previous section has documented the relative importance of locational
fundamentals, agglomeration externalities and perhaps technological transfer in determining
present income, we also �nd evidence for the negative impact of extractive institutions, even
if they did not dominate the others.27 As a proxy for extractive institutions we are able to

  27
     The negative impact of slavery cannot be taken as a foregone conclusion since disentangling the
endowment and institutional effect is difficult. Acemoglu et al. (2012) �nd that in Colombian municipalities
where slave labor was demanded poverty is higher and school enrollment, vaccination coverage and public
                                                                       ao Paulo, Brazil Summerhill (2010)
good provision is lower, than where it was not. On the other hand, in S˜
�nds no relationship between slavery and present incomes while Rocha et al. (2012) �nd slavery is positively
correlated with human capital.
                                                    33
collect data on the incidence of slavery at the subnational level for Brazil and Colombia and
the US. While data comparability and classi�cation issues are non-trivial, the average share
of the population enslaved in the mid 19th century was 28% in the American South, 13% in
Brazil and 2.9% in Colombia. We use the more expansive measure that includes free Blacks
which raises Brazil to �rst position, although the results do not change qualitatively when
we use the more narrow measure.




  Log (Y2005,ij ) = α + βg (Dprecol,ij ) + δSLAV ERYij + δint SLAV ERYij ∗ g (Dprecoli j ) +

                                                               γLOCAT ION ALij + µi +         ij   (4)


where δint captures the interaction of pre-colonial density and slavery and µi are now three
�xed effects for Brazil, Colombia and the US South with the US North as the omitted
category. Columns 1-5 in Tables 12 and 13 progressively introduce the elements of equation
3. Column 1 includes pre-colonial density along with dummies for Brazil, Colombia and the
American South.28 In the full sample in both log-level and log-log speci�cations (column 1),
pre-colonial density is signi�cant and positive, lending support from a smaller sample to the
case for persistence. Column 2 repeats the same regression with the smaller sample dictated
by the more restrictive slavery variable with a loss in signi�cance of the persistence term.
Columns 3 add the slavery term and, for both log and level speci�cations, it enters negatively
and signi�cantly. Column 4 adds slavery interacted with initial population density. It
enters negatively in both speci�cations and of similar sign although only signi�cantly in the
levels speci�cation. Further, the coefficient on pre-colonial density roughly doubles with the
inclusion of the interaction of slavery and density in the levels speci�cation and increases by
30 percent in the log speci�cation suggesting that extractive institutions did have a negative
agglomeration effect as postulated by Acemoglu et al. (2002). The same results are found in


  28
     The South is comprised of Alabama, Arkansas, Delaware, District of Columbia, Florida, Georgia,
Kentucky, Louisiana, Maryland, Mississippi, Missouri, North Carolina, South Carolina, Tennessee, Texas,
Virginia, West Virginia.
                                                  34
the MS speci�cations (column 6).


    Adding locational fundamentals (column 5) changes the coefficient little but renders
it insigni�cant in the OLS FE estimates. However, the MS estimator �nds both the free
standing and interactive terms signi�cant in both level and log speci�cations. Though the
sample is small, nonetheless, the results offer support for extractive institutions at least
reducing, if not overturning the persistence induced by agglomeration externalities and
fundamentals.


    In sum, despite legitimate concerns about the exogeneity of slavery, using a direct proxy
for exclusionary institutions is suggestive of the existence of an Acemoglu et al. (2002) effect:
slavery works against persistence. However, at the subnational level, the net effect of factors
associated with of indigenous population densities– extractive institutions, agglomeration
externalities, or locational fundamentals–tends to leave a positive correlation with prosperity
today.



4     Conclusion
    This paper documents that, within countries, economic activity in the Western Hemi-
sphere has tended to persist over the last half millennium. We construct a data set on
subnational population densities and incomes derived from poverty maps, and show that
pre-colonial population densities are strongly positively correlated with present day popula-
tion and somewhat less robustly (although never negatively), with income per capita. This
is clearly the case for low pre-colonial density countries like the US, but also for classic Latin
                                                                                       exico
conquest cases like Colombia, and, on balance for the extreme high density cases like M´
       u. We also generate proxies for suitability for agriculture and river density that
and Per´
contribute to a comprehensive set of locational fundamentals. These appear as signi�cant de-
terminants of the location of pre-colonial densities and of our present day measures, but they
do not appear to drive the observed persistence. Including locational fundamentals lowers
                                               35
the contribution of initial density only slightly in some population speci�cations, and actually
increases it in the preferred income speci�cation suggesting that agglomeration effects are
dominant on average. That said, the historical case studies clearly show that shocks to
locational fundamentals can importantly impact the spatial distribution of economic activity.


   Further, the case studies suggest reasons for both fundamentals and pre-colonial densities
to be important.    Not only would colonizers also value the rivers, coasts, fertile land,
natural resources, and climate that attracted the native populations, but they would need
the native populations themselves as sources of human capital (architects, agronomists,
and craftsmen), trading partners, sources of information, strategic bulwarks against enemy
encroachment, and souls to save. Hence, scale economies and Marshallian externalities
related to population were probably as relevant to determining where colonists located their
settlements as locational fundamentals. In turn, the contact with new technologies may
have, after the initial trauma, strengthened these agglomerations.


   That said, we also �nd evidence, at the subnational level, for the negative institutional
effects associated with pre-colonial agglomerations postulated by Acemoglu et al. (2002) at
the country level. Using the share of slaves in the population as a proxy for extractive
institutions, we �nd that regions with a higher incidence of slavery show both lower incomes
and less persistence. Hence, persistence is likely to have been stronger were such institutions
not a feature of the colonization. However, these effects do not appear strong enough to cause
reversals at the subnational level. It may be that the greater influence of agglomeration and
locational effects at lower levels of aggregation causes the net effect here to be positive
whereas at the national level this is not the case. In sum, however, many of the regions
of the very highest pre-colonial density remain among the most prosperous regions today,
and the few countries and regions exhibiting reversals seem less driven by institutional
than idiosyncratic fundamentals-related stories. At the subnational level, geographical and
especially agglomeration factors appear to dominate and to cause fortune to persist.

                                              36
References
Acemoglu, D. and Dell, M. (2010). Productivity Differences Between and Within Countries.
  American Economic Journal: Macroeconomics, 2(1):169–88.

                  ıa-Jimeno, C., and Robinson, J. A. (2012). Finding Eldorado: Slavery and
Acemoglu, D., Garc´
  Long-run Development in Colombia. Working Paper 18177, National Bureau of Economic
  Research.

Acemoglu, D., Johnson, S., and Robinson, J. A. (2001). The Colonial Origins of Comparative
  Development: An Empirical Investigation. American Economic Review, 91(5):1369–1401.

Acemoglu, D., Johnson, S., and Robinson, J. A. (2002). Reversal of Fortune: Geography
  and Institutions in The Making of The Modern World Income Distribution. The Quarterly
  Journal of Economics, 117(4):1231–1294.

Acemoglu, D. and Robinson, J. A. (2006). De Facto Political Power and Institutional
  Persistence. The American Economic Review, 96(2):pp. 325–330.

Ades, A. F. and Glaeser, E. L. (1995). Trade and Circuses: Explaining Urban Giants. The
  Quarterly Journal of Economics, 110(1):195–227.

Andrews, G. R. (1980). The Afro-Argentines of Buenos Aires 1800-1900. University of
  Winsconsin Press. Madison.

Aroca, P., Bosch, M., and Maloney, W. F. (2005). Spatial Dimensions of Trade Liberalization
  and Economic Convergence: Mexico 1985–2002. World Bank Economic Review, 19(3):345–
  378.

Ashraf, Q. and Galor, O. (2011a). Cultural Assimilation, Cultural Diffusion and the Origin
  of the Wealth of Nations. mimeo.

Ashraf, Q. and Galor, O. (2011b). Dynamics and Stagnation in the Malthusian Epoch:
  Theory and Evidence. American Economic Review.
                                            37
Ashraf, Q. and Galor, O. (2012). The� Out of Africa� Hypothesis, Human Genetic Diversity,
  and Comparative Economic Development. American Economic Review.

Banerjee, A. and Iyer, L. (2005). History, Institutions, and Economic Performance: The
  Legacy of Colonial Land Tenure Systems in India. American Economic Review, 95(4):1190–
  1213.

Becker, G. S., Glaeser, E. L., and Murphy, K. M. (1999). Population and Economic Growth.
  American Economic Review, 89(2):145–149.

Bleakley, H. and Lin, J. (2012). Portage and Path Dependence. Quarterly Journal of
  Economics.

Bonet, J. and Roca, A. M. (2006). El Legado Colonial como Determinante del Ingreso per
   apita Departamental en Colombia. Documentos de Trabajo sobre Econom´
  c´                                                                  ıa Regional
                         ublica - Econom´
  002520, Banco de la Rep´              ıa Regional.

Breen, T. H. (1984). Creative Adaptations: Peoples and Cultures. Colonial British America.

Bruhn, M. and Gallego, F. A. (2011). Good, Bad, and Ugly Colonial Activities : Studying
  Development Across the Americas. Review of Economics and Statistics.

Bushnell, D. (1993). The Making of Modern Colombia: A Nation in Spite of Itself. University
  of California Press.

Ciccone, A. and Hall, R. E. (1996). Productivity and the Density of Economic Activity.
  American Economic Review, 86(1):54–70.

Coatsworth, J. H. (2008). Inequality, Institutions and Economic Growth in Latin America.
  Journal of Latin American Studies, 40:545–569.

Comin, D., Easterly, W., and Gong, E. (2010). Was the Wealth of Nations Determined in
  1000 BC? American Economic Journal: Macroeconomics, 2(3):65–97.

Davis, D. R. and Weinstein, D. (2002). Bones, Bombs, and Break Points: The Geography of
  Economic Activity. American Economic Review, 92(5):1269–1289.
                                        38
De Vorsey, L. (1978). Amerindian Contributions to the Mapping of North America: A
  Preliminary View. Imago Mundi, 30:71–78.

De Vorsey Jr, L. (1986). The Colonial Georgia Backcountry. Colonial Augusta:“Key of the
  Indian Countrey�(Macon, Ga., 1986), pages 3–26.

Degler, C. N. (1970). Out of Our Past : The Forces that Shaped Modern America. Harper
  & Row, New York.

Dell, M. (2011). The Persistent Effects of Peru’s Mining Mita. Econometrica. Forthcoming.

Denevan, W. (1992a).    The Native Population of the Americas in 1492.     University of
  Wisconsin Press.

Denevan, W. (1992b). The Pristine Myth: the Landscape of the Americas in 1492. Annals
  of the Association of American Geographers, 82(3):369–385.

Duranton, G. (2005). Testing for Localization Using Micro-Geographic Data. Review of
  Economic Studies, 72(4):1077–1106.

Easterly, W. and Levine, R. (2003). Tropics, Germs, and Crops: How Endowments Influence
  Economic Development. Journal of Monetary Economics, 50(1):3–39.

Elbers, C., Lanjouw, J. O., and Lanjouw, P. F. (2003). Micro–Level Estimation of Poverty
  and Inequality. Econometrica, 71(1):355–364.

Ellison, G. and Glaeser, E. L. (1997). Geographic Concentration in U.S. Manufacturing
  Industries: A Dartboard Approach. Journal of Political Economy, 105(5):889–927.

Ellison, G. and Glaeser, E. L. (1999). The Geographic Concentration of Industry: Does
  Natural Advantage Explain Agglomeration? American Economic Review, 89(2):311–316.

Fujita, M. and Mori, T. (1996). The Role of Ports in the Making of Major Cities: Self-
  Agglomeration and Hub-Effect. Journal of Development Economics, 49(1):93–120.


                                           39
Gallup, J. L., Sachs, J. D., and Mellinger, A. D. (1998).        Geography and Economic
  Development. Technical report.

Galor, O. (2011). Uni�ed Growth Theory. Princeton Univ Press.

Galor, O. and Weil, D. N. (1999). From malthusian stagnation to modern growth. American
  Economic Review, 89(2):150–154.

    ıa-Jimeno, C. (2005). Colonial Institutions and Long-Run Economic Performance in
Garc´
  Colombia: Is There Evidence of Persistence? Documento CEDE.

Gennaioli, N, L. P., Lopez-de Silanes, F., and Shleifer, A. (2011). Human Capital and
  Regional Development. mimeo.

Glaeser, E. L. (2005). Urban Colossus: Why is New York America’s Largest City? NBER
  Working Papers 11398, National Bureau of Economic Research, Inc.

Glaeser, E. L., Kallal, H. D., Scheinkman, J. A., and Shleifer, A. (1992). Growth in Cities.
  Journal of Political Economy, 100(6):1126–52.

Greenstone, M., Hornbeck, R., and Moretti, E. (2010). Identifying Agglomeration Spillovers:
  Evidence from Winners and Losers of Large Plant Openings. Journal of Political Economy,
  118(3):536–598.

Krugman, P. (1991). Increasing Returns and Economic Geography. Journal of Political
  Economy, 99(3):483–99.

Krugman, P. (1992). Geography and Trade, volume 1 of MIT Press Books. The MIT Press.

Krugman, P. and Elizondo, R. L. (1996). Trade Policy and the Third World Metropolis.
  Journal of Development Economics, 49(1):137–150.

Krugman, P. R. (1993). On the Relationship between Trade Theory and Location Theory.
  Review of International Economics, 1(2):110–22.


                                            40
Lucas, R. E. (2004). The Industrial Revolution: Past and Future. Annual Report, (May):5–
  20.

Marichal, C. (1997).   Avances Recientes en la Historia de las Grandes Empresas y su
                                   omica de M´
  Importancia para la Historia Econ´         exico. In Marichal, C. and Cerutti, M.,
                                                exico, 1850-1930. Universidad Aut´
  editors, Historia de las grandes empresas en M´                                onoma
             on - Fondo de Cultura Econ´
  de Nuevo Le´                         omica.

                         es: The Great Adventurer and the Fate of Aztec Mexico. Alfred
Marks, R. L. (1994). Cort´
  A. Knopf.

Maronna, R. and Yohai, V. (2000). Robust Regression with Both Continuous and Categorical
  Predictors. Journal of Statistical Planning and Inference, 89(1-2):197–214.

Miguel, E. and Roland, G. (2011). The Long Run Impact of Bombing Vietnam. Journal of
  Development Economics, 96(1):1–15.

Mitchener, K. J. and McLean, I. W. (2003). The Productivity of US States since 1880.
  Journal of Economic Growth, 8(1):73–114.

Mora-Torres, J. (2001). The Making of the Mexican Border. University of Texas Press.

                                                                  exico. 1890-1908.
Morado, C. (2003). Empresas Mineras y Metalurgicas en Monterrey, M´
                              urgicas. Technical report.
  Parte ii. Tres Plantas Metal´

                                      c˜
Naritomi, J., Soares, R. R., and Assun¸ ao, J. J. (2007). Rent Seeking and the Unveiling of
  ‘De Facto’ Institutions: Development and Colonial Heritage within Brazil. NBER Working
  Papers 13545, National Bureau of Economic Research, Inc.

Newson, L. (1982). The Depopulation of Nicaragua in the Sixteenth Century. Journal of
  Latin American Studies, 14(2):235–286.

Nunn, N. (2008). Slavery, Inequality, and Economic Development in the Americas: An
  Examination of the Engerman-Sokoloff Hypothesis. In Helpman, E., editor, Institutions
  and Economic Performance, pages 148–180. Harvard University Press.
                                        41
Parkes, H. B. (1969). A History of Mexico. Houghton Mifflin.

Ramankutty, N., Foley, J., Norman, J., and McSweeney, K. (2002). The Global Distribution
  of Cultivable Lands: Current Patterns and Sensitivity to Possible Climate Change. Global
  Ecology and Biogeography, 11(5):377–392.

Ramirez, S. E. (1986). Patriarchs: Land Tenure and the Economics of Power in Colonial
     u. University of New Mexico Press, Albuquerque, 1st edition.
  Per´

Rappaport, J. and Sachs, J. D. (2003). The United States as a Coastal Nation. Journal of
  Economic Growth, 8(1):5–46.

Rocha, R., Ferraz, C., and Soares, R. (2012). Settlement Colonies Across Plantation Fields:
  Evidence on the Relationship Between Human Capital and Long Term Development.
  mimeo, PUC-Rio.

Safford, F. and Palacios, M. (1998). Colombia: Fragmented Land, Divided Society. 1st ed.
  edition.

Scobie, J. R. (1964). Argentina : A City and a Nation. Oxford University Press, New York.

Simpson, L. (1962). Many Mexicos. Univ of California Press.

Steckel, R. H. and Prince, J. M. (2001). Tallest in the World: Native Americans of the Great
  Plains in the Nineteenth Century. The American Economic Review, 91(1):pp. 287–294.

Summerhill, W. (2010).      Colonial Institutions, Slavery, Inequality, and Development:
                 ao Paulo, Brazil. MPRA Paper 22162, University Library of Munich,
  Evidence from S˜
  Germany.

Taylor, A. (2001). American Colonies: The Settling of North America the Penguin History
  of the United States, Volume 1. Penguin History of the United States. Penguin Books,
  New York, N.Y.

                                            es, and the Fall of Old M´
Thomas, H. (1993). Conquest: Montezuma, Cort´                        exico. Touchstone.
                                             42
          on, H., Tovar-Mora, J. A., and Tovar-Mora, C. E. (1994). Convocatoria al Poder
Tovar-Pinz´
       umero: Censos y Estad´
  del N´                    ısticas de la Nueva Granada, 1750-1830. Archivo General de
         on (Colombia).
  la Naci´

Ubelaker, D. (1988). North American Indian Population Size, AD 1500 to 1985. American
  Journal of Physical Anthropology, 77(3):289–294.

World Bank (2008). World Development Report 2009: Reshaping Economic Geography.




                                           43
                          Figure 1: Pre-colonial Population Density




Note: Pre-colonial Population Density is the number of indigenous people per square kilometer before the arrival
of Columbus, Income is per capita (PPP 2005 US dollars) in 2000. Data from national censuses, Denevan (1992),
and Bruhn and Gallego (2010). More detailed data sources and descriptions in the text.




                                                      44
Figure 2: Population Density in 2000 against Pre-colonial Population Density (United States)




     Note: Pre-colonial Population Density is the number of indigenous people per square kilometer before the arrival
     of Columbus, Current Population Density is the total population in 2000 divided by the area of the state or
     province in square kilometers. Data from national censuses, Denevan (1992), and Bruhn and Gallego (2010).
     More detailed data sources and descriptions in the text.


Figure 3: Log Income per Capita in 2005 against Pre-colonial Population Density (United
States)




     Note: Pre-colonial Population Density is the number of indigenous people per square kilometer before the arrival
     of Columbus, Income is per capita (PPP 2005 US dollars) in 2000. Data from national censuses, Denevan (1992),
     and Bruhn and Gallego (2010). More detailed data sources and descriptions in the text.

                                                           45
     Figure 4: Log Income per Capita in 2005 against Pre-colonial Population Density (Latin America)
46
47




     Note: Pre-colonial Population Density is the number of indigenous people per square kilometer before the arrival of Columbus, Income is per capita (PPP 2005 US
     dollars) in 2000. Data from national censuses, Denevan (1992), and Bruhn and Gallego (2010). More detailed data sources and descriptions in the text.
                                                    Table 1: Summary Statistics- Population Density and Income

                          Obs      Pre-colonial Population Density                 Current Population Density                                      Income
                                   Mean Coef. Var Min        Max                Mean Coef. Var Min         Max                   Mean        Coef. Var     Min            Max
          Argentina         24      0.44    1.45     0.01     2.55              626.06    4.80     1.20 14727.03               10576.16        0.46      5834.35        24328.34
          Bolivia            9      1.18    0.96     0.20     3.74               9.53     0.84     0.82    26.17                3494.36        0.25      2239.15         5219.44
          Brazil            27      2.55    0.97     0.20     8.58              53.39     1.40     1.41   346.75                7590.93        0.46      3343.24        18287.33
          Canada            13      1.22    1.06     0.02     3.00               6.34     1.19     0.01    24.40               34540.71        0.17     27479.80        48436.04
          Chile             13      2.65    0.87     0.01     4.66              53.05     1.99     1.05   393.50               12852.48        0.24      9545.53        19533.39
          Colombia          30      4.96    0.82     0.49    13.04              424.40    2.36     0.48  4310.09                4554.56        0.27      2546.91         6917.57
          Ecuador           22      5.76    0.78     0.01    12.06              56.10     0.92     2.01   182.80                5764.57        0.30      3738.26        10463.96
          El Salvador       14     24.19    0.24    15.80    39.25              326.73    1.30    95.58 1768.80                 4669.67        0.29      3378.47         8094.27
          Guatemala          8     22.95    0.35     5.64    29.08              248.97    1.57    10.23 1195.48                 3699.73        0.56      2132.71         8526.96
          Honduras          18      8.09    0.55     1.00    17.64              134.67    1.22    15.81   614.83                3171.35        0.30      1512.21         5170.91
          Mexico            32     31.90    2.38     0.40   392.34              227.55    3.36     5.61  4352.62               12119.95        0.29      6780.40        20709.32
          Nicaragua         17     29.82    0.89     1.00    60.00              103.28    1.20     8.58   473.80                1896.24        0.22      1250.37         2658.39
          Panama             9     13.40    0.67     0.06    24.78              38.66     0.88     2.42   116.80                9046.41        0.31      4880.31        13950.97
          Paraguay          18      1.27    0.56     0.20     3.29              58.62     2.28     0.10   579.36                4162.39        0.18      2923.94         5516.21
          Peru              24     17.36    1.30     0.78   100.15              31.80     0.18     1.08   222.23                5623.75        0.35      2846.11        10980.10
          US                48      0.39    1.34     0.02     2.17              169.50    0.99     5.16  1041.54               44193.13        0.14     34533.35        62765.91
          Uruguay           19      0.11    2.05     0.00     0.85              33.44     1.80     2.25   263.51                8195.26        0.21      6024.20        13965.81
48




          Venezuela         19      1.78    0.42     0.35     2.78              96.70     0.48     0.40   415.52                9788.84        0.13      7843.90        13191.90

     Note: Pre-colonial Population Density is the number of indigenous people per square kilometer before the arrival of Columbus, Current Population Density is the total population in
     2000 divided by the area of the state or province in square kilometers and Income is in per capita (PPP 2005 US dollars) in 2000. Data from national censuses, Denevan (1992), and
     Bruhn and Gallego (2010). More detailed data sources and descriptions in the text.
                     Table 2: Summary Statistics- Population Density and Income

                                                  Mean       Median        Sd      Min      Max
                               Agriculture        0.56        0.58         0.28    0.00     1.00
                               Rivers              3.28        3.29        1.23    0.00      6.92
                               Landlocked          0.57        1.00        0.50    0.00      1.00
                               Temperature        19.97       20.40        5.83    2.38     29.00
                               Altitude           0.66        0.19         0.92    0.00     4.33
                               Rainfall           1.28        1.10         0.95    0.00     8.13

Note: Agriculture is an index of probability of cultivation given cultivable land, climate and soil composition, from Ramankutty,
Foley and McSweeney (2002). Rivers captures the density of rivers as a share of land area derived from HydroSHEDS (USGS
2011). Landlocked is a dummy variable for whether the state has access to a coast or not; temperature is a yearly average in
â—¦ C; altitude measures the elevation of the capital city of the state in kilometers; and Rainfall captures total yearly rainfall in

meters, all are from Bruhn and Gallego (2010). More detailed data sources and descriptions in the text.




                                                               49
       Table 3: Pre-colonial Population Density and Locational Fundamentals (pooled)
                           (1)          (2)      (3)          (4)        (5)            (6)        (7)           (8)
                           OLS        OLS     OLS FE      OLS FE        MS              MS       MS FE        MS FE
        Agriculture        0.1        -0.04     0.06        -0.02     0.005∗∗         0.03∗∗    0.006∗∗∗       0.02∗∗
                          (0.08)     (0.11)    (0.06)      (0.11)      (0.00)         (0.01)     (0.00)        (0.01)
        Agriculture2                    0.1                  0.08                     -0.02∗                   -0.01∗
                                     (0.12)                (0.15)                     (0.01)                   (0.01)
        Rivers           -0.03∗∗∗     -0.03    -0.002        0.02      0.0004        -0.0005    -0.002∗∗∗      -0.002
                          (0.01)     (0.03)    (0.01)      (0.03)      (0.00)         (0.00)      (0.00)       (0.00)
        Rivers2                      0.0008                -0.002                   -0.00005                 -0.00009
                                     (0.00)                (0.00)                     (0.00)                   (0.00)
        Landlocked         0.01      -0.005     0.02         0.02    -0.007∗∗∗     -0.005∗∗∗      -0.001      -0.0007
                          (0.03)     (0.02)    (0.03)      (0.03)      (0.00)         (0.00)      (0.00)       (0.00)
        Temperature       0.004∗       0.01    -0.003        0.01    0.0008∗∗∗      -0.0009∗     0.0002∗     0.00009
                          (0.00)     (0.01)    (0.00)      (0.01)      (0.00)         (0.00)      (0.00)       (0.00)
        Temperature2                -0.0002               -0.0004                 0.00005∗∗∗                0.000002
                                     (0.00)                (0.00)                     (0.00)                   (0.00)
        Altitude           0.06        0.1      0.02         0.02     0.004∗∗∗       0.005∗∗     0.0006        0.0005
                          (0.04)     (0.08)    (0.04)      (0.05)      (0.00)         (0.00)     (0.00)        (0.00)
        Altitude2                     -0.02                -0.003                    -0.0006                 0.00003
                                     (0.02)                (0.00)                     (0.00)                   (0.00)
        Rainfall           -0.01      -0.02     -0.01       -0.03     -0.0005        -0.0002     0.00003        0.001
                          (0.01)     (0.03)    (0.01)      (0.03)      (0.00)         (0.00)      (0.00)       (0.00)
        Rainfall2                    0.002                  0.003                    -0.0002                -0.0004∗∗
                                     (0.00)                (0.00)                     (0.00)                   (0.00)
        Constant           0.02       -0.02     0.02        -0.10    -0.009∗∗∗         0.002     0.006∗∗        0.003
                          (0.06)     (0.06)    (0.09)      (0.09)      (0.00)         (0.01)      (0.00)       (0.00)
        N                  330         330      330          330        330             330        330           330
        R2                0.061      0.058     0.109        0.099

Note: Regression of sub national Pre-colonial Population Density on locational fundamentals. Estimation by OLS and robust MS
regression with country �xed effects. Pre-colonial Population Density is the number of indigenous people per square kilometer
before the arrival of Columbus, from Denevan (1992), and Bruhn and Gallego (2010). Agriculture is an index of probability of
cultivation given cultivable land, climate and soil composition, from Ramankutty, Foley and McSweeney (2002). Rivers captures
the density of rivers as a share of land area derived from HydroSHEDS (USGS 2011). Landlocked is a dummy variable for
whether the state has access to a coast or not; temperature is a yearly average in â—¦ C; altitude measures the elevation of the
capital city of the state in kilometers; and Rainfall captures total yearly rainfall in meters, all are from Bruhn and Gallego
(2010). More detailed data sources and descriptions in the text. Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1,
∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                             50
    Table 4: Log Pre-colonial Population Density and Locational Fundamentals (pooled)
                                 (1)         (2)       (3)          (4)        (5)         (6)        (7)           (8)
                                OLS        OLS      OLS FE      OLS FE        MS          MS        MS FE       MS FE
             Agriculture         1.2      6.9∗∗      2.2∗∗∗       5.0∗∗      2.0∗∗∗      5.4∗∗∗      0.8∗       4.6∗∗∗
                               (1.40)     (2.79)     (0.52)      (2.18)      (0.48)      (1.40)     (0.46)       (0.66)
             Agriculture2                  -5.6∗                   -2.9                  -2.9∗∗                 -3.3∗∗∗
                                          (3.15)                 (1.88)                  (1.35)                  (0.57)
             Rivers             -0.3        -0.5      0.04         0.4      -0.8∗∗∗     -2.0∗∗∗      -0.1        0.5∗∗
                               (0.24)     (0.50)     (0.21)      (0.59)      (0.09)      (0.31)     (0.07)       (0.20)
             Rivers2                        0.02                  -0.07                  0.2∗∗∗                -0.08∗∗∗
                                          (0.07)                 (0.07)                  (0.04)                  (0.02)
             Landlocked         -0.7        -0.7      -0.7∗        -0.6      0.05          0.2       -0.2∗         -0.2
                               (0.45)     (0.48)     (0.41)      (0.37)     (0.19)       (0.18)     (0.13)       (0.11)
             Temperature       0.2∗∗∗       0.10       0.02         0.2     0.06∗∗       0.3∗∗∗      -0.04         0.2∗
                               (0.05)     (0.24)     (0.05)      (0.16)     (0.03)       (0.08)     (0.06)       (0.11)
             Temperature2                 0.002                  -0.006                -0.006∗∗∗               -0.006∗∗
                                          (0.01)                 (0.00)                  (0.00)                  (0.00)
             Altitude          1.4∗∗∗       2.1∗      0.5∗         0.2       0.8∗∗∗        0.2      0.5∗∗         -0.4∗
                               (0.36)     (1.00)     (0.26)      (0.30)      (0.16)      (0.33)     (0.23)       (0.21)
             Altitude2                      -0.2                   0.06                   0.2∗                   0.1∗∗
                                          (0.25)                 (0.06)                  (0.09)                  (0.06)
             Rainfall           0.1          0.5      -0.10        0.1        0.3        1.8∗∗∗      -0.2∗          0.2
                               (0.17)     (0.41)     (0.13)      (0.27)      (0.19)      (0.33)     (0.09)       (0.13)
             Rainfall2                     -0.09                  -0.05                 -0.4∗∗∗                -0.05∗∗∗
                                          (0.05)                 (0.03)                  (0.08)                  (0.02)
             Constant          -8.3∗∗∗   -8.3∗∗∗     -7.7∗∗∗    -10.1∗∗∗    -4.2∗∗∗     -6.4∗∗∗     -5.0∗∗∗     -8.8∗∗∗
                                (1.37)    (1.48)      (1.48)     (1.96)      (0.98)      (1.00)      (1.33)      (1.26)
             N                   330        330        330         330        330         330         330          330
             R2                 0.376     0.413       0.688       0.703

Note: Regression of sub national Pre-colonial Population Density on locational fundamentals. Estimation by OLS and robust
MS regression with country �xed effects. Pre-colonial Population Density is the log of the number of indigenous people per
square kilometer before the arrival of Columbus, from Denevan (1992), and Bruhn and Gallego (2010). Agriculture is an index
of probability of cultivation given cultivable land, climate and soil composition, from Ramankutty, Foley and McSweeney (2002).
Rivers captures the density of rivers as a share of land area derived from HydroSHEDS (USGS 2011). Landlocked is a dummy
variable for whether the state has access to a coast or not; temperature is a yearly average in â—¦ C; altitude measures the elevation
of the capital city of the state in kilometers; and Rainfall captures total yearly rainfall in meters, all are from Bruhn and Gallego
(2010). More detailed data sources and descriptions in the text. Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1,
∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                                51
Table 5: Population Density in 2000 and Pre-colonial Population Density (country by
country)
                                               N         β Pop. Density           Rank
                                                     Log-Log     Level-Level   Correlation
                                Argentina      24     0.29∗∗        -602.5      0.61∗∗∗
                                                       (0.12)     (637.67)
                                Brazil         27    0.81∗∗∗          7.34       0.63∗∗∗
                                                       (0.20)       (5.29)
                                Bolivia         9    0.85∗∗∗       5.16∗∗         0.68∗∗
                                                       (0.28)       (2.02)
                                Chile          13     0.61∗∗∗       19.9∗        0.84∗∗∗
                                                       (0.08)      (11.47)
                                Canada         13    -0.86∗∗∗     -3.70∗∗∗        -0.69∗∗
                                                       (0.28)       (1.17)
                                Colombia       30     1.64∗∗∗     108.1∗∗        0.70∗∗∗
                                                       (0.32)      (51.20)
                                Ecuador        22     0.50∗∗∗       3.84∗         0.49∗∗
                                                       (0.10)       (2.24)
                                El Salvador    14     2.79∗∗       61.4∗∗∗        0.79∗∗
                                                       (0.67)      (21.40)
                                Guatemala       8     1.98∗∗∗         20.9        0.83∗∗
                                                       (0.38)      (14.85)
                                Honduras       18    0.79∗∗∗       21.2∗∗         0.47∗∗
                                                       (0.13)      (10.11)
                                Mexico         32    0.65∗∗∗       9.09∗∗∗       0.68∗∗∗
                                                       (0.12)       (2.32)
                                Mexico1        25     0.80∗∗∗      9.18∗∗∗       0.65∗∗∗
                                                       (0.17)       (2.34)
                                Nicaragua      17    0.67∗∗∗       2.99∗∗∗       0.80∗∗∗
                                                       (0.08)       (1.10)
                                Panama          9      0.034         -0.31         0.08
                                                       (0.14)       (1.11)
                                Paraguay       18     1.37∗∗∗        -12.3         0.34
                                                       (0.51)      (22.14)
                                Peru           24     0.70∗∗∗      1.11∗∗        0.74   ∗∗∗
                                                       (0.11)       (0.52)
                                Peru2          23     0.73∗∗∗       2.00∗        0.70∗∗∗
                                                       (0.12)       (1.05)
                                US             48     0.44∗∗∗     276.9∗∗∗       0.37∗∗∗
                                                       (0.15)      (71.65)
                                Uruguay        19       -0.16        -41.3         -0.25
                                                       (0.13)      (35.27)
                                Venezuela      19    0.70∗∗∗       1.11∗∗        0.76∗∗∗
                                                       (0.11)       (0.52)


     Note: Beta from OLS regression of Current Population Density on Pre-colonial Population Density in both Log-
     Log and Level-Level forms. Final column is Spearman rank correlation coefficient. Current Population Density
     is the log of the total population in 2000 divided by the area of the state or province in square kilometers, from
     national censuses, and Bruhn and Gallego (2010). Pre-colonial Population Density is the log of the number of
     indigenous people per square kilometer before the arrival of Columbus, from Denevan (1992), and Bruhn and
     Gallego (2010). More detailed data sources and descriptions in the text. 1. Mexico without border states. 2
     Peru without Lambayeque. Robust standard errors in parentheses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                            52
Table 6: Population Density in 2000, Pre-colonial Population Density, and Locational
Fundamentals (pooled)
                                    (1)       (2)          (3)           (4)             (5)         (6)          (7)
                                   OLS      Between     Within FE     Within FE     Within FE      MS FE      MS FE
           Pre-colonial Density   7.2∗∗∗      1.9        8.6∗∗∗        8.8∗∗∗        8.8∗∗∗        3.0∗∗∗      0.6∗∗∗
                                  (1.37)     (3.80)      (0.73)        (0.57)         (0.68)       (0.11)      (0.13)
           Agriculture                                                                  -4.3                     0.3
                                                                                      (8.42)                   (0.41)
           Agriculture2                                                                  8.4                      0.1
                                                                                     (11.90)                   (0.48)
           Rivers                                                                        0.2                     0.10
                                                                                      (2.89)                   (0.11)
           Rivers2                                                                      -0.2                   -0.02∗
                                                                                      (0.39)                   (0.01)
           Landlocked                                                                   -1.6                   -0.1∗∗
                                                                                      (2.03)                   (0.05)
           Temperature                                                                   0.6                     0.03
                                                                                      (0.38)                   (0.02)
           Temperature2                                                               -0.02∗                  -0.0009
                                                                                      (0.01)                   (0.00)
           Altitude                                                                     -0.8                    -0.04
                                                                                      (1.28)                   (0.06)
           Altitude2                                                                     0.3                    0.010
                                                                                      (0.40)                   (0.02)
           Rainfall                                                                    2.0∗                    0.10∗
                                                                                      (1.12)                   (0.05)
           Rainfall2                                                                  -0.3∗∗                 -0.04∗∗∗
                                                                                      (0.14)                   (0.01)
           Constant                 1.1∗      1.3∗∗       0.9∗∗∗        1.0∗∗∗          -4.2       0.07∗∗∗    -0.3∗∗∗
                                   (0.53)    (0.54)       (0.07)        (0.05)        (7.31)        (0.01)     (0.11)
           N                        365        365         365           330            330          330         330
           R2                      0.045     -0.045       0.057         0.060          0.068

Note: Regression of Current Population Density against Pre-colonial Population Density. Estimation by OLS and robust MS
regression with country �xed effects. Current Population Density is the total population in 2000 divided by the area of the
state or province in square kilometers, from national censuses, and Bruhn and Gallego (2010). Pre-colonial Population Density
is the number of indigenous people per square kilometer before the arrival of Columbus, from Denevan (1992), and Bruhn
and Gallego (2010). Agriculture is an index of probability of cultivation given cultivable land, climate and soil composition,
from Ramankutty, Foley and McSweeney (2002). Rivers captures the density of rivers as a share of land area derived from
HydroSHEDS (USGS 2011). Landlocked is a dummy variable for whether the state has access to a coast or not; temperature
is a yearly average in â—¦ C; altitude measures the elevation of the capital city of the state in kilometers; and Rainfall captures
total yearly rainfall in meters, all are from Bruhn and Gallego (2010). More detailed data sources and descriptions in the text.
Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                              53
Table 7: Log Population Density in 2000, Log Pre-colonial Population Density, and
Locational Fundamentals (pooled)
                                     (1)       (2)          (3)            (4)             (5)         (6)           (7)
                                    OLS      Between     Within FE      Within FE     Within FE      MS FE       MS FE
          Pre-colonial Density     0.3∗∗∗     0.3∗∗       0.4∗∗∗         0.5∗∗∗          0.3∗∗       0.5∗∗∗       0.4∗∗∗
                                   (0.10)     (0.12)      (0.14)         (0.13)         (0.11)       (0.06)       (0.05)
          Agriculture                                                                      2.1                      0.3
                                                                                        (1.63)                    (1.15)
          Agriculture2                                                                    -0.2                       0.3
                                                                                        (1.59)                    (1.20)
          Rivers                                                                          -0.3                    0.7∗∗∗
                                                                                        (0.38)                    (0.24)
          Rivers2                                                                        -0.07                   -0.2∗∗∗
                                                                                        (0.05)                    (0.03)
          Landlocked                                                                      -0.1                      -0.2
                                                                                        (0.25)                    (0.14)
          Temperature                                                                   0.4∗∗∗                    0.2∗∗∗
                                                                                        (0.08)                    (0.04)
          Temperature2                                                                 -0.01∗∗∗                 -0.006∗∗∗
                                                                                        (0.00)                    (0.00)
          Altitude                                                                        -0.3                     -0.3∗
                                                                                        (0.42)                    (0.16)
          Altitude2                                                                        0.1                      0.05
                                                                                        (0.08)                    (0.05)
          Rainfall                                                                         0.3                    1.0∗∗∗
                                                                                        (0.26)                    (0.23)
          Rainfall2                                                                      -0.06                   -0.2∗∗∗
                                                                                        (0.04)                    (0.05)
          Constant                  0.10       0.1          0.6            1.1∗           -2.2          0.6      -2.4∗∗∗
                                   (0.36)     (0.55)       (0.59)         (0.55)        (1.37)        (0.41)      (0.79)
          N                         365        365          365            330            330          330          330
          R2                       0.136      0.282        0.147          0.206          0.432

Note: Regression of the Log of Current Population Density against the Log of Pre-colonial Population Density. Estimation by
OLS and robust MS regression with country �xed effects. Current Population Density is the log of the total population in 2000
divided by the area of the state or province in square kilometers, from national censuses, and Bruhn and Gallego (2010). Pre-
colonial Population Density is the log of the number of indigenous people per square kilometer before the arrival of Columbus,
from Denevan (1992), and Bruhn and Gallego (2010). Agriculture is an index of probability of cultivation given cultivable land,
climate and soil composition, from Ramankutty, Foley and McSweeney (2002). Rivers captures the density of rivers as a share
of land area derived from HydroSHEDS (USGS 2011). Landlocked is a dummy variable for whether the state has access to a
coast or not; temperature is a yearly average in â—¦ C; altitude measures the elevation of the capital city of the state in kilometers;
and Rainfall captures total yearly rainfall in meters, all are from Bruhn and Gallego (2010). More detailed data sources and
descriptions in the text. Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                                54
Table 8: Income per Capita in 2005 and Pre-colonial Population Density (country by country)


                                               N       β Pop. Density           Rank
                                                     Log-Log    Log-Level    Correlation
                                Argentina     24     -0.11∗∗∗    -27.7∗∗∗     -0.53∗∗
                                                       (0.04)      (8.25)
                                Brazil        27       -0.082       -3.73       -0.22
                                                       (0.07)      (2.84)
                                Bolivia        9       0.11∗         7.86        0.47
                                                       (0.06)      (5.58)
                                Chile         13    -0.070∗∗∗    -6.57∗∗∗      -0.55∗∗∗
                                                       (0.02)      (2.44)
                                Canada        13       0.029       6.74∗         0.14
                                                       (0.02)      (3.76)
                                Colombia      30      0.19∗∗∗    4.91∗∗∗       0.75∗∗∗
                                                       (0.03)      (1.07)
                                Ecuador       22       -0.019        0.31        0.01
                                                       (0.04)      (1.32)
                                El Salvador   14      0.62∗∗     2.60∗∗∗         0.45
                                                       (0.25)      (0.80)
                                Guatemala      8        0.071        0.62       -0.07
                                                       (0.19)      (1.65)
                                Honduras      18       -0.034       -0.46       -0.04
                                                       (0.04)      (0.55)
                                Mexico        32       -0.060      0.059       -0.40∗∗∗
                                                       (0.04)      (0.04)
                                Mexico1       25        0.020     0.11∗∗∗       -0.12
                                                       (0.05)      (0.03)
                                Nicaragua     17     0.058∗∗     0.46∗∗∗        0.36∗∗
                                                       (0.03)      (0.16)
                                Panama         9       0.014        -0.70       -0.07
                                                       (0.04)      (1.31)
                                Paraguay      18       -0.012        0.75        0.02
                                                       (0.06)      (5.86)
                                Peru          24        0.041        0.39        0.13
                                                       (0.05)      (0.36)
                                Peru2         23        0.054     1.12∗∗∗        0.11
                                                       (0.06)      (0.38)
                                US            48    0.045∗∗∗     10.9∗∗∗        0.31∗∗
                                                       (0.02)      (3.22)
                                Uruguay       19       -0.030       -0.69       -0.38
                                                       (0.02)      (9.80)
                                Venezuela     19        0.041       0.39         0.10
                                                       (0.05)      (0.36)


     Note: Beta from OLS regression of Income per capita in 2000 (PPP 2005 US dollars) on Pre-colonial Population
     Density in both Log-Log and Log-Level forms. Income per capita is taken from national censuses. Final column
     is Spearman rank correlation coefficient. Pre-colonial Population Density is the number of indigenous people per
     square kilometer before the arrival of Columbus, from Denevan (1992), and Bruhn and Gallego (2010). More
     detailed data sources and descriptions in the text. 1. Mexico without border states. 2 Peru without Lambayeque.
     Robust standard errors in parentheses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                          55
Table 9: Log Income per Capita in 2005, Pre-colonial Population Density, and Locational
Fundamentals (pooled)
                                     (1)        (2)          (3)            (4)             (5)          (6)         (7)
                                     OLS      Between     Within FE      Within FE     Within FE      MS FE      MS FE
           Pre-colonial Density      -0.4       -2.8        0.1∗∗         0.09∗∗        0.09∗∗∗       0.06∗∗∗     0.1∗∗∗
                                    (0.58)     (1.70)      (0.04)         (0.04)         (0.02)        (0.02)     (0.02)
           Agriculture                                                                     -0.4                     0.3
                                                                                         (0.23)                   (0.30)
           Agriculture2                                                                     0.2                     -0.3
                                                                                         (0.24)                   (0.31)
           Rivers                                                                         -0.04                  0.0003
                                                                                         (0.10)                   (0.04)
           Rivers2                                                                      -0.0004                   -0.004
                                                                                         (0.01)                   (0.01)
           Landlocked                                                                      0.01                     0.01
                                                                                         (0.05)                   (0.03)
           Temperature                                                                     0.02                   0.006
                                                                                         (0.02)                   (0.01)
           Temperature2                                                                 -0.0008                  -0.0003
                                                                                         (0.00)                   (0.00)
           Altitude                                                                       -0.02                    -0.1∗
                                                                                         (0.09)                   (0.07)
           Altitude2                                                                      -0.03                    -0.02
                                                                                         (0.02)                   (0.02)
           Rainfall                                                                       -0.05                  -0.2∗∗∗
                                                                                         (0.04)                   (0.04)
           Rainfall2                                                                     -0.004                  0.01∗∗∗
                                                                                         (0.01)                   (0.00)
           Constant                 9.1∗∗∗     9.1∗∗∗       9.0∗∗∗         9.0∗∗∗        9.4∗∗∗        9.5∗∗∗     9.1∗∗∗
                                    (0.28)     (0.24)       (0.00)         (0.00)        (0.21)        (0.17)     (0.14)
           N                         365        365          365            330            330          330         330
           R2                       0.010      0.093        0.004          0.003         0.128

Note: Regression of the Log of Income per capita in 2000 (PPP 2005 US dollars) against Pre-colonial Population Density.
Estimation by OLS and robust MS regression with country �xed effects. Income per capita (in logs) is taken from national
censuses. Pre-colonial Population Density is the number of indigenous people per square kilometer before the arrival of Columbus,
from Denevan (1992), and Bruhn and Gallego (2010). Agriculture is an index of probability of cultivation given cultivable land,
climate and soil composition, from Ramankutty, Foley and McSweeney (2002). Rivers captures the density of rivers as a share
of land area derived from HydroSHEDS (USGS 2011). Landlocked is a dummy variable for whether the state has access to a
coast or not; temperature is a yearly average in â—¦ C; altitude measures the elevation of the capital city of the state in kilometers;
and Rainfall captures total yearly rainfall in meters, all are from Bruhn and Gallego (2010). More detailed data sources and
descriptions in the text. Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                                56
Table 10: Log Income per Capita in 2005, Log Pre-colonial Population Density, and
Locational Fundamentals (pooled)
                                      (1)        (2)         (3)            (4)              (5)        (6)           (7)
                                     OLS      Between     Within FE      Within FE      Within FE     MS FE       MS FE
           Pre-colonial Density     -0.2∗∗     -0.2∗∗       -0.01          -0.02           -0.02      0.003       -0.03∗∗
                                    (0.08)     (0.09)      (0.02)         (0.02)          (0.02)      (0.02)       (0.01)
           Agriculture                                                                      -0.3                     -0.4
                                                                                          (0.19)                   (0.39)
           Agriculture2                                                                      0.2                      0.2
                                                                                          (0.20)                   (0.35)
           Rivers                                                                          -0.03                    -0.06
                                                                                          (0.10)                   (0.09)
           Rivers2                                                                        -0.002                    0.005
                                                                                          (0.01)                   (0.02)
           Landlocked                                                                     0.005                   -0.1∗∗∗
                                                                                          (0.05)                   (0.05)
           Temperature                                                                      0.02                   -0.009
                                                                                          (0.02)                   (0.02)
           Temperature2                                                                  -0.0009                  -0.0001
                                                                                          (0.00)                   (0.00)
           Altitude                                                                        -0.01                   0.2∗∗∗
                                                                                          (0.08)                   (0.06)
           Altitude2                                                                       -0.03                 -0.09∗∗∗
                                                                                          (0.02)                   (0.01)
           Rainfall                                                                        -0.05                   -0.009
                                                                                          (0.04)                   (0.07)
           Rainfall2                                                                      -0.004                   -0.01∗
                                                                                          (0.01)                   (0.01)
           Constant                 8.3∗∗∗     8.0∗∗∗       9.0∗∗∗         8.9∗∗∗         9.3∗∗∗       9.5∗∗∗      9.3∗∗∗
                                    (0.28)     (0.39)       (0.08)         (0.09)         (0.32)       (0.17)      (0.24)
           N                         365        365          365            330             330         330          330
           R2                       0.193      0.240        0.002          0.009          0.129

Note: Regression of the Log of Income per capita in 2000 (PPP 2005 US dollars) against the Log of Pre-colonial Population
Density. Estimation by OLS and robust MS regression with country �xed effects. Income per capita (in logs) is taken from
national censuses. Pre-colonial Population Density is the log of the number of indigenous people per square kilometer before the
arrival of Columbus, from Denevan (1992), and Bruhn and Gallego (2010). Agriculture is an index of probability of cultivation
given cultivable land, climate and soil composition, from Ramankutty, Foley and McSweeney (2002). Rivers captures the density
of rivers as a share of land area derived from HydroSHEDS (USGS 2011). Landlocked is a dummy variable for whether the state
has access to a coast or not; temperature is a yearly average in â—¦ C; altitude measures the elevation of the capital city of the state
in kilometers; and Rainfall captures total yearly rainfall in meters, all are from Bruhn and Gallego (2010). More detailed data
sources and descriptions in the text. Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                                 57
Table 11: Log Income per Capita in 2005, Pre-colonial Population Density, and Locational
Fundamentals (pooled without Argentina and Chile)
                                                    Log-Level                                   Log-Log
                                       (1)         (2)       (3)          (4)      (5)        (6)       (7)        (8)
                                       FE         FE      MS FE       MS FE        FE         FE      MS FE     MS FE
            Pre-colonial Density     0.09∗∗    0.09∗∗∗    0.06∗∗∗      0.1∗∗∗    0.0003      0.003     0.02      -0.004
                                     (0.04)     (0.02)     (0.02)      (0.02)    (0.02)     (0.01)    (0.01)     (0.02)
            Agriculture                           -0.2                  0.5∗                  -0.2                  0.5
                                                (0.27)                 (0.29)               (0.26)               (0.33)
            Agriculture2                          0.04                  -0.5∗                 0.06                 -0.5
                                                (0.24)                 (0.32)               (0.23)               (0.34)
            Rivers                                -0.1                  -0.04                 -0.1                -0.05
                                                (0.07)                 (0.05)               (0.07)               (0.06)
            Rivers2                              0.01                 -0.0008                 0.01               0.002
                                                (0.01)                 (0.01)               (0.01)               (0.01)
            Landlocked                            0.03                   0.02                 0.03                -0.02
                                                (0.05)                 (0.05)               (0.05)               (0.08)
            Temperature                          0.02                  -0.003                0.02                -0.004
                                                (0.02)                 (0.04)               (0.02)               (0.03)
            Temperature2                       -0.0007               -0.00004              -0.0008              -0.0001
                                                (0.00)                 (0.00)               (0.00)               (0.00)
            Altitude                             0.03                    -0.1                0.03                  -0.1
                                                (0.07)                 (0.09)               (0.06)               (0.15)
            Altitude2                          -0.04∗∗                  -0.02              -0.04∗∗                -0.02
                                                (0.02)                 (0.03)               (0.02)               (0.04)
            Rainfall                             -0.05                -0.2∗∗∗                -0.05              -0.2∗∗∗
                                                (0.05)                 (0.06)               (0.05)               (0.05)
            Rainfall2                           -0.004                  0.010               -0.004                0.007
                                                (0.01)                 (0.01)               (0.01)               (0.01)
            Constant                 9.0∗∗∗     9.5∗∗∗    8.0∗∗∗       8.7∗∗∗    9.0∗∗∗     9.5∗∗∗    8.1∗∗∗     9.4∗∗∗
                                     (0.00)     (0.24)    (0.10)       (0.52)    (0.08)     (0.26)    (0.10)     (0.34)
            N                         293         293      293           293      293         293      293         293
            R2                       0.004       0.129                           -0.003     0.122

Note: Regression of Income per capita in 2000 (PPP 2005 US dollars) against Pre-colonial Population Density. Excluding two
countries with prominent negative correlations: Chile and Argentina. Speci�cations in Log-Level and Log-Log form. Estimation
by OLS and robust MS regression with country �xed effects. Income per capita (in logs) is taken from national censuses.
Pre-colonial Population Density is the number of indigenous people per square kilometer before the arrival of Columbus, from
Denevan (1992), and Bruhn and Gallego (2010). Agriculture is an index of probability of cultivation given cultivable land,
climate and soil composition, from Ramankutty, Foley and McSweeney (2002). Rivers captures the density of rivers as a share
of land area derived from HydroSHEDS (USGS 2011). Landlocked is a dummy variable for whether the state has access to a
coast or not; temperature is a yearly average in â—¦ C; altitude measures the elevation of the capital city of the state in kilometers;
and Rainfall captures total yearly rainfall in meters, all are from Bruhn and Gallego (2010). More detailed data sources and
descriptions in the text. Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                                58
Table 12: Log Income per Capita in 2005, Pre-colonial Population Density, and Slavery
(Brazil, Colombia and United States)
                                          (1)        (2)          (3)        (4)        (5)        (6)          (7)
                                          OLS        OLS         OLS        OLS        OLS         MS           MS
               Pre-colonial Density      2.9∗∗        1.9       2.6∗∗     5.5∗∗∗       2.5∗      6.5∗∗∗      8.6∗∗∗
                                         (1.16)     (1.33)      (1.27)     (1.45)     (1.46)     (1.56)       (1.65)
               Brazil                   -1.9∗∗∗    -2.0∗∗∗     -1.6∗∗∗    -1.6∗∗∗    -1.5∗∗∗    -2.0∗∗∗      -2.1∗∗∗
                                         (0.09)     (0.11)      (0.21)     (0.20)     (0.23)     (0.07)       (0.05)
               Colombia                 -2.5∗∗∗    -2.4∗∗∗     -2.4∗∗∗    -2.6∗∗∗    -2.3∗∗∗    -2.6∗∗∗      -2.9∗∗∗
                                         (0.07)     (0.09)      (0.08)     (0.08)     (0.23)     (0.08)       (0.16)
               South                    -0.09∗∗    -0.1∗∗∗        0.2       0.09        0.09      -0.01       0.1∗∗∗
                                         (0.04)     (0.04)      (0.13)     (0.13)     (0.14)     (0.07)       (0.03)
               Slavery                                        -0.009∗∗     -0.006    -0.006∗     -0.002    -0.006∗∗∗
                                                                (0.00)     (0.00)     (0.00)     (0.00)       (0.00)
               Slavery*Pop                                                 -0.1∗∗      -0.09    -0.1∗∗∗      -0.1∗∗∗
                                                                           (0.05)     (0.05)     (0.04)       (0.02)
               Agriculture                                                             -0.08                 0.5∗∗∗
                                                                                      (0.89)                  (0.18)
               Agriculture2                                                             -0.1                  -0.4∗∗
                                                                                      (0.75)                  (0.18)
               Rivers                                                                 -0.002                  0.09∗
                                                                                      (0.17)                  (0.05)
               Rivers2                                                                -0.009                -0.02∗∗∗
                                                                                      (0.02)                  (0.01)
               Landlocked                                                               0.02                    0.01
                                                                                      (0.10)                  (0.03)
               Temperature                                                             0.04                    -0.01
                                                                                      (0.03)                  (0.01)
               Temperature2                                                           -0.001                 0.0006
                                                                                      (0.00)                  (0.00)
               Altitude                                                                  0.3                 0.4∗∗∗
                                                                                      (0.20)                  (0.03)
               Altitude2                                                                -0.1                 -0.2∗∗∗
                                                                                      (0.08)                  (0.01)
               Rainfall                                                               -0.004                  0.1∗∗∗
                                                                                      (0.10)                  (0.02)
               Rainfall2                                                              -0.007                -0.02∗∗∗
                                                                                      (0.01)                  (0.00)
               Constant                 10.7∗∗∗    10.7∗∗∗    10.7∗∗∗     10.7∗∗∗    10.7∗∗∗    10.7∗∗∗     10.4∗∗∗
                                         (0.02)     (0.03)     (0.03)      (0.02)     (0.32)     (0.04)       (0.09)
               N                          105         78         78          78          78        78            78
               R2                        0.937      0.940      0.947       0.953       0.953

Note: Dependent variable is the Log Income per capita in 2000 (PPP 2005 US dollars). Pre-colonial Population Density is the
number of indigenous people per square kilometer before the arrival of Columbus. Estimation by OLS and robust MS regression
with country �xed effects. Income per capita (in logs) is taken from national censuses. Pre-colonial Population Density is the
number of indigenous people per square kilometer before the arrival of Columbus, from Denevan (1992), and Bruhn and Gallego
(2010). Dummies for Brazil, Colombia, and the US South (according to the US Census). Slavery is measured as a fraction of
the population and is taken from Bergad (2008) and Nunn (2008). Interaction of slavery with Pre-colonial population density.
Agriculture is an index of probability of cultivation given cultivable land, climate and soil composition, from Ramankutty, Foley
and McSweeney (2002). Rivers captures the density of rivers as a share of land area derived from HydroSHEDS (USGS 2011).
Landlocked is a duMSy variable for whether the state has access to a coast or not; temperature is a yearly average in â—¦ C; altitude
measures the elevation of the capital city of the state in kilometers; and Rainfall captures total yearly rainfall in meters, all are
from Bruhn and Gallego (2010). More detailed data sources and descriptions in the text. Robust SE for OLS and MS SE are
in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                                59
Table 13: Log Income per Capita in 2005, Log Pre-colonial Population Density, and Slavery
(Brazil, Colombia and United States)
                                         (1)        (2)         (3)       (4)         (5)       (6)          (7)
                                        OLS        OLS         OLS       OLS        OLS        MS           MS
              Pre-colonial Density    0.05∗∗       0.04       0.06∗    0.08∗∗∗     0.05∗      0.06∗       0.02∗∗
                                       (0.02)     (0.03)      (0.03)    (0.02)     (0.03)     (0.03)       (0.01)
              Brazil                  -2.0∗∗∗    -2.0∗∗∗     -1.6∗∗∗   -1.6∗∗∗    -1.5∗∗∗    -2.0∗∗∗      -1.3∗∗∗
                                       (0.10)     (0.12)      (0.20)    (0.20)     (0.23)     (0.17)       (0.03)
              Colombia                -2.5∗∗∗    -2.4∗∗∗     -2.5∗∗∗   -2.5∗∗∗    -2.3∗∗∗    -2.3∗∗∗      -2.1∗∗∗
                                       (0.08)     (0.10)      (0.09)    (0.09)     (0.19)     (0.09)       (0.04)
              South                     -0.07     -0.1∗∗        0.2       0.2         0.2       0.2       0.07∗∗
                                       (0.04)     (0.05)      (0.15)    (0.15)     (0.17)     (0.20)       (0.03)
              Slavery                                       -0.009∗∗   -0.007∗    -0.006∗     -0.007    -0.003∗∗∗
                                                              (0.00)    (0.00)     (0.00)     (0.01)       (0.00)
              Slavery*Pop                                                -0.07      -0.06     -0.002     -0.04∗∗∗
                                                                        (0.04)     (0.05)     (0.04)       (0.01)
              Agriculture                                                            -0.5                  1.0∗∗∗
                                                                                   (0.91)                  (0.21)
              Agriculture2                                                            0.2                 -0.7∗∗∗
                                                                                   (0.77)                  (0.16)
              Rivers                                                                0.006                 -0.2∗∗∗
                                                                                   (0.17)                  (0.03)
              Rivers2                                                              -0.009                0.01∗∗∗
                                                                                   (0.02)                  (0.00)
              Landlocked                                                             0.05                  -0.006
                                                                                   (0.10)                  (0.01)
              Temperature                                                            0.04                 0.02∗∗
                                                                                   (0.03)                  (0.01)
              Temperature2                                                        -0.001∗              -0.0008∗∗∗
                                                                                   (0.00)                  (0.00)
              Altitude                                                                0.3                  0.1∗∗∗
                                                                                   (0.20)                  (0.02)
              Altitude2                                                             -0.1∗                 -0.03∗∗
                                                                                   (0.08)                  (0.01)
              Rainfall                                                              -0.04                  0.2∗∗∗
                                                                                   (0.10)                  (0.04)
              Rainfall2                                                            -0.003                 -0.2∗∗∗
                                                                                   (0.01)                  (0.02)
              Constant                11.0∗∗∗    11.0∗∗∗    11.1∗∗∗    11.2∗∗∗    11.1∗∗∗   11.1∗∗∗      11.0∗∗∗
                                       (0.14)     (0.16)     (0.17)     (0.14)     (0.39)    (0.18)        (0.09)
              N                         105         78         78         78          78       78            78
              R2                       0.935      0.940      0.947      0.948       0.953

Note: Dependent variable is the Log Income per capita in 2000 (PPP 2005 US dollars). Pre-colonial Population Density is the
number of indigenous people per square kilometer before the arrival of Columbus. Estimation by OLS and robust MS regression
with country �xed effects. Income per capita (in logs) is taken from national censuses. Pre-colonial Population Density is the
log of the number of indigenous people per square kilometer before the arrival of Columbus, from Denevan (1992), and Bruhn
and Gallego (2010). Dummies for Brazil, Colombia, and the US South (according to the US Census). Slavery is measured
as a fraction of the population and is taken from Bergad (2008) and Nunn (2008). Interaction of slavery with Pre-colonial
population density. Agriculture is an index of probability of cultivation given cultivable land, climate and soil composition,
from Ramankutty, Foley and McSweeney (2002). Rivers captures the density of rivers as a share of land area derived from
HydroSHEDS (USGS 2011). Landlocked is a dummy variable for whether the state has access to a coast or not; temperature
is a yearly average in â—¦ C; altitude measures the elevation of the capital city of the state in kilometers; and Rainfall captures
total yearly rainfall in meters, all are from Bruhn and Gallego (2010). More detailed data sources and descriptions in the text.
Robust SE for OLS and MS SE are in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                              60