Was the Wealth of Nations Determined in 1000 BC

Document technical information

Format pdf
Size 722.7 kB
First found May 22, 2018

Document content analysis

Category Also themed
Language
English
Type
not defined
Concepts
no text concepts found

Persons

Peter N. Peregrine
Peter N. Peregrine

wikipedia, lookup

Organizations

Places

Transcript

Was the Wealth of Nations
Determined in 1000 B.C.?*
Diego Comin†
William Easterly ‡
Erick Gongჶ
June 2007
Abstract
We assemble a dataset on technology adoption in 1000
B.C., 0 A.D., and 1500 A.D. for the predecessors to
today’s nation states. We find that this very old history of
technology adoption is surprisingly significant for today’s
national development outcomes. Although our strongest
results are for 1500 A.D., we find that even technology as
old as 1000 BC is associated with today’s outcomes in
some plausible specifications.
Keywords: Technology adoption, technology history,
economic development.
JEL codes: O3, N7.
* We are thankful to Tobias Pfutze for comments and research assistance, to Tomoko Wada and Jason Zhan
Shi for their research assistance, for comments, to Philippe Aghion, Michael Clemens, Rafael DiTella, Simon
Gilgricht, Gene Grossman, Pete Klenow, Bob Lucas, Nathan Nunn, Peter Peregrine, Louis Putterman, Romain
Wacziarg, David Weil and seminar/conference participants at Johns Hopkins, Brown, and to the NSF (Grant
# SES-0517910) and the C.V. Starr Center for Applied Economics for their financial support.
† Harvard University and NBER.
‡ New York University and Visiting Fellow, Brookings Institution.
ჶ University of California at Berkeley.
1
1. Motivation
The study of economic development usually emphasizes modern determinants of per capita
income like quality of institutions to support free markets, economic policies chosen by
governments, human capital components such as education and health, or political factors
like violence and instability. Could this discussion be missing an important, much more long
run dimension to economic development? To the extent that history is discussed at all in
economic development, it is usually either the divergence associated with the industrial
revolution or the effects of the colonial regimes. 1 Is it possible that precolonial, preindustrial
history also matters significantly for today’s national economic outcomes?
This paper assembles a new dataset on the history of technology over 2500 years of history
prior to the era of colonization and extensive European contacts. It finds that there were
important technological differences between the predecessors to today’s modern nations as
long ago as 1000 BC, and that these differences persisted to 0 AD and to 1500 AD (which
will be the three data points in our dataset). These precolonial, preindustrial differences have
striking predictive power for the pattern of per capita incomes across nations that we
observe today. Although our strongest results are for the detailed technology dataset we
assemble for 1500 AD, we also find surprisingly significant effects under plausible
conditions for measures of technological sophistication going back to 1000 BC. Moreover,
technological history is correlated not only with per capita income today but also population
size and thus total GDP (not surprisingly, perhaps, since greater technological productivity
could either support a larger population, or a higher income for the same size population, or
both). We find these results largely continue to hold when we include continent dummies or
geographic controls.
We do not have space in this paper to explore WHY technology in 1000 BC or 1500 AD still
predicts outcomes today, a burning question on which we hope to gain insight from further
research. A very simple explanation is that technological experience has an important effect
on the ability to adopt the new technologies that have come along since the industrial
revolution, but many other explanations are consistent with our results. An alternative is that
technology is reflecting some very long run determinant of development, of which many
have already been explored such as heritable culture (Guiso et al. 2006, Fernandez 2007,
Spolaore and Wacziarg 2006), religious beliefs (Barro and McCleary 2006, 2003), ethnic
fractionalization (Easterly and Levine 1997), intellectual traditions (Mokyr 2005), and ancient
history of statehood (Bockette et al. 2002). 2 The recent emphasis on institutions also may be
consistent with our results (e.g. Acemoglu, Johnson, and Robinson 2002), to the extent that
institutions have very long run determinants. We will speculate further in the conclusion and
1 A notable, honorable, and famous exception is Jared Diamond (1997) Guns, Germs, and Steel, however, this
work did not systematically test the effect of ancient technologies on modern incomes as we will do here.
Perhaps for that reason, the Diamond work did not change much the tendency of development economics to
focus on the modern period or at most the colonial period.
2
This list is far from exhaustive, just illustrative.
2
suggest some avenues for exploration that we are pursuing in subsequent work to this
paper. 3
We are certainly aware that an attempt to collect technology data starting 3000 years ago and
reach serious conclusions is audacious, if not crazy. We will certainly acknowledge the huge
caveats inherent in such an exercise as we go along. We still think it worth doing because of
the increased interest in the literature as to what very long run tendencies can tell us about
the nature and history of economic development.
Another set of examples of such recent interest in the literature are several well known
theories of very long run development. Kremer 1993 has a dynamic story for population
(since 1 million BC!) in which better technology makes possible a larger population (a la
Malthus), and a larger population yields more inventors to make further technological
advances. The idea that larger populations cause better technology is a venerable one
associated with such economists as Simon Kuznets, Esther Boserup, and Julian Simon.
Boserup argued that population pressure induces innovation in a “necessity is the mother of
invention” type argument. Kuznets and Simon emphasized that more people means more
creators of (non-rival) ideas, which means better technology.. Galor and Weil 2000 (see also
Galor 2005) have these features in a story of very long run development with the critical
added feature that advances in technology raise the rate of return to human capital, which
causes the dynamic process to eventually switch over from extensive growth (output and
population growth at the same rate) to intensive growth (per capita income growth). Jones
(2005) emphasizes even more the non-rival nature of technological ideas, which inevitably
generates increasing returns to scale (also featuring the feedback loop between population
and ideas). If societies evolve in isolation through many eons, those who started out ahead
would be even further ahead in both population and income today.
Although we do not in this paper confirm any one particular long-run theory or mechanism,
our results can be seen as a vindication for such long run theorizing about development (as
well as for the empirical work stressing long run factors mentioned above) –at the very least
as a complement rather than necessarily a substitute for the traditional emphasis on the last
few decades.
2. Description of technology data set
The datasets presented in this paper measure the cross-country level of technology adoption
for over 100 current countries in three historical periods: 1000 B.C., 0 A.D. and the precolonial period around 1500 A.D. 4 Each dataset acts as a “snap shot” in time, capturing the
levels of technology adoption by country throughout the world. In each time period, we
This is much debated in the economic history literature. Mokyr (1990, p. 169) stresses the importance of
technology for growth but argues that technological experience has limited importance for new technology
adoption: "It is misleading to think that nothing leads to technological progress like technological progress.”
Rosenberg and Birdzell (1987) also minimize the role of previous technological experience for explaining “how
the West grew rich.” Greene (2000) argues that, in the West, Greco-Roman dynamism was part of a long
continuum from the European Iron Age to medieval technological progress and the industrial revolution.
4 The 1500 A.D. dataset measures a country’s level of technology adoption between 1500 A.D. to 1600 A.D.
Also, technically speaking, there is no year 0 AD, as the calendar moves from 1 BC to 1 AD. We use the
terminology anyway since people understand the concept of year 0 more readily than 1 BC or 1 AD.
3
3
determine a country’s level of technology adoption in five distinct sectors: communications,
agriculture, military, industry, and transportation. By aggregating these values, we determine
a country’s overall level of technology adoption.
Technology adoption is measured on the extensive margin by documenting whether a
country uses a particular technology at all, not how intensively it is used. For example, in the
dataset for 1000 B.C., we consider two transportation technologies: pack animals and
vehicles. A country’s level of technology adoption in transportation is then determined by
whether vehicles and/or draft animals were used in the country at the time. The technologies
that we examine change between the ancient period (1000 B.C. and 0 A.D.) to the early
modern period (1500 A.D.) to reflect the evolution of the technology frontier.
Our focus on the extensive margin of technology adoption is motivated by data availability
constraints. It is much easier to document whether a technology is being used in a country
(the extensive margin) rather then measuring the degree of its adoption (the intensive
margin). It is well documented that the Chinese were using iron for tools by 0 A.D; what is
more difficult to assess is the share of tools constructed from iron at the time.
Since our main objective is to analyze the effects that historic technology adoption has on
the current state of economic development, our datasets are partitioned using modern day
nation states. We use the maps from the CIA’s The World Factbook (2006) to put into
concordance the borders of present day nations with the cultures and civilizations in 1000
B.C., 0 A.D. and 1500 A.D. For example, the technologies used by the Aztecs and their
predecessors during pre-colonial times are coded as the ones used by Mexico in 1500 A.D.
In cases where a country had multiple cultures within its borders during a certain time
period, we take the culture with the highest level of technology adoption to represent that
country. This technique is justified since we are measuring the extensive margin of
technology adoption in a country. For example, in 1000 B.C. there were multiple cultures
residing within Canada’s modern day borders. The Initial Shield Woodland was the most
technologically sophisticated of these cultures and we therefore use its level of technology
adoption to represent Canada in 1000 B.C.
The use of the most advanced culture within a territory for a country’s level of technology
could induce a mechanical correlation between technology and country size (as measured
either by population or land area). The larger the size, the more cultures are being sampled,
which makes the maximum of all cultures higher. For population, this “mechanical” effect is
really the Kuznets-Simon effect of population on technology mentioned in the introduction,
if the most advanced technologies do indeed disseminate within the borders of what is today
measured as a country. We will test for this effect in our empirics. For land area, this also
could reflect a real economic phenomenon for the same reasons, but it would induce reverse
causality between land area and technology. We will examine some simple tests as to whether
this affects our results in the empirical section.
Our datasets are primarily influenced by the work that ethnologists such as George Murdock
and others have done on cross-cultural analysis (Murdock 1967; Carneiro 1970; Tuden and
Marshall 1972; Barry and Paxson 1971). Murdock and others were interested in compiling
4
data on multiple cultures and comparing their traits using analytical methods. 5 A work that
exemplifies this is “The Measurement of Cultural Complexity” (Murdock and Provost 1973).
In that paper, 186 cultures are ranked by their level of cultural complexity. Cultural
complexity was measured using ten variables; these variables included the type of
transportation a culture uses, the level of political integration and urbanization of a culture,
and the degree of technological specialization. Using these rankings, one can conclude that
the Roman Empire was culturally more complex than the Masai of East Africa (Murdock &
Provost 1973: 304).
Since our interests lie in technology adoption within a specific time period, the ethnographic
data described above hold little value for our analysis. Therefore, we adapt the methodology
used in the cross-cultural analysis work to develop our own technology adoption datasets.
Murdock & Morrow (1970) in their work “Subsistence Economy and Supportive Practices”,
provide a detailed description of the methodology that is commonly used to code a crosscultural dataset (Carneiro 1970; Tuden and Marshall 1972; Barry and Paxson 1971; Murdock
and Wilson 1972). In their work, Murdock and Morrow use over 400 sources to evaluate
180 cultures. A team of researchers survey multiple sources for each culture, take detailed
notes in the form of direct quotations, record page numbers of references, and then code
and rank each culture. Inference is used by all of the authors to assist in their coding. In
Carneiro’s appendix to his dataset, he notes (1973: 854), “the presence of the trait, while not
directly observable, may nevertheless be inferred from the presence of certain other traits
which are themselves directly observable.” All of our technology adoption datasets are
coded following this described methodology.
The datasets for 1000 B.C. and 0 A.D. are derived from the “Atlas of Cultural Evolution”
(Peregrine 2003), while we coded the dataset for 1500 A.D. in its entirety. We include a
detailed discussion about each dataset in the following sections.
2.1 Technology Datasets for 1000 B.C. and 0 A.D.
The datasets for 1000 B.C. and 0 A.D. measure the level of technology adoption for
agriculture, transportation, communications, writing, and military on 113 and 135 countries
respectively. In each sector, we examine the same technologies for the two periods. The
datasets for 1000 B.C. and 0 A.D. are based on Peter Peregrine’s (2003) “Atlas of Cultural
Evolution” 6 (henceforward abbreviated as “ACE”). In this work, Peregrine evaluates the
traits (i.e. writing and records, agriculture, transportation, urbanization) of 289 prehistoric
cultures that existed before 1000 A.D. following closely the same methodology as Murdock
& Provost (1973).
The source for the coding of the “ACE” dataset is the Encyclopedia of Prehistory (Peregrine &
Ember 2001a), which is a nine volume work that documents over 250 prehistoric cultures.
The Encyclopedia of Prehistory was compiled from contributions of over 200 authors and covers
5 See the Human Relations Area Files at Yale University for an extensive collection of source material for over
150 cultures.
6 Peregrine (2003) uses BP (Before Present) as the time variable when coding his datasets. We convert the BP
time periods to either B.C. or A.D. Peregrine’s 3000 BP dataset is used for our 1000 B.C. dataset and
Peregrine’s 2000 BP dataset is used for our 0 A.D. dataset.
5
every geographic region of the world (Peregrine & Ember 2001b:3). The Encyclopedia of
Prehistory contains a profile of each prehistoric culture and summarizes the culture’s
environment, settlements, economy, and social political organization. Using the information
from each profile, Peregrine codes the traits of each culture to construct the “ACE” dataset.
It is important to note that the “ACE” limits its survey to prehistoric cultures; prehistory
refers to the time period that precedes written records (Rouse 1972: 3). Once a culture
introduces written records, it is considered part of the historic period and excluded from the
“ACE.” Since written records were introduced at different times throughout the world,
cultures have varying dates on when they entered the historic period. For example, China,
Greece, and Mesopotamia had written records during the first millennium B.C. (Rouse 1972:
8) and are coded as historic regions in the “ACE” (Peregrine 2003). Since most of the world
in both 1000 B.C. and 0 A.D. is prehistoric, the “ACE” provides data that covers most of
the world. We then make inferences on the historic regions of the world at 1000 B.C. and 0
A.D. to complete our datasets.
The “ACE” provides us with data documenting the cultural traits of prehistoric societies;
our task was to convert this data in order to measure each country’s level of technology
adoption. The “ACE” dataset contains four variables of particular interest: “Writing and
Records,” “Agriculture”, “Technological Specialization”, and “Land Transportation.” We
use these four variables to code the adoption of the technologies in communications,
agriculture, industry, and transportation. Table 1 documents the concordance between the
“ACE” and our technology adoption datasets.
Each of the variables in the “ACE” dataset takes on one of three values as shown in the first
column of Table 1. For example, the variable “technology specialization,” can take on one
of three values: a “3” indicates that metalwork is done by a culture; a “2” indicates that
pottery is produced by a culture, and a “1” signifies an absence of both metalworking and
pottery. We take these values and convert them to signify the presence or absence of a
technology. In our technology adoption dataset, the presence of a technology was awarded a
“1” while the absence was awarded a “0”.
6
Table 1: Coding Concordance Between “ACE” Dataset and the Technology Adoption Dataset
“ACE” Dataset
Technology Dataset for 1000 B.C. & 0 A.D.
( 0 = indicates absence of technology,
1 = presence of technology)
Writing & Records
1 = None
2 = Mnemonic or nonwritten records
3 = True Writing
Communication
Technological Specialization
1 = None
2 = Pottery
3 = Metalwork (alloys, forging, casting)
Industry
Land Transport
1 = Human Only
2 = Pack or draft animals
3 = Vehicles
Transportation
Agriculture
1 = None
2 = 10% or more, but secondary
3 = Primary
Agriculture
0
1
2
0,1
0,1
0,1
0,1
0,1
0,1
Technology adoption in the agriculture sector is measured indirectly, as the “ACE” dataset
did not code the actual technologies being used. We infer that the greater the role that
agriculture plays in a culture’s subsistence the more likely that advanced agricultural
technologies were employed. The appendix contains a more detailed discussion on how the
agriculture sector is coded.
An example of how we code a country in 1000 B.C. and 0 A.D will best illustrate our
methodology.
Korea was inhabited by the Mumun peoples in 1000 B.C. The Mumuns had no tradition of
either writing or non-written records. The Mumuns however did rely on agriculture as its
primary form of subsistence and used pack animals for transportation. In addition the
Mumuns produced metalwork and used bronze for tools (Rhee 2001). The coding for the
Mumun entry in the “ACE” dataset (Peregrine 2003) therefore is:
Writing and Records = 1
Technology Specialization = 3
Land Transportation = 2
Agriculture = 3
Based on this data, we code Korea in 1000 B.C. as:
7
Communication: Mnemonic or nonwritten records = 0; True Writing = 0
Industry: Pottery = 1; Metalwork = 1
Transportation: Pack or draft animals = 1; Vehicles = 0
Agriculture: 10% or more, but secondary = 1; Primary = 1
We aggregate the technology adoption measures at the sector level by adding all the
individual technology measures in the sector and dividing the sum by the maximum possible
adoption level in the sector. In this way, the sectoral adoption level belongs to the interval
[0,1]. The overall adoption level in each country and time period is the average of the
adoption level across sectors. Obviously, the overall adoption level also belongs to the
interval [0,1].
The adoption levels in the four sectors just reported in Korea in 1000 B.C. are the following:
Communications = 0
Industry = 1
Transportation = 0.5
Agriculture = 1
Coding for the historic regions of the world in 1000 B.C. and 0 A.D. relied on a combination
of inference and additional documentation. Cultures with written records were the most
technologically sophisticated at the time. A survey of the historic regions during these
periods confirms this assumption. In 1000 B.C., the historic regions include China, Egypt,
Greece, and Mesopotamia, while in 0 A.D. the historic regions expand to encompass
Western Europe and Persia. All of these regions had advanced civilizations that were highly
innovative relative to the rest of the world. For example, by 1000 B.C., Egypt, China,
Greece, and Mesopotamia had growing city populations which relied on high productivity
agriculture (Scarre 1988:122,144; O’Brien 1999:30,36). Wheeled chariots were invented in
Mesopotamia around 3000 B.C., and were used in Egypt, Greece, and China by 1000 B.C.(
Encyclopedia Britannica 2006h). Jewelry and decorative ornaments constructed out of gold
and silver are also evident in these cultures (Scarre 1988; O’Brien 1999). We therefore code
the historic regions in our dataset as having the highest level of technology adoption in
agriculture, communications, transportation, and industry.
The “ACE” did not contain any variables that correspond to technologies used for military
purposes. To assess a country’s level of technology adoption for the military we use the
“ACE” dataset to determine which metals were available for each culture. Metallurgy is
integral for the development of more advanced weapons (Macksey 1993:216; Scarre 1988;
Collis 1997:29). The progression from stone to bronze and finally iron corresponded to a
progression of more powerful weapons; stone weapons were replaced by bronze swords and
daggers; iron weapons were considerably stronger than their bronze predecessors (Hogg
1968:19-22). The “ACE” dataset defined many cultures by the type of metals they were
using for tools. Neolithic cultures are coded as having stone weapons, while Bronze and
Iron Age cultures were coded as having bronze and iron weapons respectively. Prehistoric
cultures not adequately described in the “ACE” dataset are coded through inference. Since
the people of the New World did not use bronze until near the time of European contact, all
8
countries in North and South America are coded as not having bronze or iron weapons in
1000 B.C. and 0 A.D (Diamond 1997:259; Kipfer 2000).
The historic regions of 1000 B.C. (Mesopotamia, Northern Africa, Greece, China) did not all
use iron for weapons. We therefore differentiate iron producing regions from those that did
not use the metal. Asia Minor and Mesopotamia are coded as using iron since the Hittites
became major producers of iron in the 3rd millennium B.C. (Collis 1997:32; Kipfer
2000:257). Greece also had iron objects by 1200 B.C and is coded accordingly. The two
most prominent historic regions not possessing iron technology by 1000 B.C. are Egypt and
China. Both regions first used iron in the 6th century B.C. (Wager 1993; Lucas 1934:198).
Egypt and China however both used bronze well before 1000 B.C. (Kerr & Wood 2004:7;
Erman 1971: 461) and our dataset in 1000 B.C. reflects this.
The coding of historic regions in 0 A.D. proved much easier as iron technology had diffused
throughout Europe, the Middle East, North Africa, and China during the 1st millennium
B.C. (Kipfer 2000:258). We therefore code all historic regions as using iron weapons in the
0 A.D. dataset.
2.2 Technology Dataset for 1500 A.D.
The technology dataset for 1500 A.D. encompasses 113 countries and evaluates the level of
technology adoption across the same five sectors (agriculture, transportation, military,
industry, and communications) as the previous datasets. The technology adoption dataset
for 1500 A.D. differs from the prehistoric datasets in that it is not based on an existing work.
While the datasets for 1000 B.C. and 0 A.D. relied on the “ACE” (Peregrine 2003) for a
preponderance of data, the dataset for 1500 A.D. is coded using over 170 source materials.
Our technology measures outside Europe are estimated before European colonization. It is
important to stress, therefore, that our technology measures in 1500 A.D. do not incorporate
the technology transferred by Europeans to the rest of the world after European exploration
began around 1500.
Obviously, there is a larger number of sources covering the technology adoption patterns in
1500 A.D. than in 1000 B.C. or 0 A.D. This allows us to collect adoption data for 20
technologies in the four sectors other than agriculture vs. the eight technologies covered in
the data sets for 1000 B.C. and 0 A.D. As a result, our estimate of the level of technology
adoption in 1500 A.D. is likely to be more precise than for the earlier periods. Table 2
presents the various technologies measured in 1500 A.D.
Our technology datasets for 1500 A.D. involve surveying multiple sources (atlases, history
books, journal articles) and determining whether a technology was used in a country.
However, as with our datasets for 1000 B.C. and 0 A.D, the dataset for 1500 A.D. does
include a proxy for the level of agricultural technology adoption.
We must of course stress that there are several possible weak links in the chain to go from
the source material on old cultures to our dataset corresponding to today’s nation states –
such as the possibly tenuous link between ancient cultures and the territories of modern day
nation states, and the possible errors of commission and omission on whether technologies
9
are present given incomplete records, just to mention two. There also is likely to be selection
bias in that more technologically advanced cultures are likely to leave better records. 7
Despite these caveats, there are also important reasons to believe in the quality of our data.
First, as we describe below, it builds on the methodological contributions of the existing
literature. Second, it is based on a very extensive documentation described in detail in a
separate appendix. 8 Third, it is much easier to code extensive than intensive measures of
technology adoption for pre-colonial periods. The former is feasible, after a significant effort
such as ours. The latter is just impossible. Third, as we shall see below, our technology
adoption measures for 1500 A.D. are highly correlated to the technology adoption measures
for 1000 B.C. and 0 A.D. from ACE. We find this supportive of the quality of our data given
that they were constructed in a completely independent way. Finally, as we shall show below,
the overall technology adoption measure is highly correlated to contemporaneous measures
of the development of societies such as the urbanization rate. These arguments lead us to
persist nevertheless in making the best of the always shaky nature of very old data in order to
see whether our measures have any signal along with the noise.
Of course, in the presence of this bias, the resulting technology adoption measure would be highly correlated
across with actual technology adoption.
8 And in even more detail in a second appendix available from the authors that documents the information
used to code each technology for each country.
7
10
Table 2: Variables in the 1500 A.D. dataset
Variable
Description
Values
Military
Standing Army
Cavalry
Firearms
Muskets
Field Artillery
Warfare capable ships
Heavy Naval Guns
Ships (+180 guns), +1500 ton
deadweight
An organization of professional soldiers.
The use of soldiers mounted on horseback.
Gunpowder based weapons
The successor to the harquebus (the common firearm of European armies)
was larger and a muzzle-loading firearm.
Large guns that required a team of soldiers to operate. It had a larger caliber
and greater range than small arms weapons.
Ships that were used in battle are considered "warfare" capable.
Ships required significant advances in hull technology before they were
capable of carrying heavy guns.
0,1
0,1
0,1
Large warships that only state navies had the capability of building.
0,1
0,1
0,1
0,1
0,1
Agriculture
Hunting & Gathering
Pastoralism
Hand Cultivation
Plough Cultivation
The primary form of subsistence.
The primary form of subsistence.
The primary form of subsistence.
The primary form of subsistence.
0
1
2
3
Transportation
Ships Capable of Crossing the
Atlantic Ocean
Ships Capable of Crossing the
Pacific Ocean
Ships Capable of Reaching the
Indian Ocean
Wheel
Magnetic Compass
Horse powered vehicles
Any ship that had successfully crossed the Atlantic Ocean.
0,1
Any ship that had successfully crossed the Pacific Ocean.
0,1
Any ship that had reached the Indian Ocean from either Europe or the Far
East.
The use of the wheel for transportation purposes. The most common use
was for carts.
The use of the compass for navigation.
The use of horses for transportation.
0,1
0,1
0,1
0,1
Communications
Movable Block Printing
The use of movable block printing.
0,1
Woodblock or block printing
The use of woodblock printing.
0,1
Books
Paper
The use of books.
The use of paper.
0,1
0,1
Industry
Steel
Iron
The presence of steel in a civilization.
The presence of iron in a civilization.
0,1
0,1
The methodology for coding 1500 A.D. datasets follow the works mentioned previously by
Murdock and Morrow (1970), Murdock and Provost (1973), Peregrine (2003), and Carneiro
(1970). We rely on two principal inference techniques while coding the dataset: 1)
technological continuity (Basalla 1988) and 2) temporal extrapolation (Murdock & Morrow
1970: 314). Technological continuity is the idea that innovations are a result of previous
11
antecedents; innovations typically do not spontaneously arise without preexisting
technologies. 9 Technological continuity allows us to infer that countries with advanced
technologies also have more primitive ones. The use of military technology in 1500 A.D.
illustrates this technique. Large warships with over 180 guns on deck were considered the
pinnacle of military technology in 1500 A.D. (Black 1996). We find that many countries
with large warships also had advanced land weapons such as muskets and field artillery. It is
not unreasonable to assume a country must first acquire land-based arms technology before
producing ships with large naval guns. Therefore, in cases such as Portugal and Germany,
where large warships were present we infer that these countries also had advance land
weaponry.
Temporal extrapolation is an inference technique we use in the 1500 A.D. dataset. This
technique assumes that a technology maintains persistency over time. It is not unreasonable
to assume that a technology that is adopted by a country at a certain point in time will
continue to be in use in that country fifty to one hundred years later. Temporal
extrapolation allows us to code countries where documentation for a specific technology is
not available for 1500 A.D. When a technology’s presence cannot be documented in a
country in 1500 A.D., we look at preceding time periods. If a technology is used by a
country before 1500 A.D. we infer that it was used during 1500 A.D. as well. An example of
this is our coding of transportation technology in Turkey. We are able to document that the
magnetic compass was in use in the Ottoman Empire by 1450. Using temporal
extrapolation, we code Turkey as having the magnetic compass in the 1500 A.D. dataset.
Clearly there are limits to this technique; the longer the extrapolation period, the less
confidence we have in inferring if a technology was still being used. By consulting a very
large number of sources, we have been able to code the 1500 A.D. data set based on
information from the XVth century. 10
Country concordance for the 1500 A.D. dataset follows the methodology we described in
the introduction. We assume that a technology used by a civilization diffuses throughout the
regions it controlled. An example is the Ottoman Empire. The Ottomans controlled a wide
swath of territory during 1500 A.D., including but not limited to modern day Egypt, Libya,
Greece, and Iraq. Technologies used by the Ottoman Empire were assumed to have
diffused from Turkey to all the countries we cited as being under Ottoman control.
The following passages briefly describe the process of determining levels of technology
adoption for the military, agriculture, communications, transportation, and industrial sectors.
Further discussion on our coding methodology is in the appendix.
Military technology in 1500 A.D.
We measure a country’s level of military technology adoption by documenting the presence
of land and sea based weapons in a country. In total, we document the presence of eight
variables for each country.
9 See Basalla (1988:30-57) for a number of case studies documenting technological continuity or technological
evolution.
10
In many of the cases where we have used temporal extrapolation, we have also been able to document the
presence of the technology during the XVIth century.
12
The variables that represent technology for land weaponry include: the presence of a
standing army, the use of firearms, muskets, cavalry, and field artillery. Sea based weapons
are measured by the presence of naval warships and their armaments. The types of sea
based weapons we document are: warfare capable ships, ships with heavy naval guns, and
heavy warships that have over 180 guns and weigh over 1500 tons.
Agricultural Technology in 1500 A.D.
As with the datasets for 1000 B.C. and 0 A.D., we use a country’s primary form of
subsistence (hunting and gathering, pastoralism, agriculture) as a proxy for technology
adoption in 1500 A.D. This measure is rationalized by the fact that the adoption of some
important agricultural technologies is necessary for a country to move from a hunter and
gathering society to an agrarian one. In addition to this indirect measure, for those countries
whose primary form of subsistence was agriculture, we also measure the adoption of plough
cultivation.
Transportation Technology in 1500 A.D.
A country’s level of transportation technology adoption is measured by the forms of naval
and land based transportation. We examine six variables, four of which measure a country’s
naval technology, while the remaining two measure land-based technology. Land-based
technologies include the wheel and animals used for transportation. Naval-based
transportation technology adoption is measured by whether a country’s seamen used
magnetic compasses for navigation and the distances that a country’s exploration fleet sailed.
Communications Technology in 1500 A.D.
We measure a country’s adoption of communications technologies by examining the
technologies used to disseminate written information. We directly measure these
technologies by documenting in a country the presence of the following items: paper, books,
woodblock printing tools, and movable type printing presses.
The technologies we document represent the stages that many countries went through as
they developed their communications technology. By 1500 A.D., paper and books had
diffused throughout most of Asia and Europe. These technologies were also adopted in
parts of North Africa. More advanced technological countries adopted means of more rapid
reproduction of written communication, such as the moveable type press.
Industrial Technology in 1500 A.D.
Industrial technology measures a country’s adoption of metallurgical technology. We
measure a country’s extensive margin of technology adoption by documenting the presence
of iron and steel production in the country.
By 1500 A.D., iron and steel were being produced in Europe, the Middle East, and East
Asia. While iron was being used for tools throughout Africa in 1500 A.D., steel was not
present in Sub-Saharan Africa before contact with the Europeans. Also, the technology used
to produce iron and steel was not present in the New World until after European contact.
13
3. Descriptive statistics
We start the data analysis by presenting in Table 3 some descriptive statistics for the overall
technology adoption level in 1000BC, 0 A.D. and 1500 A.D. The descriptive statistics for the
technology adoption measures at the sector level are relegated to Table A2 in the appendix.
Table 3: Descriptive statistics of Overall Technology Adoption
Period
Number
Obs.
Average
Std. Dev.
Min
Max
1000BC
113
0.45
0.28
0
1
0
135
0.73
0.28
0
1
1500AD
123
0.46
0.32
0
1
The increase in the cross-country average of the overall technology adoption level between
1000 B.C. and 0 indicates the diffusion of the technologies described in the ACE. Recall that
the technology adoption data set for 1500 A.D. contains different technologies than the first
two periods. The decline in the average level of adoption in 1500 A.D. indicates that these
technologies had diffused less than the technologies from ACE in 0 A.D.
An important question that the descriptive statistics can answer is how large is the crosscountry dispersion in technology adoption. The binary nature of our measures of technology
adoption for individual technologies provides two benchmarks to interpret the cross-country
dispersion in technology adoption. 11 First, the maximum range for the average adoption level
across countries is the interval [0,1]; 0 for a country that has not adopted any of the
technologies and 1 for a country that has adopted all the technologies. Second, the
maximum cross-country dispersion in adoption would occur when half of the countries have
adopted all the technologies and the other half has adopted none. In this case the standard
deviation of the average adoption level across countries would be 0.5.
In Table 3 we can observe how the range of the average adoption level across countries was
[0, 1] in all three periods. The fact that these ranges are the maximum possible signals a large
cross-country dispersion in overall technology adoption.
Figures 1 through 3 and Table 4 explore the cross-country variation in the overall technology
adoption level. Table 4 explores the variation across continents in overall technology
adoption. Figures 1 through 3 present a world map with the overall technology adoption
11
The exceptions to this rule are the measures of technology adoption in agriculture.
14
level in each country and historical period. We use four colors to indicate technology
adoption levels between 0 and 0.25, between 0.25 and 0.5, between 0.5 and 0.75 and
between 0.75 and 1. Darker colors represent a higher overall technology adoption level.
Missing values are represented in white.
Table 4: Descriptive statistics of Overall Technology Adoption by Continent
Period
Continent
Number Obs.
Average
Std. Dev.
Min
Max
1000BC
Europe
30
0.66
0.16
0.5
1
Africa
34
0.36
0.31
0
1
Asia
23
0.58
0.25
0.1
1
America
24
0.24
0.12
0
0.4
Oceania
2
0.2
0.14
0.1
0.3
Europe
33
0.88
0.15
0.7
1
Africa
40
0.77
0.2
0.6
1
Asia
34
0.88
0.15
0.6
1
America
25
0.33
0.17
0
0.6
Oceania
3
0.17
0.11
0.1
0.3
Europe
26
0.87
0.074
0.69
1
Africa
39
0.32
0.2
0.1
0.78
Asia
25
0.66
0.19
0.07
0.88
America
24
0.14
0.07
0
0.13
Oceania
9
0.12
0.04
0
0.13
0AD
1500AD
15
In all three periods, Europe and Asia present the highest average levels of overall technology
adoption, while America and Oceania present the lowest.
____________ INSERT FIGURES 1-3 HERE ______________________
A glimpse to the figures suffices to note that there is substantial variance in overall
technology adoption both across and within continents. To make observation more precise,
we decompose the cross-country variation in overall technology adoption between the
variation within continents and the variation across between continents. In 1000BC, about
65 percent of the variance in overall technology adoption is due to variation within
continents and 35 percent due to variation between continents. These proportions are
reversed in 0 A.D. and in 1500 A.D. the share of total variance due to the between continent
component rises to 78 percent.
Table 5 provides a more detailed comparison of the evolution of overall technology
adoption in the most advanced countries. These countries correspond to four civilizations:
Western Europe, China, the Indian civilization and the Middle Eastern peoples. Western
Europe includes Spain, Portugal, Italy, France, United Kingdom, Germany, Belgium and
Netherlands. The Indian civilization includes India, Pakistan and Bangladesh. Finally, the
Middle Eastern civilization includes Saudi Arabia, UAE, Yemen, Oman, Iraq, Iran, Syria,
Lebanon, Jordan, Egypt, Libya, Tunisia, Algeria and Morocco.
Table 5: Average Overall Technology Adoption in Advanced Civilizations
Civilization
1000BC
0 AD
1500 AD
W. Europe
0.65
0.96
0.94
China
0.9
1
0.88
Indian
0.67
0.9
0.7
Arab
0.95
1
0.7
Note: W. Europe includes Spain, Portugal, Italy, France, United Kingdom, Germany,
Belgium and Netherlands. Indian Empire includes India, Pakistan and Bangladesh.
Arab Empire includes Saudi Arabia, UAE, Yemen, Oman, Iraq, Iran, Syria, Lebanon, Jordan,
Egypt, Libya, Tunisia, Algeria and Morocco
In 1000 B.C. the Middle Eastern empires and China have an overall technology adoption
level of 0.95 and 0.9 respectively, while in India and Western Europe the average adoption
level are 0.67 and 0.65 respectively. In 0 A.D. India and Western Europe catch up with
China and the Middle Eastern empires. In 1500 A.D. Western Europe has completed the
16
transition and is the most advanced of the four great empires with an average overall
adoption level of 0.94. China remains ahead of most countries with 0.88. But the Indian and
the Middle Eastern empires have fallen behind. The average overall adoption levels in these
empires are 0.7.
4. Technology history and current development
Without more delay, we turn next to the question that motivates our exploration. Namely,
whether centuries-old, pre-colonial technology history is correlated with development today.
To answer this question, we estimate the following regression
yc = α + βTc + uc
(1)
where yc is the log of PPP adjusted per capita income in 2002 A.D., Tc is the measure of
technology adoption and uc is the error term.
17
Table 6: Technology History and Current Development
Dependent Variable
Log Income per capita in 2002
I
Overall Technology adoption level:
in 1000BC
II
III
IV
0.73
(1.96)
in year 0
V
VI
VII
1.45
(3.05)
0.09
(0.23)
in 1500AD
1.46
(2.83)
1.64
(5.14)
2.96
(8.33)
Major European Involvement
1.83
2.47
(12.08) (10.78)
2.83
(8.18)
3.22
(12.86)
Minor European Involvement
0.16
(1.05)
0.63
(2.72)
0.82
(3.23)
1.43
(5.9)
8.2
8.45
7.75
8.43
(40.5) (30.23) (37.42) (69.64)
7.68
(27.1)
7.21
(17.35)
6.74
(27)
Constant
N
105
124
107
130
105
124
107
R2
0.03
0
0.19
0.08
0.17
0.13
0.5
Note: t-statistics in parenthesis computed using robust standard errors.
Major European Involvement is a dummy that is 1 for the "Neo-Europes": US, Canada, New Zealand and Australia.
Minor European Involvement is a dummy that is 1 for areas of partial European settlement in Latin America, the
Caribbean and southern Africa.
The first three columns of Table 6 report the estimates of regression (1) when Tc is measured
successively by the overall adoption level in 1000 B.C., in 0 and in 1500.A.D. (T-statistics are
in parentheses.) The technology adoption level in 1000 B.C. is positively and significantly
associated with the log of per capita GDP in 2002. Technology adoption in 0 A.D. is not
significantly correlated to current development. The overall technology adoption level in
18
1500 A.D. is positively and significantly associated with current income per capita. This
measure of technology in 1500 A.D. explains 18 percent of the variation in log per capita
GDP in 2002.
In addition to being statistically significant, the effect is quantitatively large. Changing from
the maximum (i.e. 1) to the minimum (i.e. 0) the overall technology adoption level in 1500
A.D. is associated with a reduction in the level of income per capita in 2002 by a factor of 5.
Figure 4 presents the scatter plot between overall technology adoption level in 1500 A.D.
and current development. The positive relationship between these two variables is quite
transparent. It is clearly not driven by outliers. In the bottom left quadrant of the plot we can
see many African countries that had adopted very few of the technologies in our 1500
sample and that are quite poor today. European countries are in the top right corner.
Countries that roughly correspond to ancient empires such as Egypt, Iran, China, India, and
Pakistan were middle-income countries in 2002 and had adopted between 70 and 90 percent
of the technologies in our 1500 A.D. sample. These countries are slightly below the
regression line in the bottom right quadrant of Figure 4. This paper does not address some
well-known puzzles, such as the failure of China to capitalize earlier on its technological
prowess, or the stagnation following the earlier technological prowess of the Islamic empire.
These are very important puzzles that deserve (and have already attracted) their own
literature, but we are concerned here with the global cross-country average relationship
between old technology and modern income, and these counter-examples are not numerous
enough to overturn the average global relationship.
___________ INSERT FIGURE 4 HERE______________
Latin American countries were behind the median country in the overall technology
adoption level in 1500 but today they are middle income countries. This very likely has
something to do with the long period of European settlement in Latin America, even though
the European settlers were generally a minority of the population. Finally, in the top left
corner of Figure 4 we find the Neo-Europes. That is the US, Canada, Australia and New
Zealand. These were among the countries with most primitive technology in 1500 A.D. but
are among the World’s richest countries today. This is very likely due to the large-scale
replacement of the original inhabitants with European settlers.
We would expect that the European settlers in the Spanish and Portuguese colonies and in
the Neo-Europes affected quite dramatically the process of technology transfer (as well as
other factors with which technology may be associated such as human capital accumulation
and institutional development) in these countries during the colonial period. Another place
where there was large scale (albeit still minority) European settlement was southern Africa.
Of course, there could be technology transfer in any colonized nation, but the duration and
intensity of the influence of the settlement processes in southern Africa, Latin America and
the Neo-Europes suggest adding special controls. Further, the difference in the degree to
which Europeans colonizers substituted for the local population justifies the distinction
between the Neo-Europes and Latin America/southern Africa.
19
To formalize this intuition, we use the fraction of European settlers in total population in
1900 from Acemoglu, Johnson and Robinson (2001). 12 This fraction was over 90 percent for
the Neo-Europes, between 15 percent and 65 percent for South Africa, Lesotho and
Swaziland, and most countries in Latin America and the Caribbean, and below 15 percent
for the rest of non-European countries.
Based on this, we create two dummies. The first captures predominant European settlers,
and takes a value of one for the US, Canada, New Zealand and Australia and is zero for the
rest of the countries. The second dummy reflects lesser European settler predominance than
in the neo-Europes, and takes a value of one for the Latin American colonies of Spain and
Portugal (see the appendix for a complete list), South Africa, Lesotho and Swaziland, and is
zero otherwise. This yields the following regression equation:
yc = α + β Tc + Majorc + Minorc + uc
(2)
_______________ INSERT FIGURE 5 HERE_________________
Columns 5 through 7 in Table 6 report the estimates of equation (2) with Tc measured
successively by the overall technology adoption level in 1000B.C., 0 and 1500 A.D.
_______________INSERT FIGURE 6 HERE_____________________
We find that the European settlement dummies have a significant positive relationship with
current per capita income. Further, when including the European settlement dummies, the
correlation between the overall technology adoption and current development increases. In
particular, the effects of the technology adoption levels in 0 on current per capita income
become statistically significant, and the effect of technology in 1000 BC and 1500 A.D.
almost doubles. In other words, once we control for the most obvious historical example of
replacement of the indigenous technology by technologies brought by new settlers,
technology in ancient times becomes an even more significant predictor of per capita income
today.
We acknowledge that there could have been other population migrations that transferred
technology, and our singling out of the international European migration may be ad hoc,
although it seems to us the primacy of European migration over the last 500 years is not
really in doubt. In any case, our results seem to hold for other population movements as
well. 13Also, we continue to find significant correlations in important specifications (such as
12 Similar results are obtained using the share of population from European descent in 1975 from Acemoglu,
Johnson and Robinson (2001) or the fraction of European settlers 100 years after first settlement from Easterly
and Levine (2006).
13 We know from a collaborative exercise with David Weil that our findings hold also when we control more
comprehensively for the international migration flows. Specifically, we use Putterman and Weil (2007)’s matrix
which gives, for each country, the distribution of its current population by its origin. We then pre-multiply the
vector of overall technology in 1500 AD by the origin matrix and find that the origin weighted measure of
technology predicts current per capita income slightly better than the regresors in column 8 of Table 6. We do
not report these results here as Putterman and Weil (2007) have not yet made their data public (nor have we,
waiting for more peer review).
20
those already reported above, and more below) even when the European dummies are
excluded.
_______________INSERT FIGURE 7 HERE_____________________
After including the settlement dummies, an increase in the overall adoption level from 0 to 1
in 1000 B.C. or in 0 A.D. is associated with an increase in income per capita in 2002 by a
factor of 4. A similar increase in the overall adoption level in 1500 A.D. is associated with an
increase in per capita income in 2002 by a factor of 19. This is half of the current difference
in income per capita between the top and bottom 5 percent of the countries in the world.
Similarly, 20 percent of the income difference between Europe and Africa is explained by
Africa’s lag in overall technology adoption in 1000 B.C., 8 percent is explained by the
technology distance in 0 A.D., and 78 percent is explained by Africa’s lag in overall
technology adoption in 1500 A.D. This gives a very different perspective on Africa’s poverty
compared to the usual emphasis on modern governments. It also shifts backward in time the
historical explanations for Africa’s poverty, compared to the usual emphasis of historians on
the slave trade and colonialism. 14
Figures 5 through 7 display the scatter plots of the current income per capita and overall
technology adoption after regressing these variables on the European influence dummies.
These figures confirm the significant association between current development and historical
technology after conditioning on the European influence dummies. Clearly, the strongest
relationship holds between overall technology adoption in 1500 A.D. and current
development.
5. Robustness and Discussion
Next we discuss the robustness and interpretation of the main fact uncovered in the
previous section, that technology history is positively and strongly associated with current
development.
A. Robustness
We start by exploring whether we are identifying the effect of historical technology on
current development through the cross-continent variation of also through the within
continent variation. To answer this question, the first three columns of Table 7 report the
estimates of regression (2) when adding four continent dummies to the control set.
14 There was some slave trade before 1500 A.D. across the Sahara and along the Indian Ocean. However, most
accounts of the negative effects of the slave trade stress the Atlantic slave trade, which only became nontrivial
after 1500 A.D.
21
Table 7: Primitive Technology and Current Development, Robustness
Dependent Variable
Log Income per capita in 2002
I
II
III
Overall Technology adoption level:
in 1000BC
0.2
(0.5)
in year 0
IV
V
VI
0.2
(0.24)
0.64
(1.51)
in 1500AD
VII
1.34
(2.2)
1.73
(7.91)
1.57
(5.22)
0.57
(1.06)
Africa dummy
-0.32
(2.15)
-0.66
(2.47)
-1.12
(3.52)
Asia dummy
0.44
(1.63)
0.39
(1.2)
-0.57
(1.27)
America dummy
0.15
(0.87)
0.11
(0.67)
-0.24
(0.73)
Distance to equator
0.55
(1.15)
1.46
(2.8)
3.9
(8.48)
IX
0.76
(1.59)
0.04
(0.09)
Europe dummy
VIII
2.43
(5.32)
4.14 2.91
(9.02) (4.1)
Tropical dummy
-1.02 -1.14 -0.45
(4.82) (5.81) (1.99)
N
105
124
107
97
114
103
105
124
107
R2
0.58
0.61
0.66
0.54
0.45
0.6
0.35
0.34
0.52
Note: t-statistics in parenthesis computed using robust standard errors.
All regressions include major and minor European involvement dummies
and a constant.
22
We extract two main conclusions from columns 1 through 3. First, much of the effect of
technology history is detected from the cross-continent variation. Adding the continent
dummies eliminates the effect of overall technology adoption in 1000 B.C. on current
development (column 1), and reduces by 60 percent the effect of technology adoption in 0
A.D. (column 2) and in 1500 A.D. (column 3) on current development. Only 1500 AD is
still significant. The flip side of this is that a significant fraction of the effects of technology
adoption history in 0 A.D. and 1500 A.D. on current development is driven by the within
continent variation. In particular, the within continent variation in overall technology
adoption in 1500 A.D. can still account for cross country variation in current income per
capita by a factor of 3.8. We will see below that the association of ancient technology with
modern total GDP and population are more robust to including continent dummies.
Gallup, Sachs and Mellinger (1999) have argued that the latitude is an important determinant
of income per capita, with the tropics at a disadvantage. Hall and Jones (1999), Acemoglu,
Johnson, and Robinson 2002, Easterly and Levine 2003 and Rodrik et al. (2003) argue that
the effect of tropical location is through institutions. Columns 4 through 9 in Table 7 report
the estimates of regression (2) after controlling for the distance to the Equator (columns 4
through 6) or whether the country is tropical (columns 7 through 9). As emphasized by the
previous literature, being far from the Equator tends to be associated with higher levels of
current income per capita. Controlling for the latitude of countries, however, does not
eliminate the strong positive effect of overall technology adoption in 1500 A.D. on current
development. This effect remains statistically significant, though the association of
technology adoption history on 1000 B.C. and in 0 A.D. on current development become
insignificant after controlling for the distance to the Equator or after including the tropical
dummy. Again, we will see next that the association of ancient technology with modern total
GDP and population is more robust to including geographic controls.
Studying whether more advanced technology also made higher population and higher total
GDP feasible as well as higher per capita GDP is natural, given the population-technology
models mentioned in the introduction. To answer this question we estimate the effect of
primitive technology on the log of real GDP (Yc) and in the log of population (Lc), both in
2002, as indicated in regressions (4) and (5).
log(Yc ) = α + β Tc + u c
(4)
log( Lc ) = α + β Tc + u c
(5)
23
Table 8a: Technology History, Current GDP, Population and Arable Land
Dependent Variable
Log GDP 2002
I
II
III
Overall Technology adoption level:
in 1000BC
1.86
(2.68)
in year 0
Log Population
2002
IV
V
1.27
(2.4)
0.93
(1.45)
in 1500AD
VII
VIII
0.97
(2.13)
0.46
(0.73)
1.85
(3.87)
NO
IX
1.6
(2.66)
3.12
(5.72)
Major and Minor european
involvement dummies
VI
Log Arable Land
1.45
(2.54)
NO
NO
N
105
124
107
114
136
118
110
132
114
R2
0.08
0.02
0.25
0.05
0.03
0.12
0.07
0
0.06
Note: t-statistics in parenthesis computed using robust standard errors.
All regressions include a constant.
Table 8a reports the estimates of these specifications for the measures of overall technology
adoption in each of the three periods. In columns 1 through 3 we observe that the measures
of primitive technology in 1000 B.C. and in 1500 A.D. have a very significant positive effect
on current GDP. The effect of technology adoption in 0 A.D. is positive but insignificant.
Columns 4 through 6 show that countries with higher overall levels of historical technology
adoption have higher population today. This is the case for each of the three measures of
primitive technology. Unlike the regressions for per capita income, the coefficient on
technology in 1000 BC for today’s GDP and population is significant even without including
the European settlement dummies, and 1500 AD also continues to be strongly significant
without these dummies.
In columns 7 through 9 of Table 8a we estimate the effect of technology adoption history on
land area by estimating the following regression:
24
log( Ac ) = α + β Tc + u c
(6)
where Ac is the arable land area. Our estimates show that the log of arable land area of
today’s nation states is also related to historical technology in that area. We interpret this as
evidence that countries with more advanced technologies could conquer more land and/or
could control more land more easily.
This could also be another mechanism by which advanced technology led to larger
populations; conversely countries with larger populations, thanks to more advanced
technology, could also conquer or settle more territory. Over the very long period that we
are considering, the size of nations in both area and population is endogenous. Our results
show that technology is one of the determinants of the size of nations. However, since both
land area and population are endogenous and we lack good instruments, we cannot separate
out the relationship between these two different dimensions of size.
Table 8b: Technology History, Current GDP, Population and Arable Land, European Influence
Dummies
Dependent Variable
Log GDP 2002
I
Overall Technology adoption level:
in 1000BC
2.85
(3.63)
in year 0
II
Log Population 2002
III
IV
V
1.74
(2.72)
3.19
(3.87)
in 1500AD
VII
VIII
2.1
(3.04)
1.91
(2.14)
2.66
(4.14)
2.86
(4.69)
YES
IX
2.25
(3.78)
5.2
(11.11)
Major and Minor
european involvement
dummies
VI
Log Arable Land
YES
YES
N
105
124
107
114
136
118
110
132
114
R2
0.21
0.16
0.53
0.09
0.09
0.21
0.21
0.1
0.2
Note: t-statistics in parenthesis computed using robust standard errors.
25
All regressions include a constant.
Table 8b estimates specifications (4) through (6) adding the two European settlement
dummies. This increases the effects (and makes all 3 dates significant) of technology
adoption history on current GDP, on current population and on current arable land area.
Hence, we conclude that historical technology adoption was associated with both a larger
population and a higher average income.
We next check the robustness of this conclusion to controlling for the distance from the
Equator which affected the significance of the ancient technology variables in the per capita
income regressions. Columns 1 through 9 in Table 8c show that controlling for distance to
Equator does not affect the strong positive effect of technology adoption history on current
GDP, on current population, and on current land area. It is interesting to note that, while
distance to Equator is positively and significantly associated with current GDP in the
regressions where technology adoption history is measured at 1000 B.C. and 0 A.D., it is
insignificantly associated with current GDP when technology adoption is measured in 1500
A.D. Similarly, while distance to Equator is insignificantly associated with current population
in the regressions for technology adoption in 1000 B.C. and 0 A.D., it is negatively and
significantly associated to current population when technology adoption is measured in 1500
A.D. We interpret these significant changes in the mechanism by which latitude affects
current income per capita as a signal that the association of latitude and current development
is not invariably causal and direct. In contrast, the association of technology adoption history
with current GDP and population is robust to measuring technology in any of the three
periods.
26
Table 8c: Technology History, Current GDP, Population and Arable Land, Distance from
Equator
Dependent Variable
Log GDP 2002
I
Overall Technology adoption level:
in 1000BC
2.42
(2.68)
in year 0
II
IV
V
VI
2.47
(3.18)
2.33
(2.78)
in 1500AD
Distance from Equator
III
Log Population 2002
Log Arable land
VII
VIII
1.98
(3.07)
2.57
(3.49)
5.43
(7.23)
2.96
3.5
-0.07
(2.56) (4.33) (0.08)
-0.52
(0.6)
IX
1.55
(1.77)
4.66
(5.68)
2.53
(2.18)
-0.46 -3.2
(0.71) (3.06)
1.38
2.1
0.44
(1.56) (2.46) (0.27)
N
97
114
103
105
125
113
104
124
111
R2
0.3
0.28
0.54
0.13
0.11
0.3
0.23
0.15
0.2
Note: t-statistics in parenthesis computed using robust standard errors.
All regressions include major and minor European Influence dummies and a constant.
Table 8d shows the regressions for total GDP, population, and land area when continent
dummies are included. The association of ancient technology with these modern outcomes is
much more robust to including continent dummies than the results with per capita income.
This suggests that the legacy of ancient technology for these other aspects of the “wealth of
nations” is not driven only by differences between continents.
27
Table 8d: Technology History, Current GDP, Population and Arable Land, Continent Dummies
Dependent Variable
Log GDP 2002
I
Overall Technology adoption level:
in 1000BC
1.45
(1.86)
in year 0
II
III
Log Population 2002
IV
V
1.59
(2.57)
2.14
(2.55)
in 1500AD
VII
VIII
1.45
(2.05)
0.95
(1.06)
3.07
(3.69)
YES
IX
1.59
(2.9)
4.5
(5.71)
Continent Dummies
VI
Log Arable land
2.55
(3.52)
YES
YES
N
105
124
107
114
136
118
110
132
114
R2
0.37
0.38
0.56
0.21
0.14
0.43
0.33
0.16
0.31
Note: t-statistics in parenthesis computed using robust standard errors.
All regressions include major and minor European Influence dummies and a constant.
As we have noted above, the association between land area and ancient technology could be
reverse causality, since a larger land area contained a larger sample of cultures and
technologies from which we are coding the “best.” Moreover, total GDP and population
are correlated with land area, so this reverse causality could contaminate these results also.
As one check on this potential problem, we include land area as a right hand side variable in
these regressions (although there are still major concerns about endogeneity of land area).
Table 9 shows that the same results hold for total GDP and population when we include
land area. It is also possibly illuminating that per capita GDP today is uncorrelated with land
area, so the association between contemporaneous technology (as reflected in today’s per
capita GDP) and land area does not seem to reflect any dominant “sampling” effect
(although this could have changed from ancient times). These results provide suggestive
evidence that the results for GDP and Population are not driven by possible reverse
causality between land area and ancient technology.
28
Table 9: Technology History, Current GDP and per capita GDP after controlling for
arable land
Dependent Variable
Log GDP 2002
I
Overall Technology adoption level:
in 1000BC
1.46
(2.29)
in year 0
II
IV
V
2.11
(3.17)
Major and Minor european
involvement dummies
VII
1.47
(2.72)
4.06
(9.28)
0.79
(9.3)
VI
1.43
(2.6)
in 1500AD
Log arable land area
III
Log per capita GDP 2002
0.63
0.52
(5.18) (5.34)
3.37
(10.34)
-0.01 -0.03
(0.14) (0.54)
YES
-0.18
(4.48)
YES
0.02
(0.43)
NO
N
102
121
105
102
121
105
127
R2
0.61
0.49
0.74
0.17
0.13
0.8
0
Note: t-statistics in parenthesis computed using robust standard errors.
All regressions include a constant.
To explore further the persistence of technology, we construct a measure of current
technology level based on Comin, Hobijn and Rovito (2006). This measure captures (minus)
the average gap in the intensity of adoption of ten major current technologies with respect to
the US. 15 More specifically, for each technology, Comin, Hobijn and Rovito (2006) measure
In particular, these technologies are electricity (in 1990), internet (in 1996), pc’s (in 2002), cell phones (in
2002), telephones (in 1970), cargo and passenger aviation (in 1990), trucks (in 1990), cars (in 1990) and tractors
(in 1970) all in per capita terms.
15
29
how many years ago did the United States last have the usage of technology ‘x’ that country
‘c’ currently has. We take these estimates, normalize them by the number of years since the
invention of the technology to make them comparable across technologies, take the average
across technologies and multiply the average lag by minus one to obtain a measure of the
average intensity gap with respect to the US.
Note that this measure of current technology adoption differs from the historical measures
in that it includes the intensive margin. This is the case because in the last 100 years or so,
the first unit of technology has diffused very quickly across countries. Therefore, the
intensive margin of technology adoption has now become the relevant margin to explain
cross-country differences in technology.
The first three columns of Table 10 present the association between technology adoption in
the three historical periods and current technology adoption. The main finding is that
current technology is correlated with historical technology adoption in all three periods. As
one would expect, the correlation is higher the more recent is the historical period. This
remarkably high persistence of technological differences over 3000 years of human history
reinforces the key finding of our paper. (It is also reassuring that the error rate on our
technological measures is not disastrously high.)
30
Table 10: Effect of ancient technology on current technology
Dependent
Variable
Current technology adoption
Overall Technology adoption level:
in 1000BC
0.18
-0.01
-0.02
(2.69)
(0.15)
(0.38)
in year 0
0.24
(3.11)
in 1500AD
0.03
(0.54)
0.44
(8.17)
Distance from
Equator
0.21
(2.6)
0.15
(2.05)
0.65 0.63 0.47
(7.57) (8.01) (3.72)
Continent
dummies
N
R2
0.16
(2.36)
NO
110
0.23
131
0.24
NO
111
0.51
102
0.56
121
0.55
YES
106
0.52
110
0.62
131
0.6
111
0.66
Note: t-statistics in parenthesis
All regressions include major and minor European involvement dummies.
Columns 4 through 6 of Table 10 shows that from 1500 AD to the present, technology
adoption is also highly persistent after controlling for the distance to the Equator , although
1000 BC and 0 AD are not robust to this control. Lastly, controlling for continent dummies,
the within-continent technology differences are also persistent for 0 and 1500 AD, although
not for 1000 BC. The persistence of technology across the last 500 years, or the last 2000
years, is not just due to differences between continents. Again, we think of this robust
persistence of technology differences over very long periods as the main finding of this
paper.
An important question is how our findings of technology and income persistence relate to
the “reversal of fortune”finding of Acemoglu, Johnson, and Robinson (2002). As shown in
Table 6 earlier and Table 10 in this section, we found that there is a strong positive
association between overall technology adoption in 1500 A.D. and current
31
development/technology. Based on these results, we have found some kind of “persistence
of fortune”. When controlling for the European influence dummies we found that the effect
of historical technology adoption in 1500 A.D. on current development for the former
colonies becomes even more positive. However, the strong effect of the European influence
dummies could themselves be capturing precisely the AJR story that European settlers
brought good institutions that dramatically changed later “fortunes.” We could alternatively
interpret the dummies as representing technology transfer, but we do not really have strong
enough evidence to contradict AJR’s institutional interpretation. We plan to investigate this
further in future work.
6. Conclusions
The main finding of this paper is a simple one: centuries-old technological history is
associated with the wealth of nations today. This is largely robust to including continent
dummies and geographic controls, so it is not just driven by “Europe vs. Africa” or “tropical
vs. temperate zones.” The most surprising part of the finding is just how old the history can
be and still be correlated with modern outcomes. Our most robust finding is that technology
in 1500 AD is correlated with development outcomes today, itself remarkably old when we
consider that most history discussions of developing countries start with European contact
and colonization. Even more surprising is that technology in 1000 BC and 0 AD has a
significant correlation with modern outcomes in many specifications. While of course this
finding is subject to standard caveats about the quality of data from ancient periods, the
finding has important implications to the extent that it survives those caveats.
The burning question about our results is WHY do technology/income differences persist
for such long periods. Is it that old technology is complementary to new technology, that
technology is reflecting the effect of institutions, is it the positive technology-population
feedback discussed in many recent papers, or is it one of the many other long-run factors
previous empirical researchers have stressed? Exploring these many questions adequately
would require a complete paper in itself, which we are presently pursuing.
We think our results might also provide food for thought to the policymakers and
international institutions who seem to overemphasize the instruments under their control,
with a seemingly excessive weight being placed on the behavior of modern-day governments
and development strategies as a determinant of development outcomes. We do not claim
that history is destiny. Our technology history only explained a partial share of the modern
day variance of development outcomes, and even then may be proxying for some other very
long run factor, and so history is obviously not everything. Yet our results show very old
history displays a surprisingly high association with today’s outcomes.
32
References
Acemoglu, D., S. Johnson and J. Robinson,(2002), “Reversal of Fortune: Geography and
Institutions in the Making of the Modern World Income Distribution” Quarterly Journal of
Economics, (November), v. 117, iss. 4, pp. 1231-94
Adshead, S.A.M. (1988) China in World History. New York: St. Margin's Press, Inc.
Barro, Robert J. and Rachel M. McCleary. 2006. Religion and Economy. Journal of Economic
Perspectives—Volume 20, Number 2—Spring 2006—Pages 49–72
Barro, Robert J. and Rachel M. McCleary. 2003. “Religion and Economic Growth.”
American Sociological Review. October, 68, pp. 760–781.
Barry, H. , III, & L.M. Paxson (1971) "Infancy and Early Childhood: Cross Cultural Codes
2," Ethnology. 10, pp. 466-508.
Basalla, G. (1988) The Evolution of Technology. Cambridge, United Kingdom: Cambridge
University Press.
Black, J. (1996) The Cambridge Illustrated Atlas of Warfare: Renaissance to Revolution 1492-1792.
Cambridge, United Kingdom: University of Cambridge.
Bockstette, Valerie, Areendam Chanda, and Louis Putterman, (2002)“States and Markets:
The Advantage of an Early Start,” Journal of Economic Growth 7: 347-69, 2002.
Carneiro, R.L. (1970) "Scale Analysis, Evolutionary Sequences, and the Rating of Cultures"
In: R. Naroll & R. Cohen eds. A Handbook of Method in Cultural Anthropology. Garden City,
New York: The Natural History Press. pp: 834-871.
Collis, J. (1984) The European Iron Age. New York: Routledge.
Diamond, J. (1997) Guns, Germs, & Steel. New York: W.W. Norton & Company.
Easterly, W. and R. Levine (2003) “Tropics, germs, and crops: the role of endowments in
economic development” Journal of Monetary Economics, 50:1, January 2003.
Easterly, W. and R. Levine (1997) "Africa’s Growth Tragedy: Policies and Ethnic Divisions,”
November 1997, Quarterly Journal of Economics. CXII (4), 1203-1250.
Encyclopedia Britannica (2006h) "chariot," Encyclopædia Britannica Online.
http://search.eb.com/eb/article-9022532
Erman, A. (1971) Life in Ancient Egypt. New York: Dover Publications Inc.
Fernandez, Raquel. 2007. "Culture and Economics," in the New Palgrave Dictionary of
Economics, 2nd edition, edited, by Steven N. Durlauf and Lawrence E. Blume, Palgrave
Macmillan (Basingstoke and New York), forthcoming, 2007.
33
Gallup,J., J. Sachs, and A. Mellinger (1999) "Geography and Economic Development,",
International Regional Science Review, Vol. 22, No. 2, pp. 179-232, August 1999.
Galor O., "From Stagnation to Growth: Unified Growth Theory", in P. Aghion and S.
Durlauf, editors, Handbook of Economic Growth, 2005, 171-293.
Galor, O., and D.N. Weil (2000), “Population, Technology and Growth: From the
Malthusian regime to the Demographic Transition”, American Economic Review 110: 806828.
Guiso, Luigi, Paola Sapienza and Luigi Zingales. 2006. Does Culture Affect Economic
Outcomes? Journal of Economic Perspectives—Volume 20, Number 2—Spring 2006—
Pages 23–48
Greene, K. (2000). Technological Innovation and Economic Progress in the Ancient World:
M.I. Finley Re-Considered. The Economic History Review, New Series, Vol. 53, No. 1 (Feb.
2000), pp. 29-59.
Hall, R. and C. Jones (1999) "Why Do Some Countries Produce So Much More Output per
Worker than Others?", Quarterly Journal of Economics, February 1999, Vol. 114, pp. 83-116.
Hogg, O.F.G. (1968) Clubs to Cannon, Warefare and Weapons before the introduction of Gunpowder.
London: Gerald Duckworth & Co. Ltd.
Jones, C.. “Growth and Ideas.”in P. Aghion and S. Durlauf, editors, Handbook of Economic
Growth, 2005
Kerr, R. & N. Wood (2004) Joseph Needham Science and Technology in China, Vol 5 Chemistry and
Chemical Technology, Part XII: Ceramic Technology. Cambridge, United Kingdom: Cambridge
University Press.
Kipfer, B.A. ed. (2000) Encyclopedic Dictionary of Archaeology. New York: Lluwker
Academic/Plenum Publishers.
Klenow, P. and A. Rodriguez-Clare (1997) "The Neoclassical Revival in Growth Economics:
Has It Gone Too Far?" with Andrés Rodríguez-Clare, NBER Macroeconomics Annual
1997, B. Bernanke and J. Rotemberg ed., Cambridge, MA: MIT Press, 73-102
Kremer, M. “Population Growth and Technological Change: One Million B.C. to 1990,” The
Quarterly Journal of Economics, Vol. 108, No. 3 (Aug., 1993), pp. 681-716.
Lucas, A. (1934) Ancient Egyptian Materials & Industries. London: Edward Arnold & Co.
Macksey, K. (1993) The Penguin Encyclopeida of Weapons and Military Technology, Prehistory to the
Present Day. New York: Penguin Books USA.
34
McEvedy, C. and R. Jones (1978) Atlas of World Population History Penguin Books, London.
Mokyr, J. (1990) The Lever of Riches: Technological Creativity and Economic Progress. Oxford; New
York; Toronto and Melbourne: Oxford University Press.
Mokyr, J. (2005) , The Intellectual Origins of Modern Economic Growth, The Journal of
Economic History, VOLUME 65 JUNE 2005 NUMBER 2, 285-351.
Murdock, G.P. (1967) Ethnographic Atlas. University of Pittsburgh Press.
Murdock, G.P. & D.O. Morrow (1970) "Subsistence Economy and Supportive Practices" In:
A. Spoher ed. Ethnology, 9. No. 3. pp: 302-330.
Murdock, G.P. & C. Provost (1973) "Measurement of Cultural Complexity" In: A. Tuden ed.
Ethnology, 12, No. 4. pp: 379-392.
Murdock, G.P. & S.F. Wilson (1972) "Settlement Patterns and Community Organization,"
Ethnology. 11, pp. 254-295.
O'Brien, P.K. ed. (1999) Atlas of World History. New York: Oxford University Press.
Peregrine, P.N. (2003) "Atlas of Cultural Evolution" In:J.P. Gray ed. World Cultures,14, no. 3.
pp: 1-75.
Peregrine, P.N. , & M. Ember eds. (2001a) Encyclopedia of Prehistory. Vols 1-8. New York:
Kluwer Academic/Plenum Publishers.
Peregrine, P.N. , & M. Ember eds. (2001b) Encyclopedia of Prehistory. Vol 9.New York: Kluwer
Academic/Plenum Publishers.
Putterman, Louis and David N. Weil (2007), “Post-1500 Population Flows and the Long
Run Determinants of Economic Growth and Inequality,” mimeo, Brown University,
2007.
Rodrik, D. A. Subramanian, F. Trebbi (2004) “Institutions Rule: The Primacy of Institutions
over Geography and Integration in Economic Development.” Journal of Economic Growth,
June 2004, v. 9, iss. 2, pp. 131-65
Rosenberg, N. and L.E. Birdzell Jr. How the West Grew Rich: The Economic Transformation of the
Industrial World, Basic Books: New York, paperback reprint, 1987.
Rouse, I. (1972) Introduction To Prehistory, A Systematic Approach. New York: McGraw-Hill
Book Company.
Scarre, C. ed. (1988) Hammond Past Worlds The Times Atlas of Archaeology. Maplewood, New
Jersey: Hammond Inc.
35
Spolaore, Enrico and Romain Wacziarg, The Diffusion of Development, mimeo, Tufts and
Stanford, March 2006
Tuden, A. & C. Marshall (1972) "Political Organization," Ethnology. 11, pp. 436-464.
Wager, D. B. (1993) Iron and Steel in Ancient China. New York: E.J. Brill.
36
Figure 1: Overall technology adoption in 1000 B.C.
37
Figure 2: Overall technology adoption in 0 A.D.
38
Figure 3: Overall technology adoption in 1500 A.D.
39
Figure 4: Technology in 1500 and current development
AUL
IRE NOR
SWZDEN
AUS
NTH
GER
FRAUNK
JAP BLG
FIN
ITA
SWE
SNG
SPA
10
GRE
POR
MTA
KOR
CZE
HUN
USA
CAN
NEW
SAU
ARG
SAF
CHL
BOT
MEX
COS
URG
BRA
TON
COL
GABBLZ
NAM
PAN
FIJ
VNZ
PER ELS
PRG
GUY GUA
ECU
NIC
HON
BOL
LES
PNG
ANG
CAM
POL
LTH
RUS
THA
ROM
TUN
IRN
TUR
ALG
BOS
UKR
CHN
PHI
EGY
SYR
IDS
IND
VTM
PAK
CMB
BAN
LAO
MLY
MOR
GHA
GUI
SUD
UZB
SEN MON
UGA
NEP
CAR
BKF
BENKEN
CHDNGA
CON
MAL
ZAM
MAD NIG ETH
GNB
ZAI
TAN
SRL
MAU
IVO
0
.2
.4
.6
.8
1
Overall technology adoption level in 1500AD
9
8
7
log per capita income in 2002
11
6
40
Figure 5: (Conditional) overall technology adoption in 1000 B.C. and
(conditional) current development
BOT
NAM
MOZ
BKF
BEN
KEN
CON
MAL
NGA TAJ
ZAM
ETH
ZAI
TAN
2
ITA
GRE
1
CRO
IRN
TUR
ALG BOS
CHN
0
EGY MOR
PAK
SUD
-1
MAU
CHD
Residual log per capita income in 2002
JAP
NOR
DEN
SWZ
AUS
NTH
BLG
FRA
GER
FIN
UNK
SWE
SPA
POR
KOR
CZE
HUN
EST
POL
LTH
ARG
MLY LTV
SAF
RUS
CHL
MEX
COS
BUL
THA
ROM
GAB
URG BRA
BLZ
KAZ USA
BEL
COL
PAN
TKM UKR
CAN
AUL
VNZ
GUY
ELS PER
SWA PRG
GUA
IDS
ECU
NIC IND
PNG
VTM
ANG
HON
GHA
GUI
BOL
CAM
LES
CMB
GAM
BAN
LAO
UZB
MON
SEN
IVO
UGA NEP MOL
-2
-.5
0
.5
Residual overall technology adoption level in 1000BC
41
JAP
SNG
HUN
SVK
EST
GUY
PNG
ARG
BOT
URG
GAB
BLZ
NAM
PRG
ANG
GHA
CAM
GAM
MON
IVO
BKF
BEN
MOZ
CON
NGA
ZAM
GNB
ZAI
MLW
SRL
LTH
CRO
LTV
MLY
CHL
COS
BUL
THA
ROM
BRA
BOS
BELDOM
COL
PAN
UKR
CAN
VNZ
AUL
ALB
PHI
NEW
IDS
ECU
INDNIC
VTM
HON
CMB
LAO
NEP
NOR
IRE
2
DEN
SWZ
AUS
NTH
BLG
FRA
GER
ITA
FIN
UNK
SWE
SPA
ISR
GRE
POR
KOR
CZE
OMA
SAU
POL
1
Residual log per capita income in 2002
Figure 6: (Conditional) overall technology adoption in 0 A.D. and (conditional)
current development
SAF
RUS
MEX
MAC
IRN
TUR
KAZ
ALG
USA
TKM
LEB
CHN0
PERJORELS
MOR
EGY
SYRSWA
GUA
BOLPAKLES
SUD
BAN
UZB
SEN-1
MAU
UGA
CAR
KEN
CHD
MAL
TAJ
YEM
NIG
MAD
ETH
TAN -2
-.6
-.4
-.2
0
.2
Residual overall technology adoption level in 0AD
42
SAU
POL
LTH
MLY
SAF
RUS
CHL
MEX THA
COS
TON
TUN
ROM
IRN
URG
BRA
TUR
NAMGABBLZ
ALG
BOS
USA
FIJ
COL
PAN
UKR
CAN
CHN
AUL VNZ
GUY
PER
PHIELS
PRG
MOR
EGY
SYR
NEW
GUA
IDS
ECU
IND
NIC
PNG
VTM
ANG
HON
GHA
GUI
BOL
CAM
LES
PAK
CMB
SUD
BAN
LAO
UZB
SEN MON
MAU
IVO
UGA
NEP
CAR
BKF
BENKEN
CHDNGA
CON
MAL
ZAM
NIG
ETH
GNBMAD
ZAI
TAN
SRL
BOT
ARG
IRE NOR
2
SWZDEN
AUS
NTH
BLG
FRAUNK
JAP GER
FIN
ITA
SWE
SNG
SPA
GRE
POR
MTA
KOR
CZE
HUN
1
0
-1
Residual log per capita income in 2002
Figure 7: (Conditional) overall technology adoption in 1500 A.D. and
(conditional) current development
-2
-.5
0
.5
Residual overall technology adoption level in 1500AD
43
×

Report this document