Reversal of Fortune: Geography and Institutions in the ...tion and Marginalization” in Bergen, The Canadian Institute of Advanced Re-search, Brown University, the University of Chicago,

REVERSAL OF FORTUNE: GEOGRAPHY ANDINSTITUTIONS IN THE MAKING OF THE MODERN

WORLD INCOME DISTRIBUTION*

DARON ACEMOGLUSIMON JOHNSON

JAMES A. ROBINSON

Among countries colonized by European powers during the past 500 years,those that were relatively rich in 1500 are now relatively poor. We document thisreversal using data on urbanization patterns and population density, which, weargue, proxy for economic prosperity. This reversal weighs against a view thatlinks economic development to geographic factors. Instead, we argue that thereversal reects changes in the institutions resulting from European colonialism.The European intervention appears to have created an “institutional reversal”among these societies, meaning that Europeans were more likely to introduceinstitutions encouraging investment in regions that were previously poor. Thisinstitutional reversal accounts for the reversal in relative incomes. We providefurther support for this view by documenting that the reversal in relative incomestook place during the late eighteenth and early nineteenth centuries, and resultedfrom societies with good institutions taking advantage of the opportunity toindustrialize.

I. INTRODUCTION

This paper documents a reversal in relative incomes amongthe former European colonies. For example, the Mughals in Indiaand the Aztecs and Incas in the Americas were among the richestcivilizations in 1500, while the civilizations in North America,New Zealand, and Australia were less developed. Today theUnited States, Canada, New Zealand, and Australia are an orderof magnitude richer than the countries now occupying the terri-tories of the Mughal, Aztec, and Inca Empires.

* We thank Joshua Angrist, Abhijit Banerjee, Olivier Blanchard, AlessandraCassella, Jan de Vries, Ronald Findlay, Jeffry Frieden, Edward Glaeser, HerschelGrossman, Lawrence Katz, Peter Lange, Jeffrey Sachs, Andrei Shleifer, FabrizioZilibotti, three anonymous referees, and seminar participants at the All-Univer-sities of California History Conference at Berkeley, the conference on “Globaliza-tion and Marginalization” in Bergen, The Canadian Institute of Advanced Re-search, Brown University, the University of Chicago, Columbia University, theUniversity of Houston, Indiana University, Massachusetts Institute of Technol-ogy, National Bureau of Economic Research summer institute, Stanford Univer-sity, the Wharton School of the University of Pennsylvania, and Yale Universityfor useful comments. Acemoglu gratefully acknowledges nancial help from TheCanadian Institute for Advanced Research and the National Science FoundationGrant SES-0095253. Johnson thanks the Massachusetts Institute of TechnologyEntrepreneurship Center for support.

© 2002 by the President and Fellows of Harvard College and the Massachusetts Institute ofTechnology.The Quarterly Journal of Economics, November 2002

1231

Our main measure of economic prosperity in 1500 is urban-ization. Bairoch [1988, Ch. 1] and de Vries [1976, p. 164] arguethat only areas with high agricultural productivity and a devel-oped transportation network can support large urban popula-tions. In addition, we present evidence that both in the timeseries and the cross section there is a close association betweenurbanization and income per capita.1 As an additional proxy forprosperity we use population density, for which there are rela-tively more extensive data. Although the theoretical relationshipbetween population density and prosperity is more complex, itseems clear that during preindustrial periods only relativelyprosperous areas could support dense populations.

With either measure, there is a negative association betweeneconomic prosperity in 1500 and today. Figure I shows a negativerelationship between the percent of the population living in townswith more than 5000 inhabitants in 1500 and income per capitatoday. Figure II shows the same negative relationship betweenlog population density (number of inhabitants per square kilome-ter) in 1500 and income per capita today. The relationships shownin Figures I and II are robust—they are unchanged when wecontrol for continent dummies, the identity of the colonial power,religion, distance from the equator, temperature, humidity, re-sources, and whether the country is landlocked, and when weexclude the “neo-Europes” (the United States, Canada, New Zea-land, and Australia) from the sample.

This pattern is interesting, in part, because it provides anopportunity to distinguish between a number of competing theo-ries of the determinants of long-run development. One of the mostpopular theories, which we refer to as the “geography hypothe-sis,” explains most of the differences in economic prosperity bygeographic, climatic, or ecological differences across countries.The list of scholars who have emphasized the importance ofgeographic factors includes, inter alia, Machiavelli [1519], Mon-

1. By economic prosperity or income per capita in 1500, we do not refer to theeconomic or social conditions or the welfare of the masses, but to a measure oftotal production in the economy relative to the number of inhabitants. Althoughurbanization is likely to have been associated with relatively high output percapita, the majority of urban dwellers lived in poverty and died young because ofpoor sanitary conditions (see, for example, Bairoch [1988, Ch. 12]).

It is also important to note that the Reversal of Fortune refers to changes inrelative incomes across different areas, and does not imply that the initial in-habitants of, for example, New Zealand or North America themselves becamerelatively rich. In fact, much of the native population of these areas did notsurvive European colonialism.

1232 QUARTERLY JOURNAL OF ECONOMICS

tesquieu [1748], Toynbee [1934 –1961], Marshall [1890], andMyrdal [1968], and more recently, Diamond [1997] and Sachs[2000, 2001]. The simplest version of the geography hypothesisemphasizes the time-invariant effects of geographic variables,such as climate and disease, on work effort and productivity, andtherefore predicts that nations and areas that were relatively richin 1500 should also be relatively prosperous today. The reversalin relative incomes weighs against this simple version of thegeography hypothesis.

More sophisticated versions of this hypothesis focus on thetime-varying effects of geography. Certain geographic character-istics that were not useful, or even harmful, for successful eco-nomic performance in 1500 may turn out to be benecial later on.A possible example, which we call “the temperate drift hypothe-sis,” argues that areas in the tropics had an early advantage, butlater agricultural technologies, such as the heavy plow, croprotation systems, domesticated animals, and high-yield crops,have favored countries in the temperate areas (see Bloch [1966],Lewis [1978], and White [1962]; also see Sachs [2001]). Althoughplausible, the temperate drift hypothesis cannot account for the

FIGURE ILog GDP per Capita (PPP) in 1995 against Urbanization Rate in 1500

Note. GDP per capita is from the World Bank [1999]; urbanization in 1500 ispeople living in towns with more than 5000 inhabitants divided by total popu-lation, from Bairoch [1988] and Eggimann [1999]. Details are in Appendices 1and 2.

1233REVERSAL OF FORTUNE

reversal. First, the reversal in relative incomes seems to be re-lated to population density and prosperity before Europeans ar-rived, not to any inherent geographic characteristics of the area.Furthermore, according to the temperate drift hypothesis, thereversal should have occurred when European agricultural tech-nology spread to the colonies. Yet, while the introduction of Eu-ropean agricultural techniques, at least in North America, tookplace earlier, the reversal occurred during the late eighteenth andearly nineteenth centuries, and is closely related to industrializa-tion. Another version of the sophisticated geography hypothesiscould be that certain geographic characteristics, such as the pres-ence of coal reserves or easy access to the sea, facilitated indus-trialization (e.g., Pomeranz [2000] and Wrigley [1988]). But we donot nd any evidence that these geographic factors caused indus-trialization. Our reading of the evidence therefore provides littlesupport to various sophisticated geography hypotheses either.

An alternative view, which we believe provides the best ex-planation for the patterns we document, is the “institutions hy-pothesis,” relating differences in economic performance to theorganization of society. Societies that provide incentives and op-portunities for investment will be richer than those that fail to doso (e.g., North and Thomas [1973], North and Weingast [1989],

FIGURE IILog GDP per Capita (PPP) against Log Population Density in 1500

Note. GDP per capita from the World Bank [1999]; log population density in1500 from McEvedy and Jones [1978]. Details are in Appendix 2.


and Olson [2000]). As we discuss in more detail below, we hy-pothesize that a cluster of institutions ensuring secure propertyrights for a broad cross section of society, which we refer to asinstitutions of private property, are essential for investment in-centives and successful economic performance. In contrast, ex-tractive institutions, which concentrate power in the hands of asmall elite and create a high risk of expropriation for the majorityof the population, are likely to discourage investment and eco-nomic development. Extractive institutions, despite their adverseeffects on aggregate performance, may emerge as equilibriuminstitutions because they increase the rents captured by thegroups that hold political power.

How does the institutions hypothesis explain the reversal inrelative incomes among the former colonies? The basic idea isthat the expansion of European overseas empires starting at theend of the fteenth century caused major changes in the organi-zation of many of these societies. In fact, historical and econo-metric evidence suggests that European colonialism caused an“institutional reversal”: European colonialism led to the develop-ment of institutions of private property in previously poor areas,while introducing extractive institutions or maintaining existingextractive institutions in previously prosperous places.2 Themain reason for the institutional reversal is that relatively poorregions were sparsely populated, and this enabled or inducedEuropeans to settle in large numbers and develop institutionsencouraging investment. In contrast, a large population and rela-tive prosperity made extractive institutions more protable forthe colonizers; for example, the native population could be forcedto work in mines and plantations, or taxed by taking over existingtax and tribute systems. The expansion of European overseasempires, combined with the institutional reversal, is consistentwith the reversal in relative incomes since 1500.

Is the reversal related to institutions? We document that thereversal in relative incomes from 1500 to today can be explained,

2. By the term “institutional reversal,” we do not imply that it was societieswith good institutions that ended up with extractive institutions after Europeancolonialism. First, there is no presumption that relatively prosperous societies in1500 had anything resembling institutions of private property. In fact, theirrelative prosperity most likely reected other factors, and even perhaps geo-graphic factors. Second, the institutional reversal may have resulted more fromthe emergence of institutions of private property in previously poor areas thanfrom a deterioration in the institutions of previously rich areas.


at least statistically, by differences in institutions across coun-tries. The institutions hypothesis also suggests that institutionaldifferences should matter more when new technologies that re-quire investments from a broad cross section of the society be-come available. We therefore expect societies with good institu-tions to take advantage of the opportunity to industrialize, whilesocieties with extractive institutions fail to do so. The data sup-port this prediction.

We are unaware of any other work that has noticed or docu-mented this change in the distribution of economic prosperity.Nevertheless, many historians emphasize that in 1500 the Mu-ghal, Ottoman, and Chinese Empires were highly prosperous, butgrew slowly during the next 500 years (see the discussion andreferences in Section III).

Our overall interpretation of comparative development in theformer colonies is closely related to Coatsworth [1993] and En-german and Sokoloff [1997, 2000], who emphasize the adverseeffects of the plantation complex in the Caribbean and CentralAmerica working through political and economic inequality,3 andto our previous paper, Acemoglu, Johnson, and Robinson [2001a].In that paper we proposed the disease environment at the timeEuropeans arrived as an instrument for European settlementsand the subsequent institutional development of the former col-onies, and used this to estimate the causal effect of institutionaldifferences on economic performance. Our thesis in the currentpaper is related, but emphasizes the inuence of population den-sity and prosperity on the policies pursued by the Europeans (seealso Engerman and Sokoloff [1997]). In addition, here we docu-ment the reversal in relative incomes among the former colonies,show that it was related to industrialization, and provide evi-dence that the interaction between institutions and the opportu-nity to industrialize during the nineteenth century played a cen-tral role in the long-run development of the former colonies.4

3. In this context, see also Frank [1978], Rodney [1972], Wallerstein [1974–1980], and Williams [1944].

4. Our results are also relevant to the literature on the relationship betweenpopulation and growth. The recent consensus is that population density encour-ages the discovery and exchange of ideas, and contributes to growth (e.g., Boserup[1965], Jones [1997], Kremer [1993], Kuznets [1968], Romer [1986], and Simon[1977]). Our evidence points to a major historical episode of 500 years where highpopulation density was detrimental to economic development, and therefore shedsdoubt on the general applicability of this recent consensus.


The rest of the paper is organized as follows. The next sectiondiscusses the construction of urbanization and population densitydata, and provides evidence that these are good proxies for eco-nomic prosperity. Section III documents the “Reversal of For-tune”—the negative relationship between economic prosperity in1500 and income per capita today among the former colonies.Section IV discusses why the simple and sophisticated geographyhypotheses cannot explain this pattern, and how the institutionshypothesis explains the reversal. Section V documents that thereversal in relative incomes reects the institutional reversalcaused by European colonialism, and that institutions startedplaying a more important role during the age of industry. SectionVI concludes.

II. URBANIZATION AND POPULATION DENSITY

II.A. Data on Urbanization

Bairoch [1988] provides the best single collection and assess-ment of urbanization estimates. Our base data for 1500 consist ofBairoch’s [1988] urbanization estimates augmented by the workof Eggimann [1999]. Merging the Eggimann and Bairoch seriesrequires us to convert Eggimann’s estimates, which are based ona minimum population threshold of 20,000, into Bairoch-equiva-lent urbanization estimates, which use a minimum populationthreshold of 5000. We use a number of different methods toconvert between the two sets of estimates, all with similar re-sults. Appendix 1 provides details about data sources and con-struction. Briey, for our base estimates, we run a regression ofBairoch estimates on Eggimann estimates for all countries wherethey overlap in 1900 (the year for which we have most Bairochestimates for non-European countries). This regression yields aconstant of 6.6 and a coefcient of 0.67, which we use to generateBairoch-equivalent urbanization estimates from Eggimann’sestimates.

Alternatively, we converted the Eggimann’s numbers using auniform conversion rate of 2 as suggested by Davis’ and Zipf ’sLaws (see Appendix 1 and Bairoch [1988, Ch. 9]), and also testedthe robustness of the estimates using conversion ratios at theregional level based on Bairoch’s analysis. Finally, we con-structed three alternative series without combining estimatesfrom different sources. One of these is based on Bairoch, the


second on Eggimann, and the third on Chandler [1987]. All fouralternative series are reported in Appendix 3, and results usingthese measures are reported in Table IV.

While the data on sub-Saharan Africa are worse than for anyother region, it is clear that urbanization in sub-Saharan Africabefore 1500 was at a higher level than in North America orAustralia. Bairoch, for example, argues that by 1500 urbaniza-tion was “well-established” in sub-Saharan Africa.5 Becausethere are no detailed urbanization data for sub-Saharan Africa,we leave this region out of the regression analysis when we useurbanization data, although African countries are included in ourregressions using population density.

Table I gives descriptive statistics for the key variables ofinterest, separately for the whole world, for the sample of ex-colonies for which we have urbanization data in 1500, and for thesample of ex-colonies for which we have population density datain 1500. Appendix 2 gives detailed denitions and sources for thevariables used in this study.

II.B. Urbanization and Income

There are good reasons to presume that urbanization andincome are positively related. Kuznets [1968, p. 1] opens his bookon economic growth by stating: “we identify the economic growthof nations as a sustained increase in per-capita or per-workerproduct, most often accompanied by an increase in populationand usually by sweeping structural changes. . . . in the distribu-tion of population between the countryside and the cities, theprocess of urbanization.”

Bairoch [1988] points out that during preindustrial periods alarge fraction of the agricultural surplus was likely to be spent ontransportation, so both a relatively high agricultural surplus anda developed transport system were necessary for large urbanpopulations (see Bairoch [1988, Ch. 1]). He argues “the existenceof true urban centers presupposes not only a surplus of agricul-

5. Sahelian trading cities such as Timbuktu, Gao, and Djenne (all in modernMali) were very large in the middle ages with populations as high as 80,000. Kano(in modern Nigeria) had a population of 30,000 in the early nineteenth century,and Yorubaland (also in Nigeria) was highly urbanized with a dozen towns withpopulations of over 20,000 while its capital Ibadan possibly had 70,000 inhabit-ants. For these numbers and more detail, see Hopkins [1973, Ch. 2].


TA

BL

EI

DE

SC

RIP

TIV

ES

TA

TIS

TIC

S

Who

lew

orld

(1)

Bas

esa

mpl

efo

rur

ban

izat

ion

(2)

Bas

esa

mpl

efo

rpo

pula

tion

dens

ity

(3)

Bel

owm

edia

nu

rban

izat

ion

in15

00(4

)

Abo

vem

edia

nur

bani

zati

onin

1500

(5)

Bel

owm

edia

npo

pula

tion

dens

ity

in15

00(6

)

Abo

vem

edia

npo

pula

tion

den

sity

in15

00(7

)

Log

GD

Ppe

rca

pita

(PP

P)

in19

958.

38.

57.

98.

88.

18.

37.

5(1

.1)

(0.9

)(1

.0)

Urb

aniz

atio

nin

1995

53.0

57.5

45.4

64.9

49.7

53.5

36.7

(23.

8)(2

2.4)

(22.

2)U

rban

izat

ion

in15

007.

36.

46.

42.

410

.52.

39.

5(5

.0)

(5.0

)(5

.0)

Log

popu

lati

onde

nsit

yin

1500

1.0

0.2

0.5

20.

91.

42

0.6

1.6

(1.6

)(1

.9)

(1.5

)P

opul

atio

nde

nsi

tyin

1500

9.2

6.3

4.8

1.2

11.7

0.8

9.1

(24.

3)(1

6.4)

(11.

7)L

ogpo

pula

tion

dens

ity

in10

000.

60.

110.

082

1.20

1.22

20.

941.

04(1

.5)

(2.0

)(1

.5)

Ave

rage

prot

ecti

onag

ain

stex

prop

riat

ion,

1985

–199

57.

16.

96.

57.

56.

36.

86.

2(1

.8)

(1.5

)(1

.4)

Con

stra

int

onth

eex

ecut

ive

in19

903.

64.

93.

75.

14.

64.

03.

5(2

.3)

(2.1

)(2

.3)

Con

stra

int

onth

eex

ecut

ive

inr

stye

arof

inde

pen

den

ce

3.6

3.3

3.4

3.8

2.8

3.6

3.3

(2.4

)(2

.5)

(2.3

)

Eur

opea

nse

ttle

men

tsin

1900

29.6

23.2

12.5

30.5

6.0

18.7

4.7

(41.

7)(2

8.7)

(22.

1)N

umbe

rof

obse

rvat

ion

s16

241

9121

2047

44

Sta

ndar

dde

viat

ions

are

inpa

rent

hese

s.N

umbe

rof

obse

rvat

ion

sva

ries

acro

ssro

ws

due

tom

issi

ngda

ta.

The

rst

thre

eco

lum

ns

repo

rtm

ean

valu

esfo

rth

esa

mpl

ein

dica

ted

atth

ehe

adof

the

colu

mn.

The

last

four

colu

mn

sre

port

mea

nva

lues

for

form

erco

lon

ies

belo

wan

dab

ove

the

med

ian,

sepa

rate

lyfo

rth

eba

seur

bani

zati

onan

dpo

pula

tion

den

sity

sam

ples

.F

orde

tail

edso

urce

san

dde

scri

ptio

ns

see

App

endi

x2.


tural produce, but also the possibility of using this surplus intrade” [p. 11].6 See de Vries [1976, p. 164] for a similar argument.

We supplement this argument by empirically investigatingthe link between urbanization and income in Table II. Columns(1)–(6) present cross-sectional regressions. Column (1) is for 1900,the earliest date for which we have data on urbanization andincome per capita for a large number of countries. The regressioncoefcient, 0.038, is highly signicant, with a standard error of0.006. It implies that a country with 10 percentage points higherurbanization has, on average, 46 percent (38 log points) greaterincome per capita (throughout the paper, all urbanization ratesare expressed in percentage points, e.g., 10 rather than 0.1—seeTable I). Column (2) reports a similar result using data for 1950.Column (3) uses current data and shows that even today there isa strong relationship between income per capita and urbanizationfor a large sample of countries. The coefcient is similar, 0.036,and precisely estimated, with a standard error of 0.002. Thisrelationship is shown diagrammatically in Figure III.

Below, we draw a distinction between countries colonized byEuropeans and those never colonized (i.e., Europe and non-Euro-pean countries not colonized by Western Europe). Columns (4) and(5) report the same regression separately for these two samples. Theestimates are very similar: 0.037 for the former colonies sample, and0.033 for the rest of the countries. Finally, in column (6) we addcontinent dummies to the same regression. This leads to only aslightly smaller coefcient of 0.030, with a standard error of 0.002.

Finally, we use estimates from Bairoch [1978, 1988] to con-struct a small unbalanced panel data set of urbanization andincome per capita from 1750 to 1913. Column (7) reports a re-

6. The view that urbanization and income (productivity) are closely related isshared by many other scholars. See Ades and Glaeser [1999], De Long andShleifer [1993], Tilly and Blockmans [1994], and Tilly [1990]. De Long andShleifer, for example, write “The larger preindustrial cities were nodes of infor-mation, industry, and exchange in areas where the growth of agricultural pro-ductivity and economic specialization had advanced far enough to support them.They could not exist without a productive countryside and a ourishing tradenetwork. The population of Europe’s preindustrial cities is a rough indicator ofeconomic prosperity” [p. 675].

A large history literature also documents how urbanization accelerated inEurope during periods of economic expansion (e.g., Duby [1974], Pirenne [1956],and Postan and Rich [1966]). For example, the period between the beginning ofthe eleventh and mid-fourteenth centuries is an era of rapid increase in agricul-tural productivity and industrial output. The same period also witnessed a pro-liferation of cities. Bairoch [1988], for example, estimates that the number of citieswith more than 20,000 inhabitants increased from around 43 in 1000 to 107 in1500 [Table 10.2, p. 159].


TA

BL

EII

UR

BA

NIZ

AT

ION

AN

DP

ER

CA

PIT

AIN

CO

ME

Cro

ss-s

ecti

onal

regr

essi

onin

1913

,al

lco

un

trie

s(1

)

Cro

ss-s

ecti

onal

regr

essi

onin

1950

,al

lco

un

trie

s(2

)

Cro

ss-s

ecti

onal

regr

essi

onin

1995

,al

lco

untr

ies

(3)

Cro

ss-s

ecti

onal

regr

essi

onin

1995

,on

lyfo

rex

-col

onie

s(4

)

Cro

ss-s

ecti

onal

regr

essi

onin

1995

,n

ever

colo

niz

edco

untr

ies

only

(5)

Cro

ss-s

ecti

onal

regr

essi

onin

1995

,al

lco

un

trie

s,w

ith

cont

inen

tdu

mm

ies

(6)

Pan

elda

tase

tth

rou

gh19

13(7

)

Dep

end

ent

vari

able

islo

gG

DP

per

capi

ta

Urb

aniz

atio

n0.

038

0.02

60.

036

0.03

70.

033

0.03

00.

026

(0.0

06)

(0.0

02)

(0.0

02)

(0.0

03)

(0.0

07)

(0.0

02)

(0.0

04)

R2

0.69

0.57

0.63

0.69

0.34

0.68

0.93

Nu

mbe

rof

obse

rvat

ion

s22

128

162

9351

162

55

Sta

ndar

der

rors

are

inpa

rent

hese

s.L

ogG

DP

per

capi

tath

roug

h19

13is

from

Bai

roch

[197

8].U

rban

izat

ion

ispe

rcen

tof

popu

lati

onliv

ing

into

wns

wit

hat

leas

t50

00pe

ople

,fr

omB

airo

ch[1

988]

thro

ugh

1900

wit

hsu

pple

men

tary

sour

ces

asde

scri

bed

inA

ppen

dix

1.L

ogG

DP

per

capi

tain

1950

isfr

omM

addi

son

[199

5];t

his

regr

essi

onus

esur

bani

zati

onin

1960

from

the

Wor

ldB

ank’

sW

orld

Dev

elop

men

tIn

dica

tors

[199

9].

Log

GD

Ppe

rca

pita

(PP

P)

and

Urb

aniz

atio

nda

tafo

r19

95ar

efr

omth

eW

orld

Ban

k’s

Wor

ldD

evel

opm

ent

Indi

cato

rs[1

999]

.Pop

ulat

ion

den

sity

isto

tal

popu

lati

ondi

vide

dby

arab

lela

nd

area

,bot

hfr

omM

cEve

dyan

dJo

nes

[197

8].F

orde

tail

edso

urc

esan

dde

scri

ptio

nsse

eA

ppen

dix

2.T

heco

unt

ries

and

appr

oxim

ate

year

sfo

rw

hic

hw

eha

veda

ta(u

sed

inth

eu

nbal

ance

dpa

nelr

egre

ssio

nin

colu

mn

(7))

are

Aus

tral

ia(1

830,

1860

,and

1913

),A

ust

ria

(183

0,18

60,

1913

),B

elgi

um(1

830,

1860

,19

13),

Bri

tain

(175

0,18

30,

1860

,19

13),

Bul

gari

a(1

860,

1913

),C

anad

a(1

830,

1860

,19

13),

Chi

na

(183

0,18

60),

Den

mar

k(1

830,

1860

,19

13),

Fin

lan

d(1

830,

1860

,191

3),F

ranc

e(1

750,

1830

,186

0,19

13),

Ger

man

y(1

830,

1860

,191

3),G

reec

e(1

860,

1913

),In

dia

(183

0,19

13),

Ital

y(1

830,

1860

,191

3),J

amai

ca(1

830,

1913

),Ja

pan

(175

0,18

30,1

913)

,Net

herl

ands

(183

0,18

60,1

913)

,Nor

way

(183

0,18

60,1

913)

,Por

tuga

l(18

30,1

860,

1913

),R

oman

ia(1

830,

1860

,191

3),R

ussi

a(1

750,

1830

,186

0,19

13),

Spa

in(1

830,

1860

,19

13),

Swed

en(1

830,

1860

,191

3),S

wit

zerl

and

(183

0,18

60,1

913)

,U

nite

dS

tate

s(1

750,

1830

,186

0,19

13),

and

Yug

osla

via

(183

0,18

60,1

913)

.


gression of income per capita on urbanization using this paneldata set and controlling for country and period dummies. Theestimate is again similar: 0.026 (s.e. 5 0.004). Overall, we con-clude that urbanization is a good proxy for income.

II.C. Population Density and Income

The most comprehensive data on population since 1 A.D.come from McEvedy and Jones [1978]. They provide estimatesbased on censuses and published secondary sources. Whilesome individual country numbers have since been revised andothers remain contentious (particularly for pre-Columbian Meso-America), their estimates are consistent with more recent re-search (see, for example, the recent assessment by the Bureauof the Census, www.census.gov/ipc/www/worldhis.html). We useMcEvedy and Jones [1978] for our baseline estimates, and testthe effect of using alternative assumptions (e.g., lower or higherpopulation estimates for Mexico and its neighbors before thearrival of Cortes).

FIGURE IIILog GDP per Capita (PPP) in 1995 against the Urbanization Rate in 1995

Note. GDP per capita and urbanization are from the World Bank [1999]. Ur-banization is percent of population living in urban areas. The denition of urbanareas differs between countries, but the usual minimum size is 2000–5000 inhabi-tants. For details of denitions and sources for urban population in 1995, see theUnited Nations [1998].


http://www.census.gov/ipc/www/worldhis.html

We calculate population density by dividing total populationby arable land (also estimated by McEvedy and Jones). Thisexcludes primarily desert, inland water, and tundra. As much aspossible, we use the land area of a country at the date we areconsidering.

The theoretical relationship between population density andincome is more nuanced than that between urbanization andincome. With a similar reasoning, it seems natural to think thatonly relatively rich areas could afford dense populations (seeBairoch [1988, Ch. 1]). This is also in line with Malthus’ classicwork. Malthus [1798] argued that high productivity increasespopulation by raising birthrates and lowering death rates. How-ever, the main thrust of Malthus’ work was how a higher thanequilibrium level of population increases death rates and reducesbirthrates to correct itself.7 A high population could therefore bereecting an “excess” of population, causing low income per cap-ita. So caution is required in interpreting population density as aproxy for income per capita.

The empirical evidence regarding the relationship betweenpopulation density and income is also less clear-cut than therelationship between urbanization and income. In Acemoglu,Johnson, and Robinson [2001b] we documented that populationdensity and income per capita increased concurrently in manyinstances. Nevertheless, there is no similar cross-sectional rela-tionship in recent data, most likely because of the demographictransition—it is no longer true that high population density isassociated with high income per capita because the relationshipbetween income and the number of children has changed (e.g.,Notestein [1945] or Livi-Bacci [2001]).

Despite these reservations, we present results using popula-tion density, as well as urbanization, as a proxy for income percapita. This is motivated by three considerations. First, popula-tion density data are more extensive, so the use of populationdensity data is a useful check on our results using urbanizationdata. Second, as argued by Bairoch, population density is closely

7. A common interpretation of Malthus’ argument is that these populationdynamics will force all countries down to the subsistence level of income. In thatcase, population density would be a measure of total income, but not necessarilyof income per capita, and in fact, there would be no systematic (long-run) differ-ences in income per capita across countries. We view this interpretation asextreme, and existing historical evidence suggests that there were systematicdifferences in income per capita between different regions even before the modernperiod (see the references below).


related to urbanization, and in fact, our measures are highlycorrelated. Third, variation in population density will play animportant role not only in documenting the reversal, but also inexplaining it.

III. THE REVERSAL OF FORTUNE

III.A. Results with Urbanization

This section presents our main results. Figure I in the intro-duction depicts the relationship between urbanization 1500 andincome per capita today. Table III reports regressions document-ing the same relationship. Column (1) is our most parsimoniousspecication, regressing log income per capita in 1995 (PPP basis)on urbanization rates in 1500 for our sample of former colonies.The coefcient is 20.078 with a standard error of 0.026.8 Thiscoefcient implies that a 10 percentage point lower urbanizationin 1500 is associated with approximately twice as high GDP percapita today (78 log points 108 percent). It is important to notethat this is not simply mean reversion—i.e., richer than averagecountries reverting back to the mean. It is a reversal. To illustratethis, let us compare Uruguay and Guatemala. The native popu-lation in Uruguay had no urbanization, while, according to ourbaseline estimates Guatemala had an urbanization rate of 9.2percent. The estimate in column (1) of Table II, 0.038, for therelationship between income and urbanization implies that Gua-temala at the time was approximately 42 percent richer thanUruguay (exp (0.038 3 9.2) 2 1 0.42). According to our estimatein column (1) of Table III, we expect Uruguay today to be 105percent richer than Guatemala (exp (0.078 3 9.2) 2 1 1.05),which is approximately the current difference in income per cap-ita between these two countries.9

The second column of Table III excludes North African coun-tries for which data quality may be lower. The result is un-

8. Because China was never a formal colony, we do not include it in oursample of ex-colonies. Adding China does not affect our results. For example, withChina, the baseline estimate changes from 20.078 (s.e. 5 0.026) to 20.079 (s.e. 50.025). Furthermore, our sample excludes countries that were colonized by Euro-pean powers briey during the twentieth century, such as Iran, Saudi Arabia, andSyria. If we include these observations, the results are essentially unchanged. Forexample, the baseline estimate changes to 20.072 (s.e. 5 0.024).

9. Interestingly, these calculations suggest that not only have relative rank-ings reversed since 1500, but income differences are now much larger than in1500.


changed, with a coefcient of 20.101 and standard error of 0.032.Column (3) drops the Americas, which increases both the coef-cient and the standard error, but the estimate remains highlysignicant. Column (4) reports the results just for the Americas,where the relationship is somewhat weaker but still signicant atthe 8 percent level. Column (5) adds continent dummies to checkwhether the relationship is being driven by differences acrosscontinents. Although continent dummies are jointly signicant,the coefcient on urbanization in 1500 is unaffected—it is 20.083with a standard error of 0.030.

One might also be concerned that the relationship is beingdriven mainly by the neo-Europes: United States, Canada, NewZealand, and Australia. These countries are settler colonies builton lands that were inhabited by relatively undeveloped civiliza-tions. Although the contrast between the development experi-ences of these areas and the relatively advanced civilizations ofIndia or Central America is of central importance to the reversal andto our story, one would like to know whether there is anything morethan this contrast in the results of Table III. In column (6) we dropthese observations. The relationship is now weaker, but still nega-tive and statistically signicant at the 7 percent level.

In column (7) we control for distance from the equator (theabsolute value of latitude), which does not affect the pattern ofthe reversal—the coefcient on urbanization in 1500 is now20.072 instead of 20.078 in our baseline specication. Distancefrom the equator is itself insignicant. Column (8), in turn, con-trols for a variety of geography variables that represent the effectof climate, such as measures of temperature, humidity, and soiltype, with little effect on the relationship between urbanization in1500 and income per capita today. The R2 of the regressionincreases substantially, but this simply reects the addition ofsixteen new variables to this regression (the adjusted R2 in-creases only slightly, to 0.27).

In column (9) we control for a variety of “resources” whichmay have been important for post-1500 development. These in-clude dummies for being an island, for being landlocked, and forhaving coal reserves and a variety of other natural resources (seeAppendix 2 for detailed denitions and sources). Access to the seamay have become more important with the rise of trade, andavailability of coal or other natural resources may have differenteffects at different points in time. Once again, the addition ofthese variables has no effect on the pattern of the reversal.


TA

BL

EII

IU

RB

AN

IZA

TIO

NIN

1500

AN

DG

DP

PE

RC

AP

ITA

IN19

95F

OR

FO

RM

ER

EU

RO

PE

AN

CO

LO

NIE

S

Dep

ende

nt

vari

able

islo

gG

DP

per

capi

ta(P

PP

)in

1995

Bas

esa

mpl

e(1

)

Wit

hou

tN

orth

Afr

ica

(2)

Wit

hou

tth

eA

mer

icas

(3)

Just

the

Am

eric

as(4

)

Wit

hco

nti

nen

tdu

mm

ies

(5)

Wit

hou

tne

o-E

uro

pes

(6)

Con

trol

lin

gfo

rla

titu

de(7

)

Con

trol

lin

gfo

rcl

imat

e(8

)

Con

trol

lin

gfo

rre

sou

rces

(9)

Con

trol

lin

gfo

rco

lon

ial

orig

in(1

0)

Con

trol

lin

gfo

rre

ligi

on(1

1)

Urb

aniz

atio

nin

1500

20.

078

20.

101

20.

115

20.

053

20.

083

20.

046

20.

072

20.

088

20.

058

20.

071

20.

060

(0.0

26)

(0.0

32)

(0.0

51)

(0.0

29)

(0.0

30)

(0.0

26)

(0.0

25)

(0.0

30)

(0.0

29)

(0.0

28)

(0.0

33)

Asi

adu

mm

y2

1.33

(0.6

1)A

fric

adu

mm

y2

0.53

(0.7

7)A

mer

ica

dum

my

20.

96(0

.57)

Lat

itu

de1.

42(0

.92)

P-v

alu

efo

rte

mpe

ratu

re[0

.51]

P-v

alu

efo

rhu

mid

ity

[0.4

0]

P-v

alu

efo

rso

ilqu

alit

y[0

.96]

P-v

alu

efo

rre

sou

rces

[0.1

6]


Lan

dloc

ked

20.

54(0

.48)

Isla

nd

0.27

(0.3

3)C

oal

0.11

(0.2

8)F

orm

erF

renc

hco

lon

y2

0.59

(0.3

9)F

orm

erS

pan

ish

colo

ny

0.06

(0.2

9)P

-val

ue

for

reli

gion

[0.4

7]

R2

0.19

0.22

0.26

0.13

0.32

0.09

0.24

0.53

0.45

0.27

0.25

Nu

mbe

rof

obse

rvat

ions

4137

1724

4137

4141

4141

41

Sta

ndar

der

rors

are

inpa

ren

thes

es.P

-val

ues

from

F-t

ests

for

join

tsi

gni

can

cear

ein

squ

are

brac

kets

.Dep

ende

ntva

riab

leis

log

GD

Ppe

rca

pita

(PP

P)

in19

95.

Bas

esa

mpl

eis

allf

orm

erco

loni

esfo

rw

hic

hw

eha

veda

ta.U

rban

izat

ion

in15

00is

perc

ent

ofth

epo

pula

tion

livi

ng

into

wns

wit

h50

00or

mor

ein

habi

tant

s.T

he

regr

essi

onth

atin

clud

esco

ntin

ent

dum

mie

sha

sO

cean

iaas

the

base

cate

gory

.The

neo-

Eu

rope

sar

eth

eU

nite

dS

tate

s,C

anad

a,A

ust

rali

a,an

dN

ewZe

alan

d.In

the

“cli

mat

e”re

gres

sion

we

incl

ude

ve

mea

sure

sof

tem

pera

ture

,fo

urm

easu

res

ofhu

mid

ity,

and

seve

nm

easu

res

ofso

ilqu

alit

y.In

the

“res

ourc

es”

regr

essi

onw

ein

clud

ere

serv

esof

gold

,iro

n,zi

nc,

silv

er,a

ndoi

l.C

oali

sa

dum

my

for

the

pres

ence

ofco

al,l

andl

ocke

dis

adu

mm

yfo

rno

tha

ving

acce

ssto

the

sea,

and

isla

ndis

adu

mm

yfo

rbe

ing

anis

lan

d.T

here

gres

sion

that

cont

rols

for

colo

nial

orig

inin

clud

esdu

mm

ies

for

form

erF

renc

hco

lony

,Spa

nis

hco

lon

y,P

ortu

gues

eco

lony

,Bel

gian

colo

ny,

Ital

ian

colo

ny,G

erm

anco

lony

,and

Dut

chco

lony

.B

riti

shco

loni

esar

eth

eba

seca

tego

ry.

The

reli

gion

vari

able

sar

epe

rcen

tof

the

popu

lati

onw

hoar

eM

uslim

,C

atho

lic,

and

“oth

er”;

perc

ent

Pro

test

ant

isth

eba

seca

tego

ry.F

orde

taile

dso

urc

esan

dde

scri

ptio

nsse

eA

ppen

dix

2.


Finally, in columns (10) and (11) we add the identity of thecolonial power and religion, which also have little effect on ourestimate, and are themselves insignicant.

The urbanization variable used in Table III relies on work byBairoch and Eggimann. In Table IV we use data from Bairoch andEggimann separately, as well as data from Chandler, who pro-vided the starting point for Bairoch’s data. We report a subset ofthe regressions from Table III using these three different seriesand an alternative series using the Davis-Zipf adjustment toconvert Eggimann’s estimates into Bairoch-equivalent numbers(explained in Appendix 1). The results are very similar to thebaseline estimates reported in Table III: in all cases, there is anegative relationship between urbanization in 1500 and incomeper capita today, and in almost all cases, this relationship isstatistically signicant at the 5 percent level (the full set ofresults are reported in Acemoglu, Johnson, and Robinson [2001b]).

III.B. Results with Population Density

In Panel A of Table V we regress income per capita today onlog population density in 1500, and also include data for sub-Saharan Africa. The results are similar to those in Table IV (alsosee Figure II). In all specications we nd that countries withhigher population density in 1500 are substantially poorer today.The coefcient of 20.38 in column (1) implies that a 10 percenthigher population density in 1500 is associated with a 4 percentlower income per capita today. For example, the area now corre-sponding to Bolivia was seven times more densely settled thanthe area corresponding to Argentina; so on the basis of thisregression, we expect Argentina to be three times as rich asBolivia, which is more or less the current gap in income betweenthese countries.10

The remaining columns perform robustness checks, andshow that including a variety of controls for geography and re-sources, the identity of the colonial power, religion variables, ordropping the Americas, the neo-Europes, or North Africa has very

10. The magnitudes implied by the estimates in this table are similar to thoseimplied by the estimates in Table III. For example, the difference in the urban-ization rate between an average high and low urbanization country in 1500 is 8.1(see columns (4) and (5) in Table I), which using the coefcient of 20.078 fromTable III translates into a 0.078 3 8.1 0.63 log points difference in current GDP.The difference in log population density between an average high-density andlow-density country in 1500 is 2.2 (see columns (6) and (7) in Table I), whichtranslates into a 0.38 3 2.2 0.84 log points difference in current GDP.


little effect on the results. In all cases, log population density in1500 is signicant at the 1 percent level (although now some ofthe controls, such as the humidity dummies, are also signicant).

TABLE IVALTERNATIVE MEASURES OF URBANIZATION

Dependent variable is log GDP per capita (PPP) in 1995

Basesample

(1)

With continentdummies

(2)

Withoutneo-Europes

(3)

Controllingfor latitude

(4)

Controllingfor resources

(5)

Panel A: Using our base sample measure of urbanization

Urbanization in 1500 20.078 20.083 20.046 20.072 20.058(0.026) (0.030) (0.026) (0.025) (0.029)

R2 0.19 0.32 0.09 0.24 0.45Number of observations 41 41 37 41 41

Panel B: Using only Bairoch’s estimates

Urbanization in 1500 20.126 20.107 20.089 20.116 20.092(0.032) (0.034) (0.033) (0.036) (0.037)


Panel C: Using only Eggimann’s estimates

Urbanization in 1500 20.041 20.043 20.022 20.036 20.022(0.019) (0.019) (0.018) (0.019) (0.023)


Panel D: Using only Chandler’s estimates

Urbanization in 1500 20.057 20.072 20.040 20.054 20.049(0.019) (0.021) (0.019) (0.019) (0.025)


Panel E: Using Davis-Zipf Adjustment for Eggimann’s series

Urbanization in 1500 20.039 20.048 20.024 20.040 20.031(0.015) (0.020) (0.014) (0.015) (0.017)


Standard errors are in parentheses. Dependent variable is log GDP per capita (PPP) in 1995. Basesample is all former colonies for which we have data. Urbanization in 1500 is percent of the population livingin towns with 5000 or more people. In Panels B, C, D, and E, we use, respectively, Bairoch’s estimates,Eggimann’s estimates, Chandler’s estimates, and a conversion of Eggimann’s estimates into Bairoch-equiva-lent numbers using the Davis-Zipf adjustment. Eggimann’s estimates (Panel C) and Chandler’s estimates(Panel D) are not converted to Bairoch-equivalent units. The continent dummies, neo-Europes,and resourcesmeasures are as described in the note to Table III. For detailed sources and descriptions see Appendix 2. Thealternative urbanization series are shown in Appendix 3.


TA

BL

EV

PO

PU

LA

TIO

ND

EN

SIT

YA

ND

GD

PP

ER

CA

PIT

AIN

FO

RM

ER

EU

RO

PE

AN

CO

LO

NIE

S

Dep

ende

nt

vari

able

islo

gG

DP

per

capi

ta(P

PP

)in

1995

Bas

esa

mpl

e(1

)

Wit

hou

tA

fric

a(2

)

Wit

hou

tth

eA

mer

icas

(3)

Just

the

Am

eric

as(4

)

Wit

hco

nti

nen

tdu

mm

ies

(5)

Wit

hou

tn

eo-

Eu

rope

s(6

)

Con

trol

lin

gfo

rla

titu

de(7

)

Con

trol

lin

gfo

rcl

imat

e(8

)

Con

trol

ling

for

reso

urc

es(9

)

Con

trol

lin

gfo

rco

lon

ial

orig

in(1

0)

Con

trol

lin

gfo

rre

ligi

on(1

1)

Pan

elA

:L

ogpo

pula

tion

den

sity

in15

00as

ind

epen

den

tva

riab

le

Log

popu

lati

onde

nsi

tyin

1500

20.

382

0.40

20.

322

0.25

20.

262

0.32

20.

332

0.31

20.

302

0.32

20.

37(0

.06)

(0.0

5)(0

.07)

(0.0

9)(0

.05)

(0.0

6)(0

.06)

(0.0

6)(0

.06)

(0.0

6)(0

.07)

Asi

adu

mm

y2

0.91

(0.5

5)A

fric

adu

mm

y2

1.67

(0.5

2)A

mer

ica

dum

my

20.

69(0

.51)

Lat

itu

de2.

09(0

.74)

P-v

alu

efo

rte

mpe

ratu

re[0

.18]

P-v

alu

efo

rh

um

idit

y[0

.00]

P-v

alu

efo

rso

ilqu

alit

y[0

.10]

P-v

alu

efo

rn

atur

alre

sou

rces

[0.3

4]

Lan

dloc

ked

20.

58(0

.23)

Isla

nd

0.62

(0.2

3)


Coa

l0.

01(0

.19)

For

mer

Fre

nch

colo

ny

20.

48(0

.20)

For

mer

Spa

nis

hco

lon

y0.

25(0

.22)

P-v

alu

efo

rre

ligi

on[0

.73]

R2

0.34

0.55

0.27

0.22

0.56

0.24

0.40

0.59

0.54

0.48

0.36

Nu

mbe

rof

obse

rvat

ion

s91

4758

3391

8791

9085

9185

Pan

elB

:L

ogpo

pula

tion

and

log

lan

din

1500

asse

para

tein

dep

end

ent

vari

able

s

Log

popu

lati

onin

1500

20.

342

0.30

20.

322

0.13

20.

232

0.27

20.

292

0.27

20.

272

0.28

20.

31(0

.05)

(0.0

5)(0

.07)

(0.0

7)(0

.05)

(0.0

5)(0

.05)

(0.0

5)(0

.05)

(0.0

5)(0

.06)

Log

arab

lela

nd

in15

000.

260.

270.

210.

160.

180.

150.

200.

200.

080.

210.

24(0

.06)

(0.0

6)(0

.09)

(0.0

6)(0

.05)

(0.0

6)(0

.06)

(0.0

6)(0

.07)

(0.0

6)(0

.07)

R2

0.35

0.45

0.31

0.17

0.55

0.31

0.41

0.59

0.55

0.47

0.36

Nu

mbe

rof

obse

rvat

ion

s91

4758

3391

8791

9085

9185

Pan

elC

:U

sin

gpo

pula

tion

den

sity

in10

00A

.D.

asan

inst

rum

ent

for

popu

lati

ond

ensi

tyin

1500

A.D

.

Log

popu

lati

onde

nsi

tyin

1500

20.

312

0.4

20.

152

0.38

20.

182

0.22

20.

272

0.26

20.

222

0.26

20.

25(0

.06)

(0.0

6)(0

.08)

(0.1

1)(0

.07)

(0.0

8)(0

.06)

(0.0

7)(0

.07)

(0.0

6)(0

.08)

Nu

mbe

rof

obse

rvat

ion

s83

4351

3283

8083

8378

8377

Sta

ndar

der

rors

are

inpa

ren

thes

es.P

-val

ues

from

F-t

ests

for

join

tsi

gni

can

cear

ein

squ

are

brac

kets

.Dep

ende

ntva

riab

leis

log

GD

Ppe

rca

pita

(PP

P)

in19

95.

Bas

esa

mpl

eis

allf

orm

erco

lon

ies

for

whi

chw

eha

veda

ta.P

opu

lati

onde

nsit

yin

1500

isto

talp

opul

atio

ndi

vide

dby

arab

lela

ndar

ea.S

eeT

able

III

for

anex

plan

atio

nof

the

sam

ple

and

cova

riat

esin

each

colu

mn.

For

deta

iled

sour

ces

and

desc

ript

ion

sse

eA

ppen

dix

2.


The estimates in the top panel of Table V use variation inpopulation density, which reects two components: differences inpopulation and differences in arable land area. In Panel B weseparate the effects of these two components and nd that theycome in with equal and opposite signs, showing that the speci-cation with population density is appropriate. In Panel C we usepopulation density in 1000 as an instrument for population den-sity in 1500. This is useful since, as discussed in subsection II.C,differences in long-run population density are likely to be betterproxies for income per capita. Instrumenting for population den-sity in 1500 with population density in 1000 isolates the long-runcomponent of population density differences across countries (i.e.,the component of population density in 1500 that is correlatedwith population density in 1000). The Two-Stage Least Squares(2SLS) results in Panel C using this instrumental variables strat-egy are very similar to the OLS results in Panel A.

III.C. Further Results, Robustness Checks, and Discussion

Caution is required in interpreting the results presented inTables III, IV, and V. Estimates of urbanization and population in1500 are likely to be error-ridden. Nevertheless, the rst effect ofmeasurement error would be to create an attenuation bias toward0. Therefore, one might think that the negative coefcients inTables III, IV, and V are, if anything, underestimates. A moreserious problem would be if errors in the urbanization and popu-lation density estimates were not random, but correlated withcurrent income in some systematic way. We investigate this issuefurther in Table VI, using a variety of different estimates forurbanization and population density. Columns (1)–(5), for exam-ple, show that the results are robust to a variety of modicationsto the urbanization data.

Much of the variation in urbanization and population densityin 1500 was not at the level of these countries, but at the level of“civilizations.” For example, in 1500 there were fewer separatecivilizations in the Americas, and even arguably in Asia, thanthere are countries today. For this reason, in column (6) we repeatour key regressions using variation in urbanization and popula-tion density only among fourteen civilizations (based on Toynbee[1934 –1961] and McNeill [1999]—see the note to Table VI). Theresults conrm our basic ndings, and show a statistically signi-cant negative relationship between prosperity in 1500 and today.Columns (7) and (8) report robustness checks using variants of


the population density data constructed under different assump-tions, again with very similar results.

Is there a similar reversal among the noncolonies? Column(9) reports a regression of log GDP per capita in 1995 on urban-ization in 1500 for all noncolonies (including Europe), and column(10) reports the same regression for Europe (including EasternEurope). In both cases, there is a positive relationship betweenurbanization in 1500 and income today.11 This suggests that thereversal reects an unusual event, and is likely to be related tothe effect of European colonialism on these societies.

Panel B of Table VI reports results weighted by population in1500, with very similar results. In Panel C we include urbaniza-tion and population density simultaneously in these regressions.In all cases, population density is negative and highly signicant,while urbanization is insignicant. This is consistent with thenotion, discussed below, that differences in population densityplayed a key role in the reversal in relative incomes among thecolonies (although it may also reect measurement error in theurbanization estimates).

As a nal strategy to deal with the measurement error inurbanization, we use log population density as an instrument forurbanization rates in 1500. When both of these are valid proxiesfor economic prosperity in 1500 and the measurement error isclassical, this procedure corrects for the measurement error prob-lem. Not surprisingly, these instrumental-variables estimatesreported in the bottom panel of Table VI are considerably largerthan the OLS estimates in Table III. For example, the baselineestimate is now 20.18 instead of 20.08 in Table III. The generalpattern of reversal in relative incomes is unchanged, however.

Is the reversal shown in Figures I and II and Tables III, IV,and V consistent with other evidence? The literature on thehistory of civilizations documents that 500 years ago many partsof Asia were highly prosperous (perhaps as prosperous as West-ern Europe), and civilizations in Meso-America and North Africawere relatively developed (see, e.g., Abu-Lughod [1989], Braudel[1992], Chaudhuri [1990], Hodgson [1993], McNeill [1999], Po-meranz [2000], Reid [1988, 1993], and Townsend [2000]). In con-

11. In Acemoglu, Johnson, and Robinson [2001b] we also provided evidencethat urbanization and population density in 1000 are positively correlated withurbanization and population density in 1500, suggesting that before 1500 therewas considerable persistence in prosperity both where the Europeans later colo-nized and where they never colonized.


TA

BL

EV

IR

OB

US

TN

ES

SC

HE

CK

SF

OR

UR

BA

NIZ

AT

ION

AN

DL

OG

PO

PU

LA

TIO

ND

EN

SIT

Y

Dep

ende

ntva

riab

leis

log

GD

Ppe

rca

pita

(PP

P)

in19

95

Bas

esa

mpl

e(1

)

Ass

um

ing

low

eru

rban

izat

ion

inth

eA

mer

icas

(2)

Ass

um

ing

low

eru

rban

izat

ion

inN

orth

Afr

ica

(3)

Ass

umin

glo

wer

urb

aniz

atio

nin

Indi

ansu

bcon

tin

ent

(4)

Usi

ng

leas

tfa

vora

ble

com

bin

atio

nof

assu

mpt

ion

s(5

)

Usi

ng

augm

ente

dT

oyn

bee

de

nit

ion

ofci

vili

zati

on(6

)

Usi

ng

land

area

in19

95fo

rpo

pula

tion

den

sity

(7)

Alt

ern

ativ

eas

sum

ptio

nsfo

rlo

gpo

pula

tion

den

sity

(8)

All

cou

ntr

ies

nev

erco

lon

ized

byE

uro

pe(9

)

Eu

rope

(in

clu

din

gE

aste

rnE

uro

pe)

(10)

For

mer

colo

nie

sN

ever

colo

niz

ed

Pan

elA

:U

nw

eigh

ted

regr

essi

ons

Urb

aniz

atio

nin

1500

20.

078

20.

089

20.

102

20.

073

20.

105

20.

117

0.06

80.

077

(0.0

26)

(0.0

27)

(0.0

29)

(0.0

27)

(0.0

32)

(0.0

52)

(0.0

23)

(0.0

23)

Log

popu

lati

onde

nsi

tyin

1500

20.

412

0.32

(0.0

6)(0

.07)

R2

0.20

0.22

0.24

0.16

0.21

0.30

0.35

0.21

0.18

0.27

Nu

mbe

rof

obse

rvat

ions

4141

4141

4114

9191

4332

Pan

elB

:R

egre

ssio

ns

wei

ghte

du

sing

log

popu

lati

onin

1500

Urb

aniz

atio

nin

1500

20.

072

20.

084

20.

097

20.

064

20.

099

20.

118

20.

064

20.

073

(0.0

25)

(0.0

26)

(0.0

29)

(0.0

26)

(0.0

32)

(0.0

53)

(0.0

23)

(0.0

22)

Log

popu

lati

onde

nsi

tyin

1500

20.

392

0.29

(0.0

6)(0

.07)

R2

0.18

0.22

0.23

0.14

0.20

0.29

0.32

0.19

0.17

0.24

Nu

mbe

rof

obse

rvat

ions

4141

4141

4114

9191

4332


Pan

elC

:In

clu

din

gbo

thu

rban

izat

ion

and

log

popu

lati

ond

ensi

tyas

ind

epen

den

tva

riab

les

Urb

aniz

atio

nin

1500

0.03

80.

039

0.01

70.

037

0.02

00.

072

0.01

70.

003

0.02

80.

032

(0.0

28)

(0.0

31)

(0.0

33)

(0.0

27)

(0.0

35)

(0.0

47)

(0.0

23)

(0.0

22)

(0.0

20)

(0.0

21)

Log

popu

lati

onde

nsi

tyin

1500

20.

412

0.41

20.

362

0.40

20.

372

0.48

20.

432

0.41

0.34

0.37

(0.0

7)(0

.08)

(0.0

7)(0

.07)

(0.0

7)(0

.09)

(0.0

7)(0

.07)

(0.0

7)(0

.08)

R2

0.56

0.56

0.54

0.56

0.54

0.79

0.61

0.60

0.48

0.57

Nu

mbe

rof

obse

rvat

ions

4141

4141

4114

4141

4332

Pan

elD

:In

stru

men

tin

gfo

ru

rban

izat

ion

in15

00u

sin

glo

gpo

pula

tion

den

sity

in15

00

Urb

aniz

atio

nin

1500

20.

178

20.

181

20.

215

20.

194

20.

242

20.

237

20.

217

20.

239

0.25

90.

226

(0.0

4)(0

.040

)(0

.048

)(0

.048

)(0

.057

)(0

.080

)(0

.053

)(0

.063

)(0

.090

)(0

.074

)N

um

ber

ofob

serv

atio

ns

4141

4141

4114

4141

4332

Sta

ndar

der

rors

are

inpa

rent

hese

s.D

epen

dent

vari

able

islo

gG

DP

per

capi

ta(P

PP

)in

1995

.Bas

esa

mpl

eis

all

form

erco

loni

esfo

rw

hic

hw

eha

veda

ta.I

nou

rba

sesa

mpl

e,u

rban

izat

ion

in15

00is

perc

ent

ofth

epo

pula

tion

livi

ngin

tow

ns

wit

h50

00or

mor

epe

ople

.Col

umn

(2)

assu

mes

9pe

rcen

tur

bani

zati

onin

the

And

esan

dC

entr

alA

mer

ica.

Col

um

n(3

)as

sum

es10

perc

ent

urba

niz

atio

nin

Nor

thA

fric

a.C

olum

n(4

)as

sum

es6

perc

ent

urba

niza

tion

inth

eIn

dian

subc

onti

nent

.Col

umn

(5)

com

bine

sth

eas

sum

ptio

nsof

colu

mn

s(2

),(3

),(4

),an

d(5

)to

crea

teth

ele

ast

favo

rabl

eco

mbi

nati

onof

assu

mpt

ions

for

our

hyp

othe

sis.

Col

umn

(6)

ison

lyci

vili

zati

ons

info

rmer

Eur

opea

nco

loni

es.T

heau

gmen

ted

Toy

nbee

civi

liza

tion

s,u

sed

inco

lum

n(6

),in

clud

eA

ndea

n,M

exic

,Y

uca

tec,

Ara

bic

(Nor

thA

fric

a),

Hin

du,

Pol

ynes

ian,

Esk

imo

(Can

ada)

Nor

thA

mer

ican

Indi

an,

Sou

thA

mer

ican

Indi

an(B

razi

l/Arg

enti

na/C

hile

),A

ustr

alia

nA

bori

gine

,M

alay

(Mal

aysi

aan

dIn

done

sia)

,P

hili

ppin

es,

Vie

tnam

/Cam

bodi

a,an

dB

urm

a.In

colu

mn

(7)

popu

lati

onde

nsi

tyin

1500

isto

tal

popu

lati

ondi

vide

dby

arab

lela

nd

area

in19

95.C

olum

n(8

)ha

lves

the

popu

lati

onde

nsi

tyes

tim

ates

for

Afr

ica.

For

deta

iled

sou

rces

and

desc

ript

ions

see

App

endi

x2.


trast, there was little agriculture in most of North America andAustralia, at most consistent with a population density of 0.1people per square kilometer. McEvedy and Jones [1978, p. 322]describe the state of Australia at this time as “an unchangingpalaeolithic backwater.” In fact, because of the relative backward-ness of these areas, European powers did not view them asvaluable colonies. Voltaire is often quoted as referring to Canadaas a “few acres of snow,” and the European powers at the timepaid little attention to Canada relative to the colonies in the WestIndies. In a few parts of North America, along the East Coast andin the Southwest, there was settled agriculture, supporting apopulation density of approximately 0.4 people per square kilo-meter, but this was certainly much less than that in the Aztec andInca Empires, which had fully developed agriculture with a popu-lation density of between 1 and 3 people (or even higher) persquare kilometer, and also much less than the correspondingnumbers in Asia and Africa [McEvedy and Jones 1978, p. 273].The recent work by Maddison [2001] also conrms our interpre-tation. He estimates that India, Indonesia, Brazil, and Mexicowere richer than the United States in 1500 and 1700 (see, forexample, his Table 2-22a).

III.D. The Timing and Nature of the Reversal

The evidence presented so far documents the reversal inrelative incomes among the former colonies from 1500 to today.When did this reversal take place? This question is relevant inthinking about the causes of the reversal. For example, if thereversal is related to the extraction of resources from, and the“plunder” of, the former colonies, or to the direct effect of thediseases Europeans brought to the New World, it should havetaken place shortly after colonization.

Figure IV shows that the reversal is mostly a late eighteenth-and early nineteenth-century phenomenon, and is closely relatedto industrialization. Figure IVa compares the evolution of urban-ization among two groups of New World ex-colonies, those withlow urbanization in 1500 versus those with high urbanization in1500.12 We focus on New World colonies since the societies came

12. The initially high urbanization countries for which we have data and areincluded in the gure are Bolivia, Mexico, Peru, and all of Central America, whilethe initially low urbanization countries are Argentina, Brazil, Canada, Chile, andthe United States.


under European dominance very early on. The averages plottedin the gure are weighted by population in 1500. In addition, inthe same gure we plot India and the United States separately(as well as including it in the initially low urbanization group).The gure shows that the initially low urbanization group as awhole and the United States by itself overtake India and the ini-tially high urbanization countries sometime between 1750 and 1850.

Figure IVb depicts per capita industrial production for theUnited States, Canada, New Zealand, Australia, Brazil, Mexico,and India using data from Bairoch [1982]. This gure shows thetakeoff in industrial production in the United States, Australia,Canada, and New Zealand relative to Brazil, Mexico, and India.Although the scale makes it difcult to see in the gure, percapita industrial production in 1750 was in fact higher in India, 7,than in the United States, 4 (with U. K. industrial production percapita in 1900 normalized to 100). Bairoch [1982] also reportsthat in 1750 China had industrial production per capita twice the

FIGURE IVaUrbanization Rate in India, the United States, and New World Countries

with Low and High Urbanization, 800–1920Note. Urbanization is population living in urban areas divided by total popula-

tion. Urban areas have a minimum threshold of 20,000 inhabitants, from Chan-dler [1987], and Mitchell [1993, 1995]. Low urbanization in 1500 countries areArgentina, Brazil, Canada, Chile, and the United States. High urbanization in1500 countries are Bolivia, Ecuador, Mexico, Peru, and all of Central America. Fordetails see Appendix 1.


level of the United States. Yet, as Figure IVb shows, over the next200 years there was a much larger increase in industrial produc-tion in the United States than in India (and also than in China).

This general interpretation, that the reversal in relative in-comes took place during the late eighteenth and early nineteenthcenturies and was linked to industrialization, is also consistent withthe fragmentary evidence we have on other measures of income percapita and industrialization. Coatsworth [1993], Eltis [1995], Enger-man [1981], and Engerman and Sokoloff [1997] provide evidencethat much of Spanish America and the Caribbean were more pros-perous (had higher per capita income) than British North Americauntil the eighteenth century. The future United States rose in percapita income during the 1700s relative to the Caribbean and SouthAmerica, but only really pulled ahead during the late eighteenthand early nineteenth centuries. Maddison’s [2001] numbers alsoshow that India, Indonesia, Brazil, and Mexico were richer than theUnited States in 1700, but had fallen behind by 1820.

U. S. growth during this period also appears to be an indus-try-based phenomenon. McCusker and Menard [1985] and Galen-son [1996] both emphasize that productivity and income growthin North America before the eighteenth century was limited.During the critical period of growth in the United States, between

FIGURE IVbIndustrial Production per Capita, 1750–1953

Note. Index of industrial production with U. K. per capita industrialization in1900 is equal to 100, from Bairoch [1982].


1840 and 1900, there was modest growth in agricultural outputper capita, and very rapid growth in industrial output per capita;the numbers reported by Gallman [2000] imply that between1840 and 1900 agricultural product per capita increased by about30 percent, a very small increase relative to the growth in manu-facturing output per capita, which increased more than fourfold.

IV. HYPOTHESES AND EXPLANATIONS

IV.A. The Geography Hypothesis

The geography hypothesis claims that differences in eco-nomic performance reect differences in geographic, climatic, andecological characteristics across countries. There are many differ-ent versions of this hypothesis. Perhaps the most common is theview that climate has a direct effect on income through its inu-ence on work effort. This idea dates back to Machiavelli [1519]and Montesquieu [1748]. Both Toynbee [1934, Vol. 1] and Mar-shall [1890, p. 195] similarly emphasized the importance of cli-mate, both on work effort and productivity. One of the pioneers ofdevelopment economics, Myrdal [1968], also placed considerableemphasis on the effect of geography on agricultural productivity.He argued: “serious study of the problems of underdevelop-ment . . . should take into account the climate and its impacts onsoil, vegetation, animals, humans and physical assets—in short,on living conditions in economic development” [Vol. 3, p. 2121].

More recently, Diamond [1997] and Sachs [2000, 2001] haveespoused different versions of the geography view. Diamond, forexample, argues that the timing of the Neolithic revolution hashad a long-lasting effect on economic and social development.Sachs, on the other hand, emphasizes the importance of geogra-phy through its effect on the disease environment, transportcosts, and technology. He writes: “Certain parts of the world aregeographically favored. Geographical advantages might includeaccess to key natural resources, access to the coastline and sea—navigable rivers, proximity to other successful economies, advan-tageous conditions for agriculture, advantageous conditions forhuman health” [2000, p. 30]. Also see Myrdal [1968, Vol. 1, pp.691–695].

This simple version of the geography hypothesis predictspersistence in economic outcomes, since the geographic factorsthat are the rst-order determinants of prosperity are time-in-


variant. The evidence presented so far therefore weighs againstthe simple geography hypothesis: whatever factors are importantin making former colonies rich today are very different from thosecontributing to prosperity in 1500.

IV.B. The Sophisticated Geography Hypotheses

The reversal in relative incomes does not necessarily reject amore sophisticated geography hypothesis, however. Certain geo-graphic characteristics that were not useful, or that were evenharmful, for successful economic performance in 1500 may turnout to be benecial later on. In this subsection we briey discussa number of sophisticated geography hypotheses emphasizing theimportance of such time-varying effects of geography.13

The rst is the “temperate drift hypothesis,” emphasizing thetemperate (or away from the equator) shift in the center of eco-nomic gravity over time. According to this view, geography be-comes important when it interacts with the presence of certaintechnologies. For example, one can argue that tropical areasprovided the best environment for early civilizations—after all,humans evolved in the tropics, and the required calorie intake islower in warmer areas. But with the arrival of “appropriate”technologies, temperate areas became more productive. The tech-nologies that were crucial for progress in temperate areas includethe heavy plow, systems of crop rotation, domesticated animalssuch as cattle and sheep, and some of the high productivityEuropean crops, including wheat and barley. Despite the key roleof these technologies for temperate areas, they have had muchless of an effect on tropical zones [Lewis 1978]. Sachs [2001, p. 12]also implies this view in his recent paper when he adapts Dia-mond’s argument about the geography of technological diffusion:“Since technologies in the critical areas of agriculture, health, andrelated areas could diffuse within ecological zones, but not acrossecological zones, economic development spread through the tem-

13. Put differently, in the simplegeography hypothesis,geography has a main effecton economic performance, which can be expressed as Yit 5 0 1 1 z Gi 1 t 1 it,where Yit is a measure of economic performance in country i at time t, Gi is a measureof geographic characteristics, t is a time effect, and it measures other country-time-specic factors. In contrast, in the sophisticated geography view, the relationshipbetween income and geography would be Yit 5 0 1 1 z Git 1 2 z Tt z Git 1 t 1 it,where Tt is a time-varying characteristic of the world as a whole or of the state oftechnology. According to this view, the major role that geography plays in history isnot through 1, but through 2.


perate zones but not through the tropical regions” (italics in theoriginal; also see Myrdal [1968], Ch. 14).

The evidence is not favorable to the view that the reversalreects the emergence of agricultural technologies favorable totemperate areas, however. First, the regressions in Tables III, IV,and V show little evidence that the reversal was related to geo-graphic characteristics. Second, the temperate drift hypothesissuggests that the reversal should be associated with the spread ofEuropean agricultural technologies. Yet in practice, while Euro-pean agricultural technology spread to the colonies between thesixteenth and eighteenth centuries (e.g., McCusker and Menard[1985], Ch. 3 for North America), the reversal in relative incomesis largely a late eighteenth- and early nineteenth-century, andindustry-based phenomenon.

In light of the result that the reversal is related to industri-alization, another sophisticated geography hypothesis would bethat certain geographic characteristics facilitate or enable indus-trialization. First, one can imagine that there is more room forspecialization in industry, but such specialization requires trade.If countries differ according to their transport costs, it might bethose with low transport costs that take off during the age ofindustry. This argument is not entirely convincing, however,again because there is little evidence that the reversal was re-lated to geographic characteristics (see Tables III, IV, and V).Moreover, many of the previously prosperous colonies that failedto industrialize include islands such as the Caribbean, or coun-tries with natural ports such as those in Central America, India,or Indonesia. Moreover, transport costs appear to have beenrelatively low in some of the areas that failed to industrialize(e.g., Pomeranz [2000], Appendix A).

Second, countries may lack certain resource endowments,most notably coal, which may have been necessary for industri-alization (e.g., Pomeranz [2000] and Wrigley [1988]). But coal isone of the world’s most common resources, with proven reservesin 100 countries and production in over 50 countries [World CoalInstitute 2000], and our results in Table III and V offer littleevidence that either coal or the absence of any other resource wasresponsible for the reversal. So there appears to be little supportfor these types of sophisticated geography hypotheses either.14

14. Two other related hypotheses are worth mentioning. First, it could beargued that people work less hard in warmer climates and that this matters more


IV.C. The Institutions Hypothesis

According to the institutions hypothesis, societies with asocial organization that provides encouragement for investmentwill prosper. Locke [1980], Smith [1778], and Hayek [1960],among many others, emphasized the importance of propertyrights for the success of nations. More recently, economists andhistorians have emphasized the importance of institutions thatguarantee property rights. For example, Douglass North startshis 1990 book by stating [p. 3]: “That institutions affect theperformance of economies is hardly controversial,” and identieseffective protection of property rights as important for the orga-nization of society (see also North and Thomas [1973] and Olson[2000]).

In this context we take a good organization of society tocorrespond to a cluster of (political, economic, and social) institu-tions ensuring that a broad cross section of society has effectiveproperty rights. We refer to this cluster as institutions of privateproperty, and contrast them with extractive institutions, wherethe majority of the population faces a high risk of expropriationand holdup by the government, the ruling elite, or other agents.Two requirements are implicit in this denition of institutions ofprivate property. First, institutions should provide secure prop-erty rights, so that those with productive opportunities expect toreceive returns from their investments, and are encouraged toundertake such investments. The second requirement is embed-ded in the emphasis on “a broad cross section of the society.” Asociety in which a very small fraction of the population, forexample, a class of landowners, holds all the wealth and politicalpower may not be the ideal environment for investment, even if

for industry than for agriculture, thus explaining the reversal. However, there isno evidence either for the hypothesis that work effort matters more for industryor for the assertion that human energy output depends systematically on tempera-ture (see, e.g., Collins and Roberts [1988]). Moreover, the available evidence onhours worked indicates that people work harder in poorer/warmer countries (e.g.,ILO [1995, pp. 36–37]), though of course these high working hours could reectother factors.

Second, it can be argued that different paths of devel

Documents

Reversal of Fortune: Geography and Institutions in the ...tion and Marginalization” in Bergen, The Canadian Institute of Advanced Re-search, Brown University, the University of Chicago,