A Western Reversal Since the Neolithic?
The Long-Run Impact of Early Agriculture
Ola Olssony
University of Gothenburg
Christopher Paikz
NYU Abu Dhabi
August 21, 2015
Abstract
In this paper, we document a reversal of fortune within the Western agricultural
core, showing that regions that made an early transition to Neolithic agriculture are
now poorer than regions that made the transition later. The …nding contrasts recent
in‡uential works emphasizing the bene…cial role of an early transition. We advance a
model of a long-run development reversal where we argue that an early transition to
agriculture was associated with collectivistic cultural norms and autocratic institutions
that were detrimental to economic development in the long run. Using data from
a large number of carbon-dated Neolithic sites throughout the Western agricultural
area, we determine an approximate date of transition for about 60 countries, 280
medium-sized regions and 1,400 small regions. Our empirical analysis shows that
there is a robust negative, reduced-form relationship between years since transition to
agriculture and contemporary levels of income both across and within countries. Our
results further indicate that the reversal had started to emerge already before the era
of European colonization. In line with our model, we also show that countries which
made an early transition to agriculture experienced a delayed adoption of democracy.
Keywords: Neolithic agriculture, comparative development, Western reversal
JEL Codes: N50, O43
1
Introduction
A striking feature of contemporary comparative development is that regions that made an
early transition to agriculture and to civilization, like current Iraq, Egypt and Syria, are
now relatively poor and have recently experienced numerous symptoms of state collapse.
In the northern periphery of the Western cultural zone, regions that made a very late
We are grateful for useful comments from Ran Abramitzky, Daron Acemoglu, Lisa Blaydes, Carles
Boix, Matteo Cervellati, Ernesto Dal Bo, Carl-Johan Dalgaard, Jared Diamond, James Fearon, Oded
Galor, Avner Greif, Douglas Hibbs, Ian Hodder, Saumitra Jha, Timur Kuran, Anastasia Litina, Neil
Malhotra, Peter Martinsson, Stelios Michalopoulos, Louis Putterman, James Robinson, Gerard Roland,
Jacob Shapiro, Pablo Spiller, Enrico Spolaore, Romain Wacziarg, David Weil, and from seminar partipants
at Berkeley Haas, Brown, East Anglia, Gothenburg, Stanford, the Zeuthen Workshop in Copenhagen and
the CAGE Workshop in Warwick. Harish S.P. provided excellent research assistance. This work was
supported in part by a grant from the Air Force O¢ ce of Scienti…c Research (AFOSR) award number
FA9550-09-1-0314.
y
Email: [email protected]
z
Email: [email protected]
1
transition to civilization, like current Sweden and the Netherlands, now host prosperous
and stable democracies. In comparison, these observations stand in stark contrast to some
of the seminal works in the literature proposing a positive association between an early
agricultural transition and high income levels today.
In this paper, we document a strong pattern of a Western reversal of fortune from the
Neolithic Revolution to the current day. The Neolithic Revolution was a momentous event
which introduced agriculture to humans around 12,000 BP.1 It marked for the …rst time
humans’departure from their hunter-gatherer lifestyle for agriculture and sedentary living.
We argue that although industrialization in particular contributed critically to the massive
divergence between countries that emerged during the last 200 years, our analysis suggests
that the relative prosperity of North European regions followed a longer trajectory with
roots in the Neolithic.
We outline a model of a long-run development reversal within an agricultural core
region, linking the time of transition to agriculture several thousand years ago, to contemporary levels of income per capita. We hypothesize that an early transition to agriculture
was associated with high parasite stress, an interdependent form of agricultural production,
and an imminent threat of predation by less advanced neighbours. These circumstances,
in turn, tended to give rise to collectivist cultural norms in history, favoring in-group
orientation, obedience, and an acceptance of a hierchical social order. The collectivist
culture led to weak private property rights and extractive and autocratic formal institutions, which eventually resulted in a weak modern economic performance. In regions that
adopted agriculture late, all these tendencies were the reverse, i.e. there was a stronger
presence of an individualist culture with secure private property rights, democracy, and a
strong contemporary economic development.
By creating a comprehensive data set on the Neolithic transition for all Western areas,
our empirical analysis con…rms that regions which made an early transition to agriculture
do tend to be relatively poor today in terms of GDP per capita, whereas regions that
made a late transition are now relatively rich. On the country level, this basic …nding is
robust to controlling for an extensive set of geographical and historical controls, as well
as to using instrumental variable techniques. On a …ner regional level, using data on 280
medium-sized and about 1,400 small regions in Europe, there is likewise a strong overall
negative relationship also within countries. The negative relationship is particularly strong
within Italy, Spain and Turkey. In addition, we show that this reversal started to manifest
itself during the Early Modern Era 1500-1800, i.e. before the big divergence in levels of
prosperity caused by the Industrial Revolution.
Our empirical results are further consistent with key intermediate elements of our
model of a long-run reversal. We document that the time since the Neolithic transition
has a very strong positive impact on historical levels of infectious disease and a negative
impact on our measures of historical levels of democracy. Furthermore, the reduced-form
association between time since transition and current levels of income is substantially
1
BP refers to Before Present, where the present period is set to be the year 2000.
2
weakened when democracy and pathogen prevalence are included, suggesting that these
are most likely important intermediate channels of causation, in line with our model.
Our paper is related to a vast literature on Western economic and social history.2 In a
companion paper to this one, Olsson and Paik (2015) present a theoretical framework on
the migration of individualistic farmers from an agricultural core region to new settlements
in the periphery. The paper provides evidence that an extensive time since the transition
to agriculture continues to be associated with collectivistic cultural norms today, whereas
late adopters should be more individualistic.3
In his classic work on the long-run impact of plant and animal domestication, Jared
Diamond (1997) argues that the region making up the Fertile Crescent in the Middle East
(roughly Israel, Lebanon, Syria, Southeastern Turkey, Iraq, and Western Iran) was the …rst
to make the transition to agriculture by about 12,000 BP because of its superior access to
plants and animals suitable for domestication. Even though other regions such as China
also developed agriculture independently, the highly favorable biogeography in the Fertile
Crescent and in large parts of the rest of the Western agricultural core (encompassing
Europe, South Asia including western India, and North Africa) implied that this part of
the world could develop civilization, statehood, science, military technology, and political
strategy much earlier. By 1500 CE, these advantages of an early start allowed European
countries to colonize and dominate much of the rest of the World.4
In subsequent research, Diamond’s hypothesis has been tested: That current income
levels across the world should have a positive relationship with biogeographical suitability
for agriculture in prehistory and/or with the timing of the agricultural transition (Hibbs
and Olsson, 2004; Olsson and Hibbs, 2005; Putterman, 2008; Ashraf et al, 2010, Putterman
and Weil, 2010; Bleaney and Dimico, 2011). Most of these studies have con…rmed a positive
relationship on a worldwide basis, suggesting that countries that made the transition early
had a long-term advantage that is still detectable in current levels of prosperity.5
Our paper makes a fundamentally di¤erent point by showing that this positive relationship in a worldwide sample, a la Diamond, is mainly driven by di¤erences between
independent agricultural core region averages, whereas the relationship within the Western
core (the region that made the transition …rst) is actually negative. Figure 1 illustrates
this point and also summarizes one of the key insights of the paper. The graph shows
the bivariate relationship between log GDP per capita in 2005 and the time (in years)
since agricultural transition for 158 countries. The relationship in the …gure is positive
2
An incomplete listing of some of the most important works includes Wittfogel (1957), Jones (1981),
Kennedy (1988), Mokyr (1990), North (1990), Landes (1998), Pomeranz (2000), Clark (2008), and Morris
(2010).
3
As proxy variables for the collectivism/individualism dimension of culture, Olsson and Paik (2015)
employ the often-used answers to the questions in World Value Survey on the importance of obedience in
upbringing and an individual’s sense of control of one’s life.
4
Other important contributions to our understanding of the Neolithic transition include Harlan (1995),
Smith (1998), and Bellwood (2005). In the economics literature, see Ashraf and Michalopoulos (2011) for
an account of the spread of agriculture in Europe using the data from Pinhasi et al (2005). Weisdorf (2005)
reviews the arguments on the transition to agriculture.
5
For a recent overview of this research, see Spolaore and Wazciarg (2013).
3
when all 158 countries are included in the regression. However, when the relationships
within three agricultural core areas (Western, Sub-Saharan Africa, and East Asia which
all witnessed adoption of agriculture independently) are investigated separately, we see in
the graph that the relationships turn negative.6 The negative relationship for the Western
core region in the upper right corner, is the main …nding of this paper that we document
in various ways.
Figure 1
In line with this evidence, we suggest the following revised interpretation: Average
income levels per capita are higher in the Western core than in all other parts of the
world due to the advantages of an early transition to agriculture and civilization, but in
comparisons within agricultural core areas, early adoptions of agriculture led to a relatively
low level of current economic development. In the original Diamond model, Neolithic
biogeography played an important role for Eurasia as a whole by introducing agriculture
to the hunter-gatherers in the continent earlier than in any other core areas in the world.
In this paper, we analyze variations within the Western core, and argue that it was the
timing of the agricultural transition which determined the type of institutional trajectory
and economic performance that was experienced in the long run.
Our paper is not the only one addressing the dramatic rise of Europe after 1500.
Tabellini (2010) also studies long-run development on a country and regional level in
Europe and, like us, attribute a critical intermediate role to culture for understanding
how history since about 1850 a¤ected current levels of income. Our argument is that
the underlying reasons for the variation in institutions and culture can be traced back
thousands of years to the Neolithic transition.7
Acemoglu et al (2005) attribute the
fast development of Western Europe to Atlantic trade and the faster transition towards
democracy in some countries as a result of the trade. Our results show that there are signs
of a reversal already before Atlantic trade took o¤. Nothing in our …ndings rule out that
Atlantic trade indeed had an important role for strengthening democracy. Our notion of
inclusive democratic institutions is further close to how those are de…ned in Acemoglu and
Robinson (2012).
On the topic of important historical events, Voigtländer and Voth (2012) attribute a
crucial role to the Black Death of 1347 which shocked many parts of Europe into a new
pattern of rising wages and higher urbanization rates, which further sped up susceptibility
to plagues, and a greater …scal capacity which increased the means for and prevalence of
wars. The outcome was a society with lower population pressures and rising per capita
incomes. the authors use model simulations and their focus is on explaining how Europe
as a whole became richer rather than how some countries overtook others.8
6
Please see Figures C1 and C2 in Appendix C that document in more detail the negative and signi…cant
relationships within East Asia and Sub-Saharan Africa.
7
For recent treatments of the links between culture and institutions, see also Gorodnichenko and Roland
(2010, 2011, 2013), Belloc and Bowles (2013) and the overview by Alesina and Giuliano (2013).
8
In our empirical analysis, we address whether the Black Death could be a potential intermediate
4
In a similar vein as our paper, Bosker et al’s (2013) city-level analysis of the Western
zone show that the faster growth of the European cities compared to Islamic cities during
the Medieval period can be partly attributed to the more democratic and republican
traditions in Europe, especially in Italy and Flanders. Blaydes and Chaney (2013) focus
instead on ruler duration and …nd that European durations were longer than in the Islamic
world by 900 CE. The primary reason for this, according to the authors, was the feudal
system which decentralized power in Europe and gave rise to greater political stability and
more bene…cial economic development.
We believe that our paper makes the following important contributions to the literature. First, we provide strong evidence of a distinctive reversal of development levels since
the Neolithic within the Western core area. Second, we demonstrate how previous results
of a positive association between time since agriculture and current income can be reconciled with our evidence of a negative relationship within the Western agricultural core.
Lastly, consistent with our model, our empirical evidence strongly indicate that historical
levels of disease and democracy are key intermediate factors in explaining the reversal.
The paper proceeds in the following order: Section 2 introduces the theoretical framework whereas Section 3 presents the data. Section 4 outlines the empirical strategy.
Section 5 then proceeds with the results. Finally, Section 6 concludes by summarizing the
main …ndings and o¤ering avenues for future research.
2
A Theoretical Framework
In this section, we present the basic theoretical framework behind our empirical analysis.
Most importantly, we outline a model of a long-run development reversal within an agricultural core region that serves as the general theoretical framework from which we derive
the key hypotheses for our empirical analysis.
2.1
Agricultural transition and current levels of development
In his widely acclaimed work on the long-run importance of the Neolithic revolution, Jared
Diamond (1997) argues that societies which transited early into agriculture in prehistory
achieved a head start over other societies in terms of social, political, and technological
development, which ultimately explained why it was Europeans who colonized the native
Americans rather than vice versa. The early onset of agriculture in the Western core was
in turn explained by a superior access to suitable plants and animals for domestication.9
When the transition to sedentary agriculture was complete, societies typically embarked
on a path towards civilization that eventually resulted in codi…ed languages, large public
mechanism. However, we also assess the impact of the Black Death with the caveat that our …nding is
hampered by the unreliable quality of mortality data from the period.
9
Like Diamond (1997) and Morris (2010), the regions that we refer to as the "Western core" include
Europe, North Africa, the Middle East, and South Asia including India and the countries west thereof.
All these regions were connected through trade and migration and based their economies on domesticated
plants and animals from the Fertile Crescent.
5
monuments, organized religions, hierarchical power structures, and full-blown states. After
having experienced thousands of years in densely populated and intensely competitive
agricultural states, it was only natural that the Western colonial powers could dominate
most of the world outside Eurasia from 1500 CE onwards.
Later works such as Hibbs and Olsson (2004), Olsson and Hibbs (2005), Putterman
(2008), and Putterman and Weil (2010) took this analysis one step further and argued that
the timing of the transition to agriculture should also matter for understanding contemporary comparative levels of wealth and institutional development around the world. Indeed,
when we use data from Putterman (2006) on the date of …rst transition to agriculture for
all 158 countries in the world and correlate that with log of GDP per capita in 2005, we
…nd a clearly positive overall relationship as in Figure 1. Results like these might thus be
interpreted as giving support to the idea that an early transition to Neolithic agriculture
is advantageous to countries even today.
However, when considering the history of the …rst zone to make the transition to full‡edged agriculture - the Western core - it is a striking fact that some of the countries
that once made up the cradle of agriculture and civilization in the "Fertile Crescent" are
currently much less developed than countries in the periphery of the zone. Consider for
instance Iraq and Syria, hosting some of the very earliest agricultural sites in our sample
(from before 12,000 BP) and some of the earliest signs of civilization already by 5500 BP.
Both countries are today close to being regarded as failed states with a very weak rule of
law, a long history of autocratic government (in Iraq, at least until 2003), and stagnant
economies.
This situation might be compared to for instance a Nordic country like Sweden in the
very periphery of the Western core. Sweden adopted agriculture only 5500 BP and did
not host a state above tribal level until around 1150 CE (Bockstette et al, 2002). Despite
this very late start as a "civilization", Sweden (as well as most of its Nordic neighbors)
is considered to be one of the most economically prosperous, egalitarian, and democratic
countries in the world.
Casual observations like these suggest that although the countries in the Western
core doubtlessly owe some of their relatively high average levels of development to their
long history of agriculture and civilization, it is less apparent that an early transition to
agriculture was advantageous for comparative development within the zone. Indeed, the
separate negative regression line in Figure 1 for the countries in the Western subsample
strongly indicates that an early transition to agriculture is associated with lower levels of
income per capita today.
How can we reconcile the two tendencies in the graphs above? This is what we turn
to in the section below.
6
2.2
A model of a long-run development reversal
The condensed version of our model of a development reversal, is illustrated in Figure
2. It should be noted that our framework applies to long-run development within an
agricultural zone such as the Western or East Asian regions and does not serve to explain
di¤erences between countries in di¤erent zones.
2.2.1
Parasites, production and predation
The transition to agriculture, which …rst happened in the Fertile Crescent in current Israel,
Jordan and Syria about 12,000 BP, implied a shift from a nomadic hunting and gathering
lifestyle to one characterized by cultivation of domesticated crops by sedentary farming
populations.10 In these early farming villages with permanent dwellings and accumulated
resources, population growth increased dramatically.
Figure 2
The …rst group of causal e¤ects of this new lifestyle that we want to emphasize in our
model is that sedentary agriculture gave rise to an increased parasite stress and pathogen
prevalence, a new type of functionally interdependent production, and a greater risk of
violent predation from primitive hunter-gatherer neighbors (1a in Figure 2). The fact that
the transition to agriculture was associated with the introduction of numerous new and
often lethal health hazards is well documented in the literature on paleopathology.11 The
much greater population density and the shared quarters with domesticated animals like
goats and pigs, led to a large increase in the prevalence of infectious crowd diseases such
as smallpox, measles, and in‡uenza (Diamond, 1997). Despite these and other health
hazards of early farming, fertility and net population increased rapidly.
A second e¤ect was that sedentary agriculture introduced a greater interdependence
between individuals in the production process. In particular, in the fragile semi-arid
regions of the Fertile Crescent where agriculture was often based on irrigation, there was
a strong need for coordinated work among the early farmers. In the extremely demanding
irrigation system of Mesopotamia, involving canals and ‡ood-gates, communities faced
the risk of outright disaster if water was not managed e¢ ciently in a coordinated fashion
(Postgate, 1995). Similarly, rice production in China required intensive and interdependent
work e¤ort by farmers (Talhelm et al, 2014).
A third implication of agricultural production was the greater vulnerability to predation from violent neighbors. The accumulation of resources such as harvested crops,
tools, shrines, and domesticated animals in the farming villages were a natural target for
more primitive groups. In Mesopotamia, as well as in China, there are numerous known
10
In this paper, we will not provide an analysis on why the original domestication of plants and animals
happened where it did. See for instance Smith (1995) or Bellwood (2005) for general treatments on the
rise of agriculture.
11
Classic works in this tradition include Cohen and Armelagos (1984) and Steckel and Rose (2002).
7
occasions when "barbarian" tribes looted the most prosperous centers of civilization and
sometimes even established a new regime (Morris, 2010).
2.2.2
Collectivism vs individualism
As outlined in more detail in Olsson and Paik (2015), we argue that the three factors
above contributed importantly to the emergence of collectivist cultural norms in the early
farming communities (link 1b in Figure 2).12 In line with the voluminous literature in social
science on collectivism-individualism, we de…ne collectivist norms as being characterized
by in-group orientation, conformity, obedience, importance of tradition, hierchical social
structures and autocratic leadership, whereas individualism is associated with out-group
orientation, independent individuals, openness to new ideas, egalitarianism and democratic
leadership (Triandis, 1995; Thornhill and Fincher, 2013).
The strong link between parasite stress and various facets of collectivism has been
demonstrated in a number of studies by biologists Randy Thornhill and Corey Fincher
(Fincher et al, 2008; Thornhill and Fincher, 2013). The mechanism behind this relationship is that individuals in areas with a high prevalence of infectious disease will adopt
in-group oriented behavior as a survival strategy to avoid infection. Openness to strangers
and to other groups is simply too dangerous in this kind of environment. In-group orientation, low social and geographical mobility, an emphasis on tradition, and other typical
collectivist norms thus naturally evolve. Over the longer term, collectivism emerges both
as a conscious choice but also through natural selection since individuals with collectivist
cultural norms have a greater probability of survival in a region with a high prevalence
of pathogens. In low-risk environments, on the other hand, individualist norms favoring
openness to new ideas and trade often have a comparative advantage over collectivist
norms.
The hypothesis that the nature of agricultural production might have important consequences for the broader organization of societies, goes back at least to Wittfogel (1957).13
Our argument is that the interdependent nature of production in the Western agricultural core area strongly favored the evolution of collectivist cultural norms. Early farming
in the Fertile Crescent often involved irrigation and other practices requiring collective
work e¤orts. The needs for village planning, social order, dispute settlement and work
coordination naturally led to the rise of central authorities, organized religion, and social
hierarchy.
12
Indeed, Olsson and Paik (2015) demonstrate that these collectivist cultural norms are strong even
today in regions that made an early transition to agriculture.
13
Sokolo¤ and Engerman (2000) argue that the large-scale plantation system in Tropical America with
labor-intensive crops like sugar led to a fundamentally di¤erent development trajectory than in colonies in
North America, where plantations were less feasible. More recently, Bentzen et al (2015) have shown that
agriculture based on irrigation is strongly associated with autocratic rule across the world. See also Jones
(1981) who makes a related point regarding the advantages of rain-fed agriculture. Talhelm et al (2014)
demonstrate that even within present-day China, individuals in regions that produce paddy rice have
stronger collectivist values than people in wheat-growing regions. Unlike wheat, paddy rice production
requires both irrigation and substantial coordinated labor input. Lastly, Hansen et al (2013) show that
the Neolithic revolution had a negative in‡uence on female empowerment.
8
Although we are not aware of any study that explicitly links collectivism with the
natural exposure to enemies, we argue that the early farming communities in the Fertile
Crescent and also in Europe most likely were repeatedly attacked by nomadic huntergatherers. In later periods, when farming communities had developed into city states or
empires, it is well known that attacks by primitive "barbarians" constituted a constant
threat and quite often led to the collapse of civilizations (Morris, 2010). This threat contributed to the emergence of specialized warriors, defensive public goods such as walls,
the selective location of villages on easily protected (but naturally isolated) locations such
as hills, and the rise of military organization. This militarization of society provided yet
another breeding ground for collectivist norms such as obedience and in-group orientation.
We argue that the threat of predation, the interdependent nature of agricultural production, and the high pathogen prevalence all contributed to the emergence of collectivism
in areas that made an early transition to agriculture. The corollary of this hypothesis
is that in areas that made a late transition to agriculture, these three forces behind the
emergence of collectivism should be weaker and the comparative advantage of individualist
norms should be stronger.
A potential objection might be that even if regions that made an early transition to
agriculture were exposed to a collectivist culture for a longer time, why would not people in
the periphery develop the same collectivist tendencies once they were exposed to parasites,
labor-intensive agricultural production, and predation? After all, the population in current
Sweden adopted agriculture in 5500 years BP and should have had plenty of time to develop
collectivist norms. First, it is clearly the case that a late transition to agriculture is
correlated with certain geographical features such as a temperate climate and a peripheral
location in Western Eurasia. As discussed further in the empirical framework below, we
recognize that although the agricultural revolution was in most places exogenously imposed
by migrants from the southeast, several geographic factors might explain a lower relative
potential of agriculture in northern Europe.14 Not only did agriculture arrive late in
northern Europe, it also relied more on rain-fed production rather than irrigation. Even
with agricultural production, the climate was less conducive to the survival of harmful
microbes. The peripheral geographical location of, for instance, Scandinavia meant that
it was naturally less exposed to roaming armies throughout history than Mesopotamia.
These factors are potential threats to the identi…cation of the true causal e¤ects of the
timing of the agricultural transition on later development.
However, even after taking these in‡uences into account, we propose that individuals
in regions that made a late transition to agriculture should have a strong bias towards
individualism. As developed in further detail in a Malthusian growth framework in Olsson
and Paik (2015), there are strong reasons to believe that the individual migrants from
the Fertile Crescent who spread the use of agriculture into Europe, North Africa, and
West Asia, were people with individualistic norms who were not happy with the social
order in the core. When new farming settlements had been established outside the core,
14
See for instance the analysis of the role of geography in Olsson and Hibbs (2005).
9
the strong selective pressures from parasites, production and predation most likely set in
and made even these pioneer societies more collectivistic. However, since they were from
the beginning a non-representative group with a bias towards invidualism, the selective
forces towards collectivism were not as strong as with people in the core. Furthermore,
the individuals with the strongest individualist norms in the new communities would soon
push the agricultural frontier even further north and establish new societies. In this
way, a process of repeated self-selection into migration would imply that the farmers who
eventually reached northern Europe had a very strong bias towards individualistic norms.
2.2.3
Culture, institutions, and economic performance
In Figure 2, the causal arrow (1c) in our model posits that collectivist cultural norms imply
a stronger tendency for public ownership, imitation, and extractive, autocratic institutions,
whereas individualistic cultural norms are associated with private property rights, innovation and inclusive institutions.15 The relationship between culture and institutions is a
very extensive research area in economics and there is a general agreement that cultural
norms and formal institutions are strongly interrelated.16 For instance, Tabellini (2010)
shows that norms in Europe have historical roots going back hundreds of years. Gorodnichenko and Roland (2013) argue in a theoretical model that democratizations in modern
times should be facilitated by an individualistic culture and show empirically that this is
indeed the case.17 Thornhill and Fincher (2013) document a strong positive correlation
between individualism on the one hand and democracy and innovation on the other.
Gorodnichenko and Roland (2010) is perhaps the most ambitious attempt to examine
the links between collectivism-individualism and economic performance. In a model, the
authors argue that collectivistic norms might facilitate collective action whereas individualism provides intrinsic motivation for innovation.18 They then show that individualism
has a positive impact on economic performance, even after controlling for some institutional variables.19 Borcan et al (2015) suggest a mechanism whereby the emergence of a
15
We de…ne extractive and inclusive institutions in the same way as Acemoglu and Robinson (2012).
Extractive institutions are thus de facto or de jure rules that guarantee political or economic power to be
held by a small elite whereas inclusive institutions allow a broader sharing of power and strong private
property rights. Extractive institutions are associated with autocratic leadership and inclusive institutions
with democractic leadership.
16
North (1990) provides an often cited distinction between formal and informal institutions (culture).
See Alesina and Giuliano (2014) for a recent overview of this literature.
17
As an instrument for collectivist cultural norms, the authors use data on the historical prevalence of
disease from Fincher et al (2008). Using a sample of 74 countries, the …rst stage relationship between
pathogen prevalence and collectivism is very strong (Gorodnichenko and Roland, 2013, Table 1).
18
Ashraf and Galor (2011b) present theory and evidence proposing that conformist cultural norms,
prevalent in naturally isolated regions, might have enhanced the intergenerational transmission of regionspeci…c human capital, which in turn was favorable to economic development in the agricultural era. In
the industrial era, conformism inhibited the adoption of the new technological paradigm, which in turn
implied that these regions eventually stagnated. Lagerlöf (2014) presents a model where investments
in extractive capacity in the early civilizations eventually impeded a transition to democracy, once a
"democratic window" opened.
19
See also Gorodnichenko and Roland (2011) for further tests of the relationship between cultural norms
and income levels.
10
state in the early civilizations should initially have boosted the capacity to collect taxes,
…nance public goods and improve productivity. However, with the passage of time, these
early states became more and more extractive and eventually lost most of the capacity to
transform tax revenue into public goods. This in turn led to a gradual stagnation, which
is also observed in their data.20
Apart from these works, there are numerous well-known contributions that have attempted to identify a causal relationship between inclusive institutions and economic
growth (link 2d in Figure 2). Our emphasis on inclusive institutions as a central factor for
understanding comparative levels of economic development since 1500 CE, is shared by
many authors in the literature. Continuing in the tradition of North (1990), Acemoglu et
al (2002, 2005, 2012) have in a series of works shown that e¤ective constraints on rulers
and freedom from expropriation have been crucial for the individual’s willingness to invest
and to pursue productive activities. Extractive institutions, favoring a narrow elite group
in power, can be associated with temporary spells of economic progress that are not sustainable in the long run. Only in countries with inclusive, democratic institutions will the
incentives for broad segments of society be aligned with fostering economic growth based
on innovation and long-term contracts.21
It goes beyond the scope of this paper to try to demonstrate empirically the full
set of proposed relationships in Figure 2. On the basis of the model above, our focus
will be on establishing a reduced-form relationship between the timing of the agricultural
transition and economic performance with inclusive institutions (democracy) as a potential
intermediate channel.
2.3
Empirical framework
In this section, we brie‡y discuss how we operationalize the testing of the main predictions
from the general model above. Our basic empirical model is shown in Figure 3. The …gure
shows historical time on the horizontal axis combined with a proposed causal chain of
time-invariant and historical variables.
Figure 3
Exogenous factors Xi such as geography, biogeography and climate are shown on top
of the graph because they are not generally functions of time but still a¤ect historical
developments in various ways. The …rst link to the left in the …gure posits that these
exogenous factors played an important role in the timing of the origins and spread of
Neolithic agriculture, happening in the span of 9700-5600 BP in the Western sample
of countries. This relationship has a central role in Diamond’s (1997) theory, in which
20
In Appendix E, we make an in depth comparison of the histories of Iraq and Sweden, which provides
a more detailed illustration of several of the di¤erences between areas with early and late transitions to
agriculture.
21
Other authors have shown that states with a greater degree of inclusivity were also more capable of
raising sovereign debt (Stasavage, 2011) and to conscript soldiers for successful wars (Tilly, 1990).
11
speci…c geographical and biogeographical characteristics implied that the …rst agricultural
transition happened in the Fertile Crescent.
Once the agricultural technology had taken root in the Fertile Crescent, it gradually
spread from its original area. Most recent authors tend to emphasize the role of a demic
di¤ usion process, i.e. agriculture spread through migration of farmers from the original
area towards all directions where there was habitable land (Ammerman and Cavalli-Sforza
(1984). In this di¤usion process, the geographical distance Di to the agricultural cradle
region was an important determinant of the time of the transition of area i, as suggested in
Figure 3. Furthermore, as has been shown by Ashraf and Michalopoulos (2011), exogenous
geographical factors Xi such as climate variation played a role for the timing of transition
even when distance to origin was the same.
In Figure 4 and in Appendix B, Table B1 of this paper, we document that the time since
transition to agriculture for country i or region j is negatively associated with distance to
the Fertile Crescent. Furthermore, time since transition has a negative relationship with
latitude and is positively associated with longitude. This is not at all surprising since,
as will be shown further below, agriculture spread through Europe mainly in a linear,
northwestern direction.
The horizontal set of arrows in Figure 3 contain our main hypothesis, which in turn
follows directly from the general framework in Figure 2: That the time since the Neolithic
transition to agriculture Ti has a causal negative relationship with current levels of income
Yi . The e¤ect is indirect and runs through the historical strength of collectivist and
individualist norms. While these norms are unfortunately not observable, we do have some
data on institutional quality during the last few centuries (referred to as Zi in the …gure).
Zi , in turn, had a lasting direct in‡uence on current comparative levels of development.
It should be noted that the interpretations of these main relationships are confounded
by the fact that exogenous factors such as climate (Xi in the …gure) also likely have a
direct e¤ect on the 2005 levels of income per capita, as displayed in Figure 3, apart from
in‡uencing the time of the transition. Moreover, we recognize the importance of numerous
other historical events (referred to as Hi ) that are unrelated to the agricultural transition
and the rise of collectivist or individualist culture. Examples of such events include the
impact of Roman rule, the Mongol invasion, and the opening of Atlantic trade around 1500
CE. Atlantic trade, for instance, had a direct e¤ect on both institutions and possibly on
economic performance.22 As speci…ed below, we will attempt to control for the in‡uences
of both Xi and Hi .
In this article, our focus will be on documenting a reduced-form relationship between
time since agricultural transition and current income, controlling for exogenous and other
historical factors, on a cross-country, a cross-regional, and a within-country level of analysis. This Western reversal is a striking fact in itself that has not previously been recognized
in the literature. We will also pay some attention to the issue of when the reversal actually
happened. We further test our hypothesis that observable inclusive institutions constitutes
22
See Acemoglu et al (2005) for an extensive investigation of this theory.
12
an important intermediate channel of causation between the agricultural transition and
current income levels.
3
Data
The key explanatory variable in our empirical analysis is the time since the Neolithic
transition.23 A basic building block in our empirical analysis is that agriculture in the
Western core area - based on a composite package of wheat, barley, goats and pigs originated in the Fertile Crescent around 12,000 BP and then gradually spread from there
northwestwards towards Europe and eastwards to Iran, Pakistan and the Indus valley.
According to a dominant tradition in archaeology, this process mainly had the character
of a demic di¤ usion whereby migrants from the Fertile Crescent colonized lands further
and further away from the original center (Bellwood, 2005; Pinhasi et al, 2005).
We develop two new measures for regions as well as for countries in the Western zone:
Average time since agricultural transition and Predicted time since agricultural transition.
The construction of these variables is brie‡y explained below, but we start by discussing
the characteristics of the raw data in Pinhasi et al (2005).
3.1
The Pinhasi et al (2005) data
For the construction of both our measures, we use a sample of calibrated C14-dates from
Neolithic sites in the Near East and Europe available from Pinhasi et al (2005). The data
contains a full list of excavation sites (765 in total) that spans from the Fertile Crescent
to Northwest Europe; the list includes the location coordinates as well as calibrated C14dates and standard deviations estimated for each site. The oldest site in the sample is
M’lefaat, near Mosul in Northern Iraq, dating back 12,811 years (see Table A1 in Appendix
A).
What is the quality of this data? In Appendix A, we analyze several dimensions of
the quality of the Pinhasi et al (2005) data. In brief, …ve issues will be brought up here.
First, the accuracy of measurement would have been higher if the dating of all sites had
been obtained through the highest standard of radiocarbon age determination (accelerator
mass spectrometry), which is not the case. As discussed by Pinhasi et al (2005), there is
a standard tradeo¤ to be made between many observations with somewhat lower quality
or fewer observations with higher quality. Second, the precision of estimates is typically
quite high. The standard deviations of the 765 observations range between 28-291 years.
Third, the distribution of sites across Western countries is uneven. More than 20 countries
have no site observation at all within their boundaries whereas the two countries with the
greatest numbers, United Kingdom and France, have 126 and 108 respectively.24 Figure
23
See Appendix A for a more comprehensive presentation of the data and Appendix D for de…nitions
and descriptive statistics for the variables used in the cross-country and cross-regional analyses.
24
As will be shown below, this is not necessarily a problem for assessing the approximate date of transition
for such a country as long as the dating of nearby sites is accurate.
13
A2 in Appendix A shows a map of the geographical distribution of the sites.
Fourth, certain geographic and economic factors in‡uence the number of sites in each
country. As shown in Table A2 in Appendix A, a regression analysis with the logged
number of sites in a country as the dependent variable shows that more sites are generally
found in countries close to the Fertile Crescent that have a large area, arable soils, and
a high GDP per capita. Fifth, the frequency distribution in Figure A1 in Appendix A
shows an interesting pattern of two major waves of colonization. The clearest peak is
the expansion of the so called Linear Bandkeramik-culture (LBK) around 7400 BP into
Central Europe and Southern Germany.
3.2
Average time since agricultural transition
Our main measure is called Average time since agricultural transition. For the construction of this variable, we use the location and dating of the individual sites to create a
comprehensive map of the date of transition for each grid cell in the Western core. The
method that we employ is Inverse Distance Weighted Interpolation (IDW) in ArcGIS.
The …rst step of calculating the average adoption date for a given region is interpolating
observed points of archaeological sites. For each cell S0 on the map, the formula used in
interpolation is as follows (Johnston et al, 2003):
Y^ (S0 ) =
N
X
iY
(Si )
i=1
In this expression, Y^ (S0 ) is the predicted adoption date for location S0 , N is the number
of measured sample points surrounding the prediction location that will be used in the
prediction,25
i
are the weights assigned to each measured point, and Y (Si ) is the observed
value at the location Si , one of the known archaeological sites.
The formula to determine the weights is:
i
=
di0p =
N
X
di0p
where
i=1
N
X
i
=1
i=1
The variable di0 is the air distance between location S0 and archaeological sites, Si .
The power parameter p is determined by minimizing the root-mean-square prediction error
(RMSPE) in the Geostatistical Analyst tool of ArcGIS.26 In determining the adoption
date of a given area, the IDW methodology implicitly assumes that the archaeological
site closest to the area gives best information on the approximate date of agricultural
adoption. Given that the IDW method calculates the interpolated adoption date for every
cell, the average adoption date for each subnational unit of analysis is simply calculated
25
In our study, N is set to 15.
As explained in Johnston et al (2003), each observed point is removed and compared to the predicted
value for that location. The RMSPE is the summary statistic of the error of the prediction surface. The
Geostatistical Analyst in ArcGIS tries several di¤erent powers for IDW to identify the power that produces
the minimum RMSPE.
26
14
as the average of all the estimated adoption date of location cells within that subnational
unit.2728
Taking the year 2000 as the benchmark year, the mean date of transition in years
before present (BP) in the cross-country sample is 7,611 years, the minimum is 5,608
(Denmark) and the maximum 9,743 (Syria). The mean time since agricultural transition
in our cross-regional sample of NUTS2 regions is 7,050 years with a range from 5,598 to
10,290. This translates into a mean adoption date of 5,050 BCE and a …rst adoption date
of 8,290 BCE.
Most cross-country studies that include the time since Neolithic transition as a variable have so far used the cross-country data set in Putterman (2006). For each country,
Putterman (2006) determines a date of transition by using the …rst attested date of Neolithic agriculture within the country’s borders as stated by various specialized sources.
We believe that our new methodology o¤ers several advantages as compared to Putterman
(2006). As far as we know, the data in Pinhasi et al (2005) o¤er the most recent and most
comprehensive compilation of transition dates for the Western region. Furthermore, our
methodology provides the average date of transition for a country rather than the …rst
date of transition, as in Putterman (2006).29 We believe that this practice will more accurately re‡ect the transition for the whole country since there may be large discrepancies in
dates of transition between regions within countries, as also acknowledged by Putterman
(2006). With our methodology, it is further possible to determine transition dates on a
much …ner geographical level.
3.3
Predicted time since agricultural transition
In order to further investigate the robustness of our results, we construct another variable
called Predicted time since agricultural transition. This variable is also based on the
location and dating of the 765 sites in the Pinhasi et al (2005) sample and exploits the
observed pattern of a fairly stable speed of agricultural di¤usion across space from the
area of origin in the Fertile Crescent. A similar method is used in the analysis in Pinhasi
et al’s (2005) article.
The construction of this variable …rst relies on the identi…cation of a center of agricultural origin. In line with the famous study by Ammerman and Cavalli-Sforza (1984),
we use Jericho as the center of di¤usion and then calculate the shortest air distance from
Jericho to each archaeological site in our sample, using the Great Circle Formula.30 More
27
Another class of interpolation techniques, often known as kriging, uses geostatistical properties. Kriging
relies on autocorrelation as a function of distance and assumes that the data comes from a stationary
stochastic process. Given the terrain variation and boundaries, however, the spread of agricultural adoption
does not appear to satisfy this assumption. Pinhasi et al (2005) …nds that for Eurasia the agricultural
adoption date in an area can be well approximated as a linear function of the distance from the origin in
the Fertile Crescent. The correlation between IDW and kriging estimates nevertheless remain very high
(0.9729) and have little impact on the …nal result.
28
Figure A3 in the Appendix shows an example of IDW methodology applied to the 78 Neolithic sites
within Italy.
29
We average over the calculated scores for all the cells within each country to get the country score.
30
In their investigation of the most likely centers of original domestication, Pinhasi et al (2005) …nd that
15
speci…cally, we run a regression of the form
Ts = ^ 0 + ^ 1 Ds + ^s
(1)
where Ts is the observed time since agricultural transition at site s, Ds is air distance
in kms from site s to Jericho, ^s is a random error term, ^ 0 is a constant and ^ 1 is
the regression coe¢ cient capturing the marginal e¤ect of distance from Jericho on time
since transition. The prediction is ^ 1 < 0, i.e. the longer the distance from the origin of
agriculture, the shorter the time since transition to agriculture.
Figure 4 shows the relationship between distance from Jericho in kms and the age of
the site in calibrated C14 years (or time since agricultural transition) BP (before present)
for the 765 sites in our sample. Each observation is a circle and the size of each circle
re‡ects the standard deviation in the calibration of the date such that larger circles have
a higher standard deviation.
Figure 4
As the …tted line in Figure 4 shows, there is a strong negative relationship between
the age of the site and the distance to Jericho in kms. The constant in the regression is
^ 0 = 9783:8 years BP and the slope coe¢ cient is ^ 1 =
1.018. The interpretation of the
coe¢ cient is that agriculture spread on average with a speed of approximately 1 km per
year.31
We then use the estimated equation in Figure 3 and the geodesic distances of countries
and regions to Jericho, to calculate the predicted date of agricultural transition Ti =
^ 0 + ^ 1 Di of each geographical unit i in our samples. The correlation of this measure with
Average time since agricultural transition is 0.76 on country level.
4
Empirical strategy
The basic model that guides our empirical strategy is shown in Figure 3. The …rst link
proposes that a vector of various time invariant exogenous factors Xi in country i, as well as
distance to the Fertile Crescent (Jericho) Di , arguably a¤ected the date of transition. For
instance, Diamond (1997) argued that a Mediterranean climate was particularly favorable
for the domestication of heavy grasses like emmer and einkorn wheat.
In line with such a causal e¤ect, we postulate a relationship of the form
Ti =
0
+
1 Di
+ Xi0
2
+
(2)
i
where Ti is the time since the agricultural transition and
i
is a random error term.
the sites Abu Madi (in Egypt) and Cayönu (Turkey) are the most likely centers, although Jericho gives a
very similar score. See Appendix A for further details.
31
There were several periods of more rapid expansion, such as during the LBK culture, as well as periods
of standstill. In the appendix, we discuss the agricultural di¤usion process in more detail.
16
Table B1 in Appendix B shows the results from di¤erent speci…cations with distance,
geographical, biogeographical and climate indicators as independent variables.
As expected, Distance to Jericho has a strong negative impact in Columns (1)-(5) and
so has Latitude. Longitude has a positive but not always signi…cant coe¢ cient. Distance to
Jericho, Latitude and Longitude - i.e. our proxies for how agriculture spread historically
- together explain about 86 percent of the variation in both samples. However, when
we insert several geographical variables from Xi into the regression in Columns (3)-(5),
only Log arable land area, Log distance to coast and river and a dummy for a semi-desert
biogeography in Column (5) have a signi…cant positive impact. The equivalent analysis of
the regional sample is shown in Table B2.32
Is it possible that there were pre-existing cultural di¤erences that sometimes in‡uenced
Ti ? Acemoglu and Robinson (2012), for instance, suggest that the Natu…ans, who lived in
the Middle East during the Mesolithic period, perhaps had adopted a sedentary lifestyle
as a more or less random technological innovation that had not so much to do with
how suitable the wild plants and animals in the neighborhood was for domestication.33
Although we recognize this possibility of an unobservable, omitted variable in equation
(2), we …nd no reason to believe that such variations were systematically related to either
Di or Xi and should not bias the results.
Our main interest will instead be to analyze how the time since agricultural transition
a¤ects current levels of income. In our cross-country analysis, we will run reduced-form
regressions with the straightforward format
Yi =
0
+
1 Ti
+ Xi0
2
+ Hi0
3
+ "i
(3)
where Yi is income per capita in country i, Hi is a vector of other historical developments
and "i is a normally distributed error term. The parameter of interest in this setting is
1,
showing the reduced-form impact of agricultural history on economic performance. Our
key hypothesis is that
1
< 0, i.e. an adverse long-run impact of agricultural experience
on economic prosperity within the Western core.
We know however from equation (2) that Ti is a function of Di and of Xi . By inserting
the expression for Ti above into equation (3), we see that the function can be rewritten as
Yi =
0
+
1 0
+
1 1 Di
Hence, it should be kept in mind that
+ Xi0 (
1
1 2
+
2)
+
1 i
+ Hi0
3
+ "i :
will both capture the impact of Di and that of
Xi , as well as other unobservable factors that a¤ect the time of transition. In addition, it
32
A noteworthy feature of the results in this table is that Distance to Jericho changes its coe¢ cient sign
from negative to positive in Columns (1)-(2) when Latitude and Longitude are added. The most likely
reason for this confounding e¤ect in the case of regional analysis is colinearity: Distance to Jericho and
Longitude have a correlation of -0.88.
33
A similar line of reasoning is also found in Bowles and Choi (2013) who argue that an institutional
innovation of private property in certain societies most likely preceded the transition to agriculture.
17
might be noted that the sum of the indirect and direct e¤ects of Xi is
1 2
+
34
2.
A key identifying assumption in this setup is further that Yi did not in any way
in‡uence Ti . Although local geographical and climatic conditions no doubt played a role
in the di¤usion of agriculture, we can control for a set of these exogenous factors. In
doing so we think there are no good reasons to believe that current or historical levels of
income per capita should have any plausible reverse causality on the date of agricultural
transition.35 The main reason for this is the widespread archaeological and genetic support
for the "Neolithic demic di¤usion model", arguing that agriculture mainly spread through
physical migration by people from the original farming areas in the Fertile Crescent to
areas in the periphery.36 Agriculture thus typically appeared in a region as an exogenous
intervention and did not arise indigenously.
Nonetheless, in order to further alleviate such potential concerns, we make use of the
fact that we have an exogenous source of variation of Ti in the form of Di . This makes
it possible to specify a 2SLS-regression framework where equation (2) is the …rst stage
and equation (3) is the second stage and where Di serves as an instrumental variable.
The exclusion restriction is that Cov ("i ; Di ) = 0, i.e. Di should only a¤ect the outcome
variable through Ti ; we do not …nd a good reason as to why distance to Jericho should
have any direct impact on current income per capita. Hence, we feel comfortable about
using this variable as an instrumental variable. We will present some results using this
strategy below.
As outlined in Figure 3, our hypothesis is that time since transition to agriculture
should mainly a¤ect current economic performance through its e¤ect on unobserved cultural norms and observed inclusive institutions in the form of democracy during the Modern Era. If we refer to such a democracy variable as Zt , then a relevant regression for our
purposes is:
Yi =
Our expectation is that
3
0
+
1 Ti
+ Xi0
2
+
3 Zi
+ "i :
(4)
> 0. Furthermore, our argument is that Ti and Zi are struc-
turally related such that
Zi =
where
1
0
+
1 Ti
+
i
(5)
< 0 implies that an early transition to agriculture retards the adoption of democ-
34
The indirect e¤ect will not be teased out when we use Predicted time since the agricultural transition
Ti = ^ 0 + ^ 1 Di . In that case, 2 should capture both the direct and indirect e¤ects of Xi .
35
One might imagine individual episodes whereby advancing farmers met resistance from particularly
prosperous hunter-gatherers. Bellwood (2005) suggests for instance that the relatively late colonization
of the areas surrounding the Baltic Sea might have been due to the very rich aquatic resources that the
local inhabitants had access to. This implies that the distance to coast would be an important factor in
explaining the lifestyle of these inhabitants. The marine resources would be an important part of the diet
even after farmers had come to dominate the area.
36
See Ammerman and Cavalli-Sforza (1984) and Bellwood (2005) for thorough accounts of this theory.
Demic di¤usion of agriculture is often discussed in relation to "cultural di¤usion" whereby agriculture
proposedly spreads through the di¤usion of technology rather than through migration.
18
racy. In order to assess whether Zi is the true intermediate variable, we will estimate both
1
and
1.
If the estimate
1
< 0 is signi…cant whereas
1
= 0 when Ti and Zi are both
included as in Equation (4), then we interpret these results as lending support to that Zi
is indeed the true intermediate variable.
In the cross-regional analysis, the estimated equation is
Yij =
0
+
1 Tij
0
+ Xij
2
+ Hij0
3
+ fi +
ij
where Yij is income per capita in region j of country i, Tij is the time since agricultural
transition for the region, Xij is a vector of region-speci…c exogenous variables, Hij is a
vector of region-speci…c historical variables, and
ij
is an error term. The key di¤erence
from the cross-country analysis is the inclusion of the country …xed e¤ect fi . The key
parameter of interest here is also
5
1,
which we expect to be negative.
Results
5.1
Cross-country analysis
Table 1 shows the results associated with our reduced-form estimation of equation (3).
In these speci…cations, our two main measures of time since agricultural transition are
included as independent variables alongside various exogenous measures of geography and
climate. The base sample includes around 60 Western countries in Europe, Middle East,
North Africa, and Southwestern Asia that belonged to the Western core of agricultural
di¤usion. Conley (1999) standard errors are included in all speci…cations to take into
account potential spatial correlations.
As is clear from the table, Average time since agricultural transition and Predicted
time since agricultural transition in the Western core consistently have a negative and
signi…cant impact on log income per capita in 2005. Without control variables in Column
(1), the estimate is -0.492 and strongly signi…cant.
Table 1
Studies within the tradition of long-run development typically include geographical
control variables together with their main variable of interest. However, no consensus
has yet emerged on exactly what variables to include, and theory does not provide any
useful guide. In Column (2), we use the same set of geographical controls as in Ashraf
and Galor (2013). Apart from Log latitude, this set includes a measure of land suitability
for agriculture and Log arable land area. As usual in this kind of studies, Log latitude
is positive and signi…cant, implying that northern countries tend to be richer. Perhaps
surprisingly, Log arable land area is consistently negative and signi…cant. This suggests
another negative e¤ect of agriculture since countries with good agricultural lands are now
relatively poor.
19
However, Log land suitability for agriculture and Log arable land area turn out to be
strongly correlated (0.90) and inclusion of the land suitability variable reduces the number
of observations to 55. Furthermore, although latitude levels is a strong determinant of the
time since agricultural transition, as demonstrated above, it is less clear what latitude
captures when income per capita is the dependent variable. We do not believe that a
northern location per se is an advantage for economic development. Rather we think
of latitude as being correlated with other geographical factors that potentially have a
true causal e¤ect. In the following analysis, we therefore drop latitude from our set of
geographical controls.
In Column (3), we instead introduce Log distance to coast and river. Given the historical importance of waterborne trade in Europe, we argue that this variable should be
a good proxy for trade potential. As the table shows, Log distance to coast or river is
consistently negative in all speci…cations, as one might expect. Our two main geographical
controls are therefore henceforth Log arable land area and Log distance to coast and river.
Figure 5 shows the conditional relationship between Log GDP per capita in 2005 and
Average time since agricultural transition when controlling for the two geographical variables. The point estimate (-0.485) means that a 1,000-year earlier transition to agriculture
is associated with a 48.5 percent lower GDP per capita. The numbers imply that if a country close to the mean time since transition such as Italy had experienced the transition to
agriculture in 8,300 BP instead of in the actual year 7,300 BP, their GDP per capita in
2005 would have been 9,984 $US instead of 19,386 $US. The economic signi…cance is thus
also quite strong.
Figure 5
Inclusion of other geographical controls such as Temperature and Precipitation in Column (5) brings down the level of the point estimate for Average time since agricultural
transition but signi…cance is still retained. In Column (6), we include a variable Migratory
distance from Addis Abeba which Ashraf and Galor (2013) use as an instrumental variable
for genetic diversity. The variable does not have a signi…cant estimate. The results are
not qualitatively di¤erent when we use Predicted rather than Average time in Columns
(8)-(9).
In Table 2, we check the robustness of these results in various ways while controlling
for our two geographical variables. Our …rst strategy for robustness analysis is to vary
the samples. Our …rst question to the data concerns the countries in the Fertile Crescent.
Perhaps these "founder countries" are fundamentally di¤erent in some way and drive the
results? In Column (1), we therefore include a Fertile Crescent-dummy and in Column
(2) we exclude the Fertile Crescent countries altogether. The result of this exercise is that
the estimate for Average time since agricultural transition becomes more negatively sloped
and even more signi…cant. In Column (3), we include two continental dummies for Europe
and Southwest Asia. The Europe dummy is positive and signi…cant, as one might expect.
In Columns (4)-(5), we restrict the sample to only European and only former Roman
20
countries respectively. The latter countries arguably shared a similar institutional and
cultural context during a very formative period of Western history. Even in these limited
samples and when controlling for the standard geographical and historical variables, time
since agricultural transition is still signi…cant. In Column (6), we restrict the sample to
only those countries with at least one Pinhasi site within their borders. The basic result
stays the same.
Table 2
Our second robustness strategy is to replace Average time with three other measures:
Predicted time, as used above, as well as a variable Earliest date of transition and Putterman’s Time since agricultural transition. Earliest date uses information on the earliest
incidence of agriculture within a country rather than the average date of transition. It
could be argued that the earliest date is more in line with the method used in Putterman
(2006) whose data provide the basis for the empirical results in several papers. Although
the estimates in (9)-(10) are closer to zero, the negative slope coe¢ cients are signi…cant
in both cases.
A third robustness strategy addresses concerns of a potential measurement problems
and reverse causality. Could it be the case that Yi in some way in‡uenced the level of
Ti ? Although we …nd that unlikely, another concern might be biases due to measurement
error. Both types of problems can be alleviated by using an IV-approach.
In Table 3, we introduce Distance to Jericho as an instrumental variable for Average
time since agricultural transition. Informally, Column (2) in Panel A suggests that Distance to Jericho does not in‡uence our outcome variable when included alongside Average
time. Columns (3)-(4) in Panel A then presents the coe¢ cients from the …rst stage of
the IV-2SLS estimation, with and without our vector of controls Xi . As the numbers
suggest, the …rst stage relationship is very strong. Also in the second stage, the negative
coe¢ cients for the instrumented Average time since agricultural transition are strongly
signi…cant and at levels quite similar to the equivalent OLS estimates in Table 1, Columns
(1) and (3).
In Table 4, we check the robustness of our main results to inclusion of a list of other historical variables Hi that have been proposed/used in the literature, while also controlling
for geography. In terms of Figure 3, what we are checking for is alternative intermediate
historical processes that might a¤ect both inclusive institutions and the dependent variable. We argue that if the inclusion of any of these alternative historical variables makes
1
in equation (4) insigni…cant or close to zero, this might be interpreted as a sign that the
variable in question is actually an intermediate variable.37 The estimates of the historical
variables are shown in the second column. For brevity, we have omitted to report standard
errors.
37
Loss of signi…cance for 1 might of course also indicate that the statistical relationship between Ti and
Yi that we have found so far is spurious and only a result of that Ti just happens to be correlated with
the historical variable by chance.
21
Table 4
The …rst variable that we try is Ashraf and Galor’s (2013) Predicted genetic diversity
and Predicted genetic diversity squared. For the world as a whole, this variable was shown
to have an inverted u-shaped relationship with income in the original study. In our Western
sample, the estimates have the wrong sign and are not signi…cant. Ethnic fractionalization
in Row (3) is negative and signi…cant but so is the estimated coe¢ cient for Average time
since agricultural transition.
Perhaps our main variable capturing time since the Neolithic just happens to be correlated with historical Atlantic trade, which is the true driving force behind the rise of
Western Europe, as hypothesized by Acemoglu et al (2005)? In Row (4), we introduce
an Atlantic dummy, which turns out to be strongly positive and signi…cant. However,
Average time since agricultural transition is still negative and signi…cant. An indicator
for the duration of state experience State history 1-1950 is entered in Row (5). Despite
the strong link between time since the Neolithic and subsequent state history, the control
variable does not make Average time since agricultural transition insigni…cant. Hence, the
long-term impact of agriculture does not run solely through the timing of state formation.
In Row (6), we include a whole set of historical empire proxies capturing the proportion
of the country controlled by Roman, Byzantine, Carolingian, and Ottoman rulers, as well
as a dummy whether the country was overrun by Mongols in 1300 CE.38 Having been
included in the Carolingian empire had a clear positive in‡uence while Mongolian rule
had a negative impact but our main coe¢ cient of interest is almost una¤ected.
The Legal Origin-variables in Row (7) are all signi…cant but do not seem to be a very
important intermediate variable. Generalized trust in Row (8), capturing the strength
of trust using the World Value Survey, does however seem to be a potentially important
variable. Its inclusion lowers the estimate for Average time since agricultural transition
considerably and it becomes less signi…cant. We will return to this variable in a section
below.
One obvious potential intermediate variable in our historical framework is religion.
All the old areas of civilization are currently dominated by Muslims whereas Protestants
are in majority in northern Europe. Furthermore, important scholarly works argue that
certain institutional features of Islam were harmful to economic development whereas the
emphasis on the individual within Protestantism has been discussed as a bene…cial factor
for economic development at least since Max Weber.39 When we control for the religion
population percentages in Row (9), the estimate for Average time since agricultural transition falls and becomes insigni…cant. In particular, Protestant population seems to have
the character of an intermediate variable since the estimate is strongly signi…cant. This
…nding goes well with our hypothesis that various institutions conducive to individualism
38
We have coded the latter Mongol variable ourselves. A description for the coding procedure is available
upon request.
39
See for instance Kuran (2011) for a in depth account of how Islam a¤ected economic development in
the Western core and Landes (1998) for a treatment of Protestantism.
22
should be a causal mechanism. We will return to Protestant population further below and
analyze its stength as an intermediate variable in comparison with democracy.
An additional historical shock which has been emphasized by many authors is the Black
Death of 1347-1353 CE (Clark, 2008; Voigtländer and Voth, 2012). Could it have been
an extra high mortality in Black Death that "shocked" the north European countries and
regions into a new equilibrium with higher higher wages and higher urbanization? Any
such analysis is plagued by the fact that cross-country data on mortality from the Black
Death are generally very unreliable. In Figure C5 in Appendix C, we show the relationship
between Avarage time since agricultural transition and Black Death mortality among 53
Western cities, using data from Christakos et al (2005). Interestingly, the relationship
in the …gure is positive, indicating that the cities with a long history of agriculture were
actually worse hit by the Black Death on average. Although this evidence is far from
conclusive, it does not seem to give support to a reversal happening because of a more
adverse shock in north Europe.40
In summary, Table 4 shows that our main variable is robust to the inclusion of a
range of historical variables suggested in the literature. However, inclusion of Muslim and
Protestant populations and Generalized trust all seem to be potential intermediate causal
channels between time since agricultural transition and current incomes. We will return
to those variables further below.
5.2
Cross-regional and within-country analysis
So far, we have shown that the negative relationship between time since the agricultural
transition and current levels of income is strong at the cross-country level. Are our key
…ndings robust to changing the level of aggregation? In this section, we move from the
country level to the regional level.
Table 5 shows the results for the cross-regional analysis among about 280 Western
NUTS2-regions in 30 European countries.41 In this table, we include an even broader set
of geographical control variables capturing size of region area, latitude, altitude, share of
arable land in 1600 CE as well as access to water. The dependent variable is Log GDP
per capita in 2010.
Columns (1)-(2) show the OLS regressions when we do not control for country …xed
e¤ects. The estimate for Average time since agricultural transition is then negative and
signi…cant, as before, even when we control for latitude. However, since we are now
40
An alternative hypothesis, advanced by Borsch (2005), is that although Western countries often experienced fairly similar shocks in terms of mortality rates, it was pre-existing institutional di¤erences between
collectivist and absolutist countries such as Egypt and more individual-oriented countries such as England,
that determined the response to the Black Death. In Mamluk Egypt, coercion became even worse as a
result of the plague whereas rising wages due to labor shortage gave workers in England a stronger social
position than before. One might thus view the Black Death as a key event for the emergence of the reversal
whereby the importance of existing institutional di¤erences, in turn caused by the di¤erent agricultural
histories, was magni…ed and pushed northwestern Europe into a take-o¤.
41
The data on regional GDP levels has been obtained from Eurostat (2012). Please see Appendix D2
for further details about the variables included in the regional analysis.
23
primarily interested in the within-country variation, we include country …xed e¤ects in
Columns (3)-(4). The slope coe¢ cient changes but retains its signi…cance. In Column (4),
R2 is 0.76 which indicates that our speci…cation explains a quite large proportion of the
variation in regional income levels.
Table 5
The results above suggest that there is also a within-country relationship between
time since transition and current income. In order to zoom in even more closely on the
micro level, we also use data for time since transition and average levels of GDP per
capita that we have collected on the NUTS3-level of aggregation. In Figure 6, we show
the unconditional bivariate scatter plot for 1,371 regions with available data. The …gure
also includes regression lines and slope coe¢ cients for the relationship within …ve of the
largest countries. Figure C6 in Appendix C shows the separate scatter plots for these …ve
countries and also for United Kingdom, whereas Figures C3-C4 in Appendix C show maps
of the transition to agriculture and current income levels for 107 Italian regions.
As Figure 6 indicates, there is a very strong negative relationship overall. However,
when we introduce the separate regression lines for …ve of the largest countries in the
sample with many Pinhasi sites within their borders (Germany, France, Italy, Spain, and
Turkey), the picture is somewhat modi…ed. The individual estimates for these countries,
without and with controls, are shown in Table 6. The relationships within Turkey, Spain,
and Italy remain negative and signi…cant even when adding geographical controls. The
unconditional regression coe¢ cient for Germany is positive and signi…cant but changes
sign and becomes negative and signi…cant at the 0.1-level when we add controls.42
Table 6
Figure 6
Our interpretation of these results is that there is indeed a strong negative relationship
between Average time since agricultural transition and GDP per capita both across and
within countries. The strong negative correlations on a disaggregated level within countries
like Turkey, Spain and Italy are particularly striking. A more comprehensive analysis of
these intra-state reversals is beyond the scope of this paper, and remains a potential issue
for future research.
5.3
Timing of the reversal
When did this reversal happen in the Western world? There are at least two strands in
the literature arguing for a great divergence happening in the Western world after 1500
CE. Acemoglu et al (2005) demonstrate that Atlantic trading nations, bene…tting from
42
For Germany, the situation is complicated by the recent merge of East and West Germany. When
we also include a former communist dummy to the controls in Table 6, there is neither a positive nor a
negative relationship within Germany.
24
the newly opened colonial trade, surged ahead of other countries and regions in Europe
after 1500 CE. The second and very signi…cant shift was the Industrial Revolution, starting
around 1800, which even further contributed to the economic and technological dominance
of Britain and other north European countries (Mokyr, 1990; Acemoglu and Robinson,
2012). A key question in our analysis is thus if the reversal since the Neolithic was visible
already during the First Great Divergence 1500-1800 CE?
For the Malthusian era, it is widely accepted that population density is a better indicator than GDP per capita for the level of economic development in a country (Ashraf
and Galor, 2011a). In Table 7, Columns (1)-(3), we use population density in years 1,
1000 and 1500 CE as dependent variables. In year 1, the reversal does not seem to be in
place since the estimate for Average time since agricultural transition is positive. In 1000
CE, the estimate is negative but insigni…cant. However, by 1500, the reversal is evident
since time since agricultural transition in that year is both negative and signi…cant.
Table 7
In Columns (4)-(7), we instead use Log GDP per capita from 1500 and 1820 CE as
the dependent variables.43 In (5) and (7), we also include our geographical controls and
the sample is now limited to only 21-28 countries. The estimate for Average time since
agricultural transition is negative but non-robust in 1500 but becomes strongly signi…cant
by 1820.
When we use income per capita data also for years 1000, 1600, 1700, 1913, 1980, and
2005 and run the same conditional bivariate regressions as in Table 7, we can track more in
detail how the regression coe¢ cient of Average time since agricultural transition develops
over time. Figure 7 shows the patterns, suggesting a stronger and stronger tendency for a
reversal over the course of the millennium.
Figure 7
We do not interpret these results as evidence against the critical roles played by Atlantic
trade, the Reformation, and the Industrial Revolution. Undoubtedly, these fundamental
processes contributed very importantly to the great divergence that made Britain and its
northern followers so much richer than the Mediterranean and Middle Eastern countries.
What we do suggest, however, is that the process of reversing fortunes in the Western world
seems to have followed an older trajectory rooted in the Neolithic which has previously
not been recognized and which seems to have become manifest already during the period
after 1500.
5.4
Causal mechanisms
So far, our focus has been on documenting a negative relationship between time since
agricultural transition and current income per capita. Our model in Figure 2 speci…es the
43
Detailed variable sources and data descriptions are available in Appendix D.
25
proposed causal mechanisms behind this reduced-form relationship. In this …nal section,
we will analyze whether our data support the proposed linkages with two of the key
elements of our model that are observable: parasite stress and inclusive institutions.
When we studied the timing of the reversal in the previous section, the results suggested
that the negative impact of time since agriculture seemed to have emerged sometime during
the last 500 years. In Table 8a, we therefore employ our variables Average time since
agricultural transition in 1500 and Log GDP per capita in 1500 together with a proxy for
the level of executive constraints in 1500. The variable is mainly taken from Acemoglu
et al (2005) but is complemented with information for the Ottoman countries during the
period.44 A high score indicates stronger constraints against the executive (and hence
more inclusive institutions).
Table 8a
In line with our hypothesis, Columns (1)-(2) con…rm that there indeed appears to
be a strong relationship between the timing of the agricultural transition and Executive
constraints in 1500 such that countries that adopted agriculture early had less inclusive
institutions already by 1500. Furthermore, executive constraints are positively correlated
with GDP per capita in 1500. When we run a ’horse race’ between Average time since
agricultural transition and Executive constraints for 21 countries with available data in
Columns (5)-(6), it is clear that our proxy for democracy is still positive and signi…cant
whereas the time since transition has no explanatory power left. The conditional scatter
plots associated with Column (5) are shown in Figure C7 in Appendix C. Our interpretation of the results in Table 8a is that they are consistent with the hypothesis that a long
agricultural history provided an obstacle to institutional development already in 1500 and
that the constraint against the executive was an important intermediate channel between
the agricultural transition and economic development, as suggested by our theory.
However, given various issues such as the limited number of observations and the
quality of the data from 1500, we believe the results from Table 8a should be interpreted
with caution. In order to obtain a more reliable estimate of historical inclusive institutions,
we also employ Democracy stock 1900-2000 from Gerring et al (2005). As described in
Appendix D1, the authors use individual country-year scores for the standard Polity2 variable over the 1900-2000 period. Each annual observation for year s is discounted by
0:992000
s
so that more recent years have a greater weight. Polity2 i;s ranges from +10
(full democracy) to -10 (full autocracy). The highest scorer in our Western sample is
Switzerland at 637 (full democracy throughout period) and the lowest is Saudi Arabia
at -604.6. The variable might thus be said to measure a historical accumulated stock of
democracy during the 20th century.45 Given that our main dependent variable below is
44
See Appendix D1 for a description of the historical variables used in this section.
In their original study, Gerring et al (2005) argue that a democracy stock a¤ects economic growth by
facilitating the accumulation of physical, human, social and political capital. In their panel regressions
with economic growth as the dependent variable, they …nd that the lagged accumulated stock of democracy
at di¤erent points in time had a strong positive e¤ect on growth.
45
26
GDP per capita in 2005, the endogeneity problem should be alleviated.46
We also introduce a variable capturing parasite stress; Historical pathogen prevalence
from Murray and Schaller (2010). The variable codes information from historical sources
on the prevalence of nine common infectious diseases, including typhus, malaria, and
tuberculosis. High values on this index means high disease prevalence and it is normalized
on a global scale to have a zero mean. In our sample, the highest scorer with the worst
historical disease environment is India (0.94) whereas the lowest pathogen prevalence is
found in Iceland (-1.19).
Table 8b shows the results. In Columns (1)-(2), Historical pathogen prevalence is
the dependent variable. As hypothesized in Figure 2, link 1a, there is indeed a strong
positive relationship between Average time since agricultural transition and parasite stress.
Remarkably, time since transition alone explains more than 50 percent of the variation in
pathogen prevalence in Column (1). The relationship is signi…cant also after including our
standard geographical controls.
Democracy stock 1900-2000 is the dependent variable in Columns (3)-(4). In line
with our hypothesis of democracy being an important intermediate variable, we …nd that
Average time since agricultural transition strongly predicts the level of the democracy
stock such that countries with an early transition tended to be more autocratic during
the 20th century. The estimate is strongly signi…cant both with and without geographical
controls. Figure C8 in Appendix C shows the unconditional scatter plot. As expected,
the Scandinavian countries, Great Britain and Ireland are found in the upper left corner
whereas Egypt, Syria, Jordan and Iraq are found in the lower right corner. The estimate
implies that an earlier transition to agriculture by 1,000 years is associated with a 155.4
units lower democracy stock.
Table 8b
When we include both Average time since agricultural transition and Democracy stock
1900-2000 as regressors in Column (6), Democracy stock is still highly signi…cant. The
conditional scatter plots for these regression are shown in Figure C9 in Appendix C.
In Column (7), we also add Historical pathogen prevalence, which has a negative but
weakly signi…cant relationship to levels of income. An important observation, supporting
our hypothesis that the impact of Average time since agricultural transition should work
strongly through historical levels of disease and democracy, is that the regression coe¢ cient
for Average time is insigni…cant in Columns (6)-(9).
In Column (8), we include the two religious variables related to our democracyindividualism complex discussed earlier; Muslim and Protestant population. Both appeared to be fairly strong intermediate channels in Table 4. However, when we employ
46
We can of course not rule out identi…cation problems associated with this speci…cation. For instance,
it could be argued that income and democracy have had too strong feedback loops throughout the 20th
century for any identi…cation of causal e¤ects beween these two variables to be possible. We believe that
the correlations reported in Table 8b are still useful for a deeper understanding of the historical linkages.
27
them here alongside Democracy stock 1900-2000, Historical pathogen prevalence and the
standard geographical control variables in Column (8), neither of their coe¢ cients are
signi…cant whereas the democracy estimate is. The same tendency remains when we also
include Generalized trust in Column (9). Most likely, this is due to the fact that both
the share of a Protestant population and the level of generalized trust are strongly related
to norms in society that are highly associated with individualist cultural norms, which in
turn are strongly related to democracy.
In summary, our interpretation of these results is that in particular historical levels of
inclusive institutions in the form of democracy, but also historical levels of parasite stress,
are important intermediate variables in the causal chain between time since agricultural
transition and current prosperity, consistent with our model of a long-run development
reversal. We do not claim, however, that disease, culture, and democracy are necessarily
the only channels. Further research needs to be carried out before we can reach conclusive
evidence in this regard.
6
Conclusions
In this paper, we document a historical reversal of fortune within the Western agricultural
core region in the sense that countries that adopted agriculture and complex civilizations
early in history tend to be poorer today than countries in the periphery of the core that
adopted agriculture late. We further demonstrate that there is a strong negative association between time since agricultural transition and current income levels also within
Western countries with a particularly strong relationship in Italy, Spain and Turkey. Importantly, the economic reversal appears to have started to emerge already by 1500 CE,
i.e. before Western colonization and industrialization, and then grew stronger over time.
Hence, the very large income di¤erences that emerged during the last 500 years seem to
be partly rooted in development trajectories dating back to the Neolithic.
While we do not rule out other related channels that have played crucial roles according
to the existing literature, we also show in this paper that historical pathogen prevalence and
inclusive institutions such as democracy are key intermediate channels between the time
of agricultural transition and current income levels. Using standard regression techniques,
we …nd that an early transition to agriculture strongly predicts low levels of democracy
already by 1500 CE and that this relationship persists when we use composite data from
the 20th century. However, on the basis of our results, we do not rule out that other
related channels could also have played a role. We believe that our model of long-run
development, as well as our empirical results, might serve as a reference point for further
future research on the exact mechanisms that generated the Western reversal.
References
[1] Acemoglu, D., S. Johnson and J. Robinson (2002) "Reversal of Fortune: Geography and
28
Institutions in the Making of the Modern World Income Distribution" Quarterly Journal
of Economics 117(4), 1231-1294.
[2] Acemoglu, D. et al (2005) "The Rise of Europe: Atlantic Trade, Institutional Change, and
Economic Growth" American Economic Review 95(3), 546-597.
[3] Acemoglu, D. and J. Robinson (2012) Why Nations Fail: The Origins of Power, Prosperity
and Poverty, New York: Crown Publishers.
[4] Alesina, A. and P. Giuliano (2014) "Culture and Institutions" mimeo, Harvard University.
[5] Ammerman, A. and L. Cavalli-Sforza (1984) The Neolithic Transition and the Genetics of
Populations in Europe, Princeton University Press.
[6] Ashraf, Q, O. Galor, and Ö. Özak (2010) "Isolation and Development" Journal of the European Economic Association 8(2-3), 401-412.
[7] Ashraf, Q. and O. Galor (2011a) "Dynamics and Stagnation in the Malthusian Epoch" American Economic Review 101(5), 2003-2041.
[8] Ashraf, Q. and O. Galor (2011b) "Cultural Diversity, Geographical Isolation, and the Origin
of the Wealth of Nations" NBER Working Paper 17640.
[9] Ashraf, Q. and O. Galor (2013) "The ’Out of Africa’Hypothesis, Human Genetic Diversity,
and Comparative Economic Development" American Economic Review 103(1), 1-46.
[10] Ashraf, Q. and S. Michalopoulos (2011) "The Climatic Origins of the Neolithic Revolution:
Theory and Evidence" working paper.
[11] Belloc, M. and S. Bowles (2013) "The Persistence of Inferior Cultural-Institutional Conventions" American Economic Review: Papers and Proceedings, 103(3), 1-7.
[12] Bellwood, P. (2005) First Farmers: The Origins of Agricultural Societies, Oxford: Blackwell
Publishers.
[13] Bentzen, J.S., N. Kaarsen, and A.M. Wingender (2012) "Irrigation and Autocracy" University
of Copenhagen, working paper.
[14] Blaydes, L. and E. Chaney (2013) "The Feudal Revolution and Europe’s Rise: Political
Divergence of the Christian and Muslim Worlds before 1500 CE", American Political
Science Review, 107(1),
[15] Bleaney, M. and A. Dimico (2011) "Biogeographical Conditions, the Transition to Agriculture,
and Long-Run Growth" European Economic Review 55(7), 943-954.
[16] Bockstette, V., A. Chanda and L. Putterman (2002) "States and Markets: The Advantage of
an Early Start" Journal of Economic Growth 7(4), 347-369.
[17] Borcan, O., O. Olsson and L. Putterman (2014) "State History and Long-Run Economic
Development: Evidence from Six Millennia" Brown Working Paper 2014-8, Brown University.
[18] Borsch, S. (2005) The Black Death in Egypt and England: A Comparative Study, University
of Texas Press.
[19] Bosker, M., E. Buringh, J. L. van Zanden (2013) "From Baghdad to London: Unraveling
Urban Development in Europe, the Middle East, and North Africa, 800-1800" Review of
Economics and Statistics 95(4), 1418-1437.
[20] Christakos, G., R. Olea, M. Serre, H-L Yu, L-L Wang (2005) Interdisciplinary Public Health
Reasoning and Epidemic Modelling: The Case of Black Death, Berlin: Springer.
29
[21] Clark, G. (2008) A Farewell to Alms: A Brief Economic History of the World, Princeton
University Press.
[22] Cohen, M.N. and G.J. Armelagos (1984)(eds) Paleopathology at the Origins of Agriculture,
Gainesville: University Press of Florida.
[23] Conley, T. (1999) "GMM Estimation with Cross-Sectional Dependence," Journal of Econometrics 92, 1-45.
[24] Diamond, J. (1997) Guns, Germs and Steel: The Fates of Human Societies. New York: Norton.
[25] Eurostat (2012) http://epp.eurostat.ec.europa.eu/portal/page/portal/region_cities/introduction.
[26] Fincher, C.L., R. Thornhill, D.R. Murray, and M. Schaller (2008) "Pathogen Prevalence
Predicts Human Cross-Cultural Variability in Individualism/Collectivism" Proceedings of
the Royal Society B, 275, 1279-1285.
[27] Gerring, J., P. Bond, W. Barndt, and C. Moreno (2005) "Democracy and Economic Growth:
A Historical Perspective" World Politics 57, 323-364.
[28] Gorodnichenko, Y. and G. Roland (2010) "Culture, Institutions and the Wealth of Nations"
NBER Working Paper 16368.
[29] Gorodnichenko, Y. and G. Roland (2011) "Which Dimensions of Culture Matter for Long-Run
Growth?" American Economic Review: Papers and Proceedings 101(3), 492-498.
[30] Gorodnichenko, Y. and G. Roland (2013) "Culture, Institutions and Democratization"
mimeo.
[31] Hansen, C.W., P.S. Jensen, C. Skovsgaard (2013) "Modern Gender Roles and Agricultural
History: The Neolithic Inheritance" University of Southern Denmark, working paper.
[32] Harlan, J.R. (1995) The Living Fields: Our Agricultural Heritage, Cambridge University
Press.
[33] Hibbs, D.A. and O. Olsson (2004) ”Geography, Biogeography, and Why Some Countries Are
Rich and Others Are Poor”Proceedings of the National Academy of Sciences of the USA,
March 9, 101(10), 3715-3720.
[34] Johnston, K., J.M. Ver Hoef, K. Krivoruchko, and N. Lucas (2003) ArcGIS: Using ArcGIS
Geostatistical Analyst. Redlands, CA: ESRI.
[35] Jones, E.L. (1981) The European Miracle: Environments, Economies and Geopolitics in the
History of Europe and Asia. Cambridge: Cambridge University Press.
[36] Kennedy, J.M. (1988) The Rise and Fall of the Great Powers: Economic Change and Military
Con‡ict from 1500 to 2000. New York: Vintage Books.
[37] Kuran, T. (2011) The Long Divergence: How Islamic Law Held Back the Middle East, Princeton University Press.
[38] Lagerlöf, N. (2014) "Statehood, Democracy and Preindustrial Development" working paper.
[39] Landes, D. (1998) The Wealth and Poverty of Nations: Why Some Are So Rich and Others
So Poor, New York: W.W. Norton.
[40] Mokyr, J. (1990) The Lever of Riches: Technological Creativity and Economic Progress. Oxford: Oxford University Press.
[41] Morris, I. (2010) Why the West Rules - For Now: The Patterns of History and What They
Reveal About the Future, London: Pro…le Books.
30
[42] Murray, D. and M. Schaller (2010) "Historical Prevalence of Infectious Diseases Within 230
Geopolitical Regions: A Tool for Investigating the Origins of Culture" Journal of CrossCultural Psychology 41(1), 99-108.
[43] North, D. (1990) Institutions, Institutional Change and Economic Performance. Cambridge
University Press.
[44] Olsson, O. and Hibbs, D.A. (2005) "Biogeography and Long-Run Economic Development"
European Economic Review 49(4), 909-938.
[45] Olsson, O. and C. Paik (2015) "Long-Run Cultural Divergence: Evidence from the Neolithic
Revolution" Working Papers in Economics 620, University of Gothenburg.
[46] Pinhasi, R., J. Fort, and A. Ammerman (2005) "Tracing the Origins and Spread of Agriculture
in Europe" PLOS Biology 3(12), 2220-2228.
[47] Pomeranz, K. (2000) The Great Divergence: China, Europe, and the Making of the Modern
World Economy. Princeton: Princeton University Press.
[48] Postgate, J.N. (1995) Early Mesopotamia: Society and Economy at the Dawn of History.
London: Routledge
[49] Putterman,
L. (2006) "Agricultural Transition Data Set" online resource,
http://www.econ.brown.edu/fac/louis_putterman/agricultural%20data%20page.htm.
[50] Putterman, L. (2008) "Agriculture, Di¤usion and Development: Ripple E¤ects of the Neolithic Revolution" Economica 75(300): 729-748.
[51] Putterman, L. and D. Weil (2010) "Post-1500 Population Flows and the Long-Run Determinants of Economic Growth and Inequality" Quarterly Journal of Economics 125(4),
1627-1682.
[52] Smith, B.D. (1998) The Emergence of Agriculture, New York: Scienti…c American Library.
[53] Sokolo¤, K. and S. Engerman (2000) "History Lessons: Institutions, Factor Endowments,
and Paths of Development in the New World" Journal of Economic Perspectives 14(3),
217-232.
[54] Spolaore, E. and R. Wacziarg (2013) "How Deep are the Roots of Economic Development?"
Journal of Economic Literature 51(2), 1-45.
[55] Stasavage, J. (2011) States of Credit: Size, Power and the Development of European Polities,
Princeton University Press.
[56] Steckel, R. and J. Rose (2002) The Backbone of History: Health and Nutrition in the Western
Hemisphere, Cambridge University Press.
[57] Tabellini, G. (2010) "Culture and Institutions: Economic Development in the Regions of
Europe" Journal of the European Economic Association 8(4), 677-716.
[58] Talhelm, T. et al (2014) "Large-Scale Psychological Di¤erences Within China Explained by
Rice Versus Wheat Agriculture" Science 344 (9 May), 603-608.
[59] Thornhill, R. and C. Fincher (2013) The Parasite-Stress Theory of Values and Sociality, New
York: Springer.
[60] Tilly, C. (1990) Coercion, Capital and European States: AD 990-1990, London: Blackwell.
[61] Triandis, H.C (1995) Individualism & Collectivism: New Directions in Social Psychology,
Boulder: Westview Press.
[62] Voigtländer, N. and H.J. Voth (2013) "The Three Horsemen of Riches: Plague, War, and
Urbanization in Early Modern Europe" Review of Economic Studies 80(2).
31
[63] Weisdorf, J. (2005) "From Foraging to Farming: Explaining the Neolithic Revolution", Journal of Economic Surveys 19(4), 561-586.
[64] Wittfogel, K. (1957) Oriental Despotism: A Comparative Study of Total Power. New Haven:
Yale University Press.
32
Table 1: Baseline cross-country results controlling for geography
Average time since
agricultural transition
Predicted time since
agricultural transition
Conley SE
Log latitude
Log land suitability for
agriculture
Log arable land area
Log distance to coast or
river
Log mean altitude
Log area
Roughness of terrain
Temperature
Precipitation
Migratory distance to
Addis Abeba
Longitude
(1)
(2)
(3)
(4)
-0.492***
(0.133)
-0.469**
(0.198)
-0.485***
(0.121)
-0.547***
(0.134)
-0.308**
(0.149)
-0.347**
(0.150)
-0.313**
(0.136)
[0.189]
[0.194]
[0.138]
[0.151]
[0.147]
[0.179]
[0.139]
-0.483***
(0.120)
-0.663***
(0.097)
-0.518***
(0.136)
-0.625***
(0.173)
-0.146
(0.231)
0.007
(0.133)
1.839
(1.534)
-0.615***
(0.135)
-0.653***
(0.105)
-0.482***
(0.129)
-0.689***
(0.095)
-0.510***
(0.106)
-0.582***
(0.115)
1.733**
(0.777)
0.078
(0.188)
-0.632***
(0.223)
(5)
(6)
Dependent variable:
Log GDP per capita in 2005
-0.033
(0.027)
0.006
(0.007)
0.211
(0.158)
(7)
(8)
(9)
-0.396**
(0.149)
[0.194]
-0.338***
(0.117)
[0.136]
-0.326***
(0.107)
-0.757***
(0.109)
-0.020**
(0.007)
Constant
12.435***
7.429*
16.650***
16.496***
15.722***
14.681***
15.542***
11.655***
15.485***
(0.984)
(4.252)
(1.011)
(1.604)
(1.291)
(1.795)
(1.113)
(1.138)
(1.005)
Observations
64
55
59
59
59
59
59
65
60
R-squared
0.16
0.41
0.52
0.54
0.55
0.54
0.57
0.10
0.48
Note: The estimator is OLS in all specifications. Robust standard errors are in parentheses. In calculating Conley standard errors (in square brackets), we assume that
spatial autocorrelation exists among observations which are within ten degrees of each other. *** p<0.01, ** p<0.05, * p<0.1. The sample is all countries specified as
Western in the text with available data.
Table 2: Robustness of cross-country results to sample variation and alternative indicators of time since agricultural transition
Sample
Average time since
agricultural transition
Predicted time since
agricultural transition
Earliest date of
transition
Time since agricultural
transition (Putterman)
Europe dummy
(1)
(2)
(3)
All
Outside
Fertile
Crescent
All
-0.654***
(0.119)
-0.683***
(0.117)
-0.351**
(0.132)
Observations
R-squared
1.038**
(0.460)
59
0.56
(5)
(6)
Dependent variable:
Log GDP per capita in 2005
Europe
Roman
Sites>0
empire
-0.419**
(0.177)
-0.391**
(0.160)
-0.436***
(0.127)
1.010***
(0.372)
0.367
(0.288)
Southwest Asia dummy
Fertile Crescent
dummy
(4)
53
0.56
59
0.55
(7)
(8)
(9)
(10)
All
Sites>0
All
All
-0.219*
(0.115)
-0.449***
(0.118)
-0.223**
(0.095)
0.995**
(0.418)
0.201
(0.325)
36
0.39
37
0.57
39
0.58
60
0.52
39
0.60
57
0.46
-0.139**
(0.066)
59
0.43
Note: The estimator is OLS in all specifications. Robust standard errors are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The full sample of Western countries
includes all countries specified as Western in the text. Geographical controls and a constant with unreported coefficients have been included in each regression. The set
of geographical controls are Log arable land area and Log distance to coast or river.
Table 3: 2SLS estimation using distance to Jericho as an instrumental variable
Average time since
agricultural transition
Distance to Jericho
Geographical controls
Observations
R-squared
F-stat
Panel A: 1st stage estimates
(1)
(2)
(3)
(4)
Dependent variable:
Log GDP per capita
Av. time since
in 2005
agricultural transition
OLS
OLS
IV-2SLS
IV-2SLS
-0.492***
-0.526**
(0.133)
(0.257)
-0.047
-0.806***
-0.775***
(0.282)
(0.086)
(0.081)
No
No
No
Yes
64
64
64
59
0.16
0.16
0.584
0.689
87.19
40.59
Panel B: 2nd stage IV-estimates
Average time since
agricultural transition
(IV – Dist. to Jericho)
Log arable land
Log distance to coast or
river
Observations
(3)
(4)
Dependent variable:
Log GDP per capita
in 2005
-0.468**
-0.558***
(0.190)
(0.157)
64
-0.501***
(0.112)
-0.643***
(0.133)
59
Notes: The estimator is OLS in columns 1-2 and IV-2SLS in columns 3-4. Distance to Jericho is the excluded instrumental
variable and Average time since the agricultural transition is the endogenous variable in columns (3)-(4). Robust standard errors
are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The full sample of Western countries includes all countries specified
as Western in the text. Geographical controls and a constant with unreported coefficients have been included in each
regression. The set of geographical controls are Log arable land area and Log distance to coast or river.
Table 4: Cross-country results controlling for various historical channels
Historical
control variable(s):
Dependent variable:
Log GDP per capita in 2005
OLS point
OLS point estimate for
estimate for
Average time since
Obs
historical control
agricultural transition
(1)
None
None
(2)
Predicted genetic diversity
Predicted genetic diversity sq.
-4,450.98
3,002.40
(3)
Ethnic fractionalization
-1.337**
(4)
Atlantic dummy
1.224***
(5)
State history 1-1950 CE
2.633***
(6)
Roman empire in 200 CE
Byzantine empire in 500 CE
Carolingian empire in 800 CE
Mongol invasion in 1300 CE
Ottoman empire in 1600 CE
0.346
0.229
0.886**
-0.731***
-0.001
(7)
Legal origin UK
Legal origin France
Legal origin Scandinavia
(8)
(9)
R2
-0.485***
(0.121)
-0.426***
(0.155)
59
0.52
59
0.56
-0.405***
(0.113)
-0.282**
(0.133)
-0.539***
(0.156)
-0.478***
(0.154)
58
0.59
59
0.61
50
0.57
59
0.67
0.963**
0.708**
1.430***
-0.458***
(0.122)
59
0.62
Generalized trust
2.952***
47
0.59
Muslim population
Protestant population
-0.007
0.014***
-0.289*
(0.145)
-0.165
(0.150)
58
0.59
Note: The estimator is OLS in all specifications. Each row presents results from a specific regression with historical
controls as specified. Robust standard errors have been used in all specifications and are shown for the point estimates of
Average time since agricultural transition but are not reported for the historical variables. *** p<0.01, ** p<0.05, * p<0.1. The
sample is all countries specified as Western in the text with available data. Geographical controls and a constant with
unreported coefficients have been included in each regression. The geographical controls are Log arable land area and Log
distance to coast or river.
Table 5: Cross-regional analysis among 285 European NUTS2-regions
(1)
(2)
(3)
(4)
Dependent variable:
Log GDP per capita in 2010
Average time since
-0.233*** -0.259*** -0.171*** -0.162***
agricultural transition (0.022)
(0.040)
(0.048)
(0.046)
Conley SE
[0.047]
[0.105]
[0.083]
[0.080]
Constant
11.563*** 14.006*** 11.057*** 11.860***
(0.153)
(0.543)
(0.278)
(0.720)
Geographical controls
Country FE
Observations
R-squared
No
No
285
0.25
Yes
No
283
0.45
No
Yes
285
0.69
Yes
Yes
283
0.76
Note: The estimator is OLS. Columns (3)-(4) include country fixed effects. The sample is all Western regions with
available data on NUTS2-level. Robust standard errors are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. In calculating
Conley standard errors (in square brackets) for the estimates of Average time since agricultural transition, we assume that
spatial autocorrelation exists among observations which are within ten degrees of each other. The controls are Log area,
Latitude, Altitude, Arable land in 1600 CE, as well as Access to water.
Table 6: Within-country relationships between average GDP per capita and time since agricultural transition
for NUTS3-regions in five large countries
Country
OLS point estimate for
Average time since agricultural
transition
No
With
controls
controls
Observations
Pinhasi
sites
France
-0.197**
(0.075)
R2=0.098
-0.084
(0.063)
R2=0.444
96
108
Germany
0.158***
(0.037)
R2=0.040
-0.079*
(0.040)
R2=0.396
429
57
Italy
-0.565***
(0.106)
R2=0.208
-0.251***
(0.062)
R2=0.802
107
78
Spain
-0.807***
(0.175)
R2=0.234
-0.508***
(0.251)
R2=0.439
51
8
Turkey
-0.405***
(0.047)
R2=0.407
-0.241***
(0.053)
R2=0.531
81
30
Note: The table shows estimated coefficients for the within-country relationships between Average time since agricultural
transition and Log Average GDP per capita 1997-2008 for the five largest countries with significant regression coefficients.
The estimator is OLS in all specifications and each observation is a NUTS3-region. A constant with unreported
coefficients has been included in all regressions. The set of geographical control variables includes Log Area, Latitude, and
Altitude with unreported coefficients. Robust standard errors are in ()-parentheses. Pinhasi sites refers to the number of
archaeological sites in Pinhasi et al (2005) within each country’s borders that are used for assessing the date of transition
for each region. *** p<0.01, ** p<0.05, * p<0.1.
Table 7: Historical evolution of relationship
(4)
(5)
(6)
(7)
Dependent variable:
Log Population density in:
Log GDP per capita in:
1 CE
1000 CE 1500 CE
1500 CE
1820 CE
Average time since
agric. transition
Geographical
controls
Observations
R-squared
(1)
(2)
(3)
0.215
(0.142)
-0.001
(0.125)
-0.231*
(0.137)
-0.050*
(0.024)
-0.034
(0.049)
No
No
No
No
Yes
No
Yes
54
0.04
60
0.00
61
0.04
21
0.07
21
0.26
28
0.36
28
0.60
-0.180*** -0.121***
(0.033)
(0.030)
Note: The estimator is OLS in all specifications. Average time since agricultural transition is measured as time since the
transition at the corresponding date of the dependent variable (i.e. 1 CE, 1000 CE, etc). Robust standard errors in
parenthesis. *** p<0.01, ** p<0.05, * p<0.1. The sample is always all Western countries with available data. A constant
with unreported coefficients have been included in each regression. The geographical controls are Log arable land area and
Log distance to coast or river.
Table 8a: Inclusive institutions as a causal channel: Executive constraints in 1500 CE
(1)
(2)
Exec. constraints
in 1500
Average time since agricultural
transition in 1500
Executive constraints
in 1500
Geographical controls
Observations
R-squared
(3)
(4)
(5)
Dependent variable:
Log GDP per capita
in 1500
(6)
-0.050*
(0.024)
No
0.004
(0.027)
0.225***
(0.060)
No
-0.020
(0.033)
0.222***
(0.071)
Yes
21
0.07
21
0.51
21
0.58
-0.260***
(0.055)
-0.244***
(0.064)
No
No
0.222***
(0.051)
No
47
0.23
21
0.17
21
0.51
Notes: The dependent variable is Executive constraints in 1500 in column (1)-(2) and Log GDP per capita in 1500 in columns
(3)-(6). Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The estimator is OLS in all specifications.
The sample is always all Western countries with available data except in column (2) where we have restricted the sample
to make it comparable with the same 21 country observations in columns (2)-(6). A constant with unreported coefficients
have been included in each regression. The geographical controls are Log arable land area and Log distance to coast or river.
Table 8b: Historical levels of pathogen prevalence and inclusive institutions as intermediate causal channels
(1)
Average time since
agric. transition
Democracy stock
1900-2000
Historical pathogen
prevalence
Protestant population
(2)
Historical pathogen
prevalence
0.350***
0.293***
(0.028)
(0.035)
(3)
(4)
(5)
Dependent variable:
Dem. stock 1900-2000
-155.350*** -119.684**
(39.048)
(48.265)
-0.492***
(0.133)
Muslim population
Generalized trust
Geographical controls
Observations
R-squared
No
59
0.51
Yes
56
0.58
No
64
0.22
Yes
59
0.28
No
64
0.16
(6)
(7)
(8)
Log GDP per capita in 2005
-0.114
-0.031
-0.010
(0.106)
(0.091)
(0.107)
0.243***
0.220***
0.225***
(0.035)
(0.029)
(0.033)
-0.597**
-0.585*
(0.279)
(0.313)
-0.001
(0.003)
-0.002
(0.004)
No
64
0.48
Yes
56
0.79
Yes
55
0.80
(9)
0.051
(0.121)
0.254***
(0.034)
-0.546
(0.344)
0.004
(0.004)
-0.005
(0.004)
-1.700*
(0.872)
Yes
44
0.85
Notes: The dependent variable is Historical pathogen prevalence in columns (1)-(2), Democracy stock 1900-2000 in columns (3)-(4) and Log GDP per capita in 2005 in
columns (5)-(9). Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. In columns (5)-(9), the estimates for Democracy stock 1900-2000 have been
multiplied by 100 for expositional purposes. The estimator is OLS in all specifications. The sample is always all Western countries with available data. A constant with
unreported coefficients has been included in each regression. The geographical controls are Log arable land area and Log distance to coast or river.
10
Log GDP per capita in 2005
12
Figure 1: Relationship between log GDP per capita in 2005 and time since agricultural transition within
the Western, Sub-Saharan African, and East Asian core areas.
6
8
Western
East Asia
4
Sub-Saharan Africa
0
2000
4000
6000
8000
Time since agricultural transition (Putterman, 2006)
Western
Sub-Saharan Africa
East Asia
10000
All other
Note: The figure combines three fitted lines for separate OLS regressions using the Western sample (N=62), the
East Asian sample (N=22), and the Sub-Saharan African sample (N=41). Included in the graph are also 33 other
country observations. The color and shape of each country indicates whether it belongs to the Western, East Asian,
Sub-Saharan African, or All other category. Time since agricultural transition is taken from Putterman (2006). In the total
sample of 158 observations, the (non-displayed fitted equation is Log GDP per capita in 2005 = 6.99*** +0.000147***
x Time since agricultural transition. *** indicates significance at the .01 level.
Figure 2: A model of a long-run development reversal
Early transition to agriculture
Early Neolithic
transition
(1c)
(1a)
* High parasite stress
* Interdependent production
* Predation by primitive
neighbors
* Public ownership
* Imitation
* Extractive institutions
(1d)
(1b)
Collectivist
cultural norms
Weak current
economic performance
Late transition to agriculture
Late Neolithic
transition
(2c)
(2a)
* Low parasite stress
* Independent production
* Learning from more
advanced neighbors
* Private property rights
* Innovation
* Inclusive institutions
(2d)
(2b)
Individualistic
cultural norms
Strong current
economic performance
Figure 3: Causal relationships in empirical model
Distance to
agricultural origin
- Di
Agricultural
transition - Ti
Exogenous factors:
Geography, biogeography, climate Xi
Individualistic or
collectivist culture
(unobserved)
9700 BP – 5600 BP
Major historical
events - Hi
Inclusive or
extractive
institutions - Zi
1500 – 2000 CE
Time
GDP pc Yi
2005 CE
Figure 4: Relationship between age of site and distance to Jericho among 765 Pinhasi et al (2005) sites
Note: The figure shows the unconditional relationship between age of site in calibrated radiocarbon years BP and
aerial distance to Jericho among 765 archaeological sites in Pinhasi et al (2005). Distance to Jericho is calculated
using the Great Circle Formula and takes no account of geographical constraints such as mountains or oceans. Each
site observation is represented by a circle. The size of each circle is proportional to the standard deviation of the
calibrated dating of the site such that larger circles have a larger standard deviation. The fitted curve is a linear OLS
regression line with estimated coefficients as displayed in the graph (with robust standard errors in parenthesis).
Figure 5: Conditional relationship between log GDP per capita in 2005 and average time since agricultural
transition among 64 Western countries
Note: The figure shows the added-variable plot for the conditional relationship between Log GDP per capita in 2005
and Average time since agricultural transition from the regression specification in Table 1, Column 3.
Average Log GDP per capita 1997-2008
8
9
10
11
12
Figure 6: Scatter plot and unconditional relationships within five countries between average log GDP per
capita in 1997-2008 and average time since agricultural transition among 1,371 European NUTS3-regions.
Germany (+0.158***)
France (-0.197**)
Italy (-0.565***)
Spain (-0.807***)
7
Turkey (-0.405***)
4
6
8
10
Average time since agricultural transition
12
Note: The figure shows the scatter plot for the bivariate relationship between Average Log GDP per capita 1997-2008
and Average time since agricultural transition (in 1,000 years) for 1,371 Western NUTS3-regions. It also shows the
separate unconditional regression lines for Germany, France, Italy, Spain and Turkey with slope coefficients and
levels of significance in parenthesis. German observations are green, French observations dark blue, Italian
observations light blue, Spanish observations black and Turkish observations dark orange. All other region
observations are grey. More details for the within-country regressions are shown in Table 6. The estimated
coefficients for the whole sample are (with robust standard errors in parenthesis): Average log GDP per capita 19972008 (ij) = 12.99 (0.147) – 0.484 (0.021) *Average time since agricultural transition(ij) + ε(ij).
OLS estimate for Av. time since agric. transition
.2
-.4
-.2
0
-.8
-.6
Figure 7: Historical evolution of relationship between log GDP per capita and average time since
agricultural transition among Western countries 1000-2005
1000
1500
1600
1700
1820
1913
1980
2005
1000
1100
1200
1300
1400
1500
Year
Estimate (with controls)
1600
1700
1800
1900
2000
95% conf. interval
Note: The figure shows the regression coefficients from 8 separate regressions for different time periods, all
including control variables. The specification is Log GDP per capita (in year t) = β0 + β1*Average time since agricultural
transition (in year t) + β2*Distance to coast or river + β3*Arable land + ε (in year t). The 8 β1-estimates from these
regressions are shown as blue triangles where the dashed lines represent the associated 95% confidence interval. The
included years t (in the CE period with number of Western country observations in parenthesis) are 1000 (20), 1500
(21), 1600 (19), 1700 (21), 1820 (28), 1913 (34), 1980 (43), and 2005 (64). The data on GDP per capita for all periods
except 1980 and 2005 are from Maddison (2013). Results from the regressions with controls for 1500 and 1820 are
shown in Table 7 and for 2005 in Table 1. Results from the regressions for 1600, 1700, 1913, and 1980 are available
upon request.
APPENDICES
Appendix A:
Analysis of the Pinhasi et al (2005) data
1. Introduction
In this section, we briefly discuss the properties of the Pinhasi et al (2005) data, which form the basis of our
empirical analysis. Section 2 reviews the basic characteristics of the data, section 3 discusses the geographical
distribution, whereas section 4, lastly, deals with issues regarding measurement error and potential biases.
2. Basic characteristics
The frequency distribution of the 765 archaeological sites in Pinhasi et al (2005) is shown in the histogram
in Figure A1. The distribution has two very distinct peaks; one around 7400 BP and one around 5800 BP.
These coincide with two periods of particularly rapid expansions referred to as the Linear Bandkeramikculture (LBK), in which farmers broke into central Europe and southern Germany after several centuries of
standstill, and the later Funnelbeaker-culture (TRB) when most of northern Europe and Scandinavia was
settled (Bellwood, 2005).
Figure A1: Frequency distribution of time since agricultural transition among 765 Pinhasi et al (2005) sites
The five oldest and five youngest sites are shown in Table A1. The oldest sites are all in the Fertile Crescent
area of the Middle East, comprising areas of current Iraq, Turkey, Jordan, Israel, Syria, and Lebanon. Among
the oldest five sites are the very famous Cayönü and Mureybet that are often discussed in the archaeological
i
literature and which are around 12,000 years old. The youngest site in the sample is from 5140 BP at King
Barrow Ridge, UK. The standard deviation of the observations range between 28-291 years.
Table A1: The five oldest and youngest sites in Pinhasi et al (2005)
Site
Country
St dev
Oldest 5 sites
M’lefaat
Hallan Cemi Tepsi
Cayönü
Mureybet
Wadi Faynan 16
Calibrated
date (yrs BP)
Iraq
Turkey
Turkey
Syria
Jordan
12811
12429
12300
11855
11851
75
268
200
251
103
Youngest 5 sites
King Barrow Ridge
Normanton Down.
Dorchester VIII
Flögeln
Szczecin-Ustowo
UK
UK
UK
Germany
Poland
5140
5147
5147
5148
5150
170
153
150
117
130
3. Geographical distribution
As we discuss also in the paper, the distribution of sites across Western countries is not even. United
Kingdom has by far the greatest number of site observations (126) with France second (108), followed by
Italy (78). We would thus expect that the accuracy of dating is greatest in these countries. More than 20
countries have 0 observations within their borders. Below, we will return to this issue and perform a statistical
analysis on the determinants of why some countries have more observations than others. Figure A2 shows the
geographical distribution of the 765 sites. As can be seen, large parts of North Africa and Eastern Europe lack
any observations.
ii
Figure A2: Neolithic sites from Pinhasi et al (2013) and spread of agriculture in the Western core
iii
Figure A3: Neolithic sites and the spread of agriculture in Italy
Figure A3 shows the distribution of 78 sites within Italy. The colors have been generated according to the
Inverse Distance Weighted Interpolation (IDW) method discussed in 3.2 such that dark green areas made the
agricultural transition first and the dark red areas made the transition last. The figure illustrates that there can
be quite a bit of variation also within countries.
4. Measurement issues and potential biases
The data set in Pinhasi et al (2005) has been collected from various different sources. The authors collected
the earliest date of Neolithic occupation for each site. Outlier dates that were considered anomalies to the
existing literature were excluded. Also the typical culture (LBK, TRB, etc) of each site is specified. The
material used for the dating of the site is charcoal, whenever that was available. Comparatively few of the sites
were dated by using the best available methodology in the field; accelarator mass spectrometry (AMS). The
authors preferred to include many sites rather than fewer sites with very high quality of measurement and
recognize that the choice involves a certain tradeoff. In the paper, Pinhasi et al (2005) exclude 30 observations
with a standard deviation above 200 so that their sample consist of 735 observations. Since we did not see any
particular reason for using a standard deviation cut-off of 200 for inclusion, we chose to use also these
observations so that our sample has 765 sites as observations in total.
iv
Are the observations biased in some structural way? We have already shown that the geographical
distribution across countries is uneven. Table A2 features the logged number of sites within countries as the
dependent variable and then studies what determines this number. Not surprisingly, the number of
observations increases with the size of country area. Furthermore, it is also clear that richer countries have
more observations, all else equal. This presumably reflects the fact that richer countries can spend more
resources on archaeological research. Similarly, more sites are available in countries with a greater arable land
area. One of the strongest predictors is however distance to Jericho such that when controlling for other
factors, a longer distance to Jericho is associated with fewer sites.
Does this potentially bias our results? The fact that richer countries have more observations implies that the
accuracy of regional dating should be higher in such countries. Although this might affect our regional
analysis, we do not think that it should bias our cross-country investigation to any greater extent as long as the
dating in a country with few observations is as accurate as the dating in a country with many observations.
Table A2: Determinants of the number of sites within countries
Dependent variable:
Log number of sites
Distance to Jericho
Latitude
Longitude
Log arable land area
Log distance to coast or river
Log country area
Log GDP per capita
Europe dummy
Observations
R-squared
-0.501***
(0.158)
-0.141*
(0.024)
-0.007
(0.010)
0.428**
(0.174)
-0.305
(0.243)
0.452***
(0.143)
0.669***
(0.154)
0.296
(0.586)
60
0.498
Notes: The figure shows the determinants of log number of Pinhasi archaeological sites within within our sample of
Western countries. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
v
Appendix B:
Supplementary tables
Table B1: Correlates of average and predicted time since agricultural transition among Western countries
(1)
Distance to Jericho
Latitude
-0.806***
(0.094)
Longitude
Log arable land area
(2)
(4)
Dependent variable:
Average time since
agricultural transition
-0.491***
(0.060)
-57.854***
(7.317)
5.928*
(3.273)
(3)
Desert, 10k BP
Tropical extreme, 10k
BP
Steppe-Tundra, 10k BP
Ice, 10k BP
Semi-desert, 10k BP
Dry steppe, 10k BP
Wooded steppe, 10k
BP
Mediterranean, 10k BP
Constant
Observations
R-squared
9,435.9***
(185.482)
64
0.58
(7)
Predicted time since
agricultural transition
-38.638***
(12.826)
14.332
(11.067)
-48.716***
(14.258)
15.857
(11.287)
134.449
(97.007)
-176.869
(123.824)
-258.709
(403.009)
307.867
(347.226)
60.223
(306.250)
-228.846
(556.117)
1,593.516**
(597.081)
367.312
(294.870)
410.874
(374.693)
418.746
10,991.8*** 10,467.5*** 11,375.8*** 9,918.0***
(321.221)
(415.013)
(1,289.614)
(854.042)
8,704. 3***
(749.658)
9,568.6***
(1,011.562)
65
0.31
60
0.36
Precipitation
64
0.86
59
0.89
-0.492***
(0.072)
-65.955***
(21.857)
6.584*
(3.576)
(6)
-0.270*
(0.151)
-54.506***
(12.744)
14.630
(8.881)
Log distance to coast
or river
Temperature
-0.513***
(0.048)
-64.868***
(7.110)
3.296
(2.509)
117.109***
(34.281)
139.791**
(56.313)
(5)
-10.117
(27.675)
1.359
(2.056)
62
0.86
59
0.89
Notes: The figure shows the correlates of average and predicted time since agricultural transition on country level within
our sample of Western countries. Given the demic diffusion of agriculture across the Western core, the table
demonstrates as expected that Distance to Jericho and Latitude have a negative impact and Longitude a positive impact,
despite the obvious fact that Distance to Jericho is structurally correlated with both Latitude and Longitude. The three
variables together explain 86 percent of the variation in the dependent variable in column 2. Distance to Jericho is not used
for explaining Predicted time since agricultural transition since the former variable is used directly to construct the dependent
variable. The map for the biogeographical variables that are included in column 5 are shown in figure C9. Robust
standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
vi
Table B2: Correlates of average time since agricultural transition on regional level
(1)
Distance to Jericho
Latitude
(2)
(3)
Dependent variable:
Average time since
agricultural transition
-0.797***
(0.028)
0.394***
(0.098)
-123.587***
(5.378)
56.772***
(7.007)
9,255.381***
(81.295)
1,404
0.54
Longitude
Log region area
Predicted time since
agricultural transition
11,235.225***
(162.407)
8,834.703***
(143.723)
-55.104***
(2.555)
65.888***
(1.213)
-55.322***
(5.104)
0.041
(0.028)
856.953***
(111.775)
61.297***
(20.355)
9,096.583***
(155.699)
1,404
0.72
1,404
0.74
1,457
0.93
1,457
0.94
Arable land in 1600
CE
Atlantic dummy
Observations
R-squared
(5)
0.488***
(0.095)
-124.034***
(5.949)
61.131***
(6.894)
-67.557***
(10.133)
0.279***
(0.052)
71.148
(158.656)
-80.152**
(38.917)
11,343.334***
(183.467)
Mean elevation
Constant
(4)
-54.545***
(2.701)
63.332***
(1.320)
Notes: The figure shows the correlates of average and predicted time since agricultural transition on regional level within
our sample of Western NUTS3 regions. A noteworthy feature is that whereas Distance to Jericho has a very similar estimate
in column 1 as in column 1 of table B1 (-0.797 vs -0.806), the estimate switches from being negative to positive in
columns 2-3. The most likely explanation for this is that Distance to Jericho and Longitude have a nearly perfect negative
linear relationship (Pearson correlation coefficient is -0.88). Robust standard errors in parentheses. *** p<0.01, **
p<0.05, * p<0.1.
vii
Appendix C:
Supplementary figures
Figure C1: Relationship between log GDP per capita in 2005 and time since agricultural transition among 41
Sub-Saharan African countries.
Note: The figure shows the unconditional bivariate relationship between Time since agricultural transition and Log GDP per
capita in 2005. N=41, R2=0.281, p-value < 0.001. OLS estimate using robust standard errors. Each country is identified
by a three-letter isocode. Time since agricultural transition is taken from Putterman (2006).
viii
Figure C2: Relationship between log GDP per capita in 2005 and time since agricultural transition among 22
Central and East Asian countries.
Note: The figure shows the unconditional bivariate relationship between Time since agricultural transition and Log GDP per
capita in 2005. N=22, R2=0.186, p-value = 0.074. OLS estimates using robust standard errors. Each country is identified
by a three-letter isocode. Time since agricultural transition is taken from Putterman (2006).
ix
Figure C3: Average time since agricultural transition among 107 Italian regions
x
Figure C4: Average GDP per capita 1997-2008 among 107 Italian regions
xi
Figure C5: Bivariate relationship between Black Death mortality in 1347-1353 and average time since
agricultural transition among 53 Western cities
Note: The figure shows the bivariate relationship between Black Death mortality in 1347-1353 and average time since
agricultural transition for 53 European cities. The data on Black Death mortality are from Christakos et al (2005), p 141.
The data on Average time since agricultural transition for the 53 cities have been obtained with the same method as
described in section 3.2.
xii
Figure C6: Bivariate relationships between average Log GDP per capita 1997-2008 and average time since
agricultural transition among the six largest countries in Western zone
Note: The figures show the bivariate, unconditional relationships between average log GDP per capita in 1997-2008 and
average time since agricultural transition among NUTS3-regions within the six largest countries in the Western zone:
Germany, France, Italy, Spain, United Kingdom, and Turkey. The estimated regression coefficients for all countries
except the United Kingdom are shown in Table 6.
xiii
Figure C7: Conditional relationships between dependent variable Log GDP per capita in 1500 and
independent variables Average time since agricultural transition in 1500 and Executive contraints in 1500
Note: The figures show the conditional scatter plots for 21 countries from Table 8a, column (5). Each country is
identified by a three-letter isocode.
xiv
Figure C8: Bivariate relationship between Democracy stock 1900-2000 and Average time since agricultural
transition for 64 Western countries
Note: The figure shows the unconditional scatter plot between Democracy stock 1900-2000 on the vertical axis and
Average time since agricultural transition on the horizontal axis. A fitted egression line has been included. The details
from the regression are shown in Table 8b, column (3).
xv
Figure C9: Conditional relationships between dependent variable Log GDP per capita in 2005 and
independent variables Average time since agricultural transition and Democracy stock 1900-2000
Note: The figures show the conditional scatter plots for 64 countries from Table 8b, column (6). Each country is
identified by a three-letter isocode.
xvi
Appendix D:
Data appendix
D1: Variable definitions, sources, and descriptive statistics for the main cross-country tables
Outcome variables
GDP per capita in 2005. Real GDP per capita in constant 2000 international dollars from World Development
Indicators, World Bank.
GDP per capita in 1 CE, 1000, 1500, 1820. Real GDP per capita data from Maddison (2013).
Population density in 1 CE, 1000, 1500. Total country population divided by total country area, from Ashraf and
Galor (2013).
Variables capturing time since agricultural transition
Average time since agricultural transition. Main explanatory variable showing the date of transition to agriculture for
the average region in a country, using the Inverse Distance Weighted Interpolation, on the basis of the 765
sites from Pinhasi et al (2005), as described in detail in section 3.2.
Predicted time since agricultural transition. Variable created on the basis of the regression equation specified in
Figure 3. The latitude and longitude of each country’s centroid is used to calculate their approximate Distance
to Jericho. The distance figure for each country i is then imputed into the estimated equation in Figure 3;
Predicted time since agricultural transition(i) = 9783.8 – 1.01782 x Distance to Jericho(i). See also explanation in section
3.3.
Earliest date of transition. The date of transition to agriculture for the first region in a country to make the
transition, using the Inverse Distance Weighted Interpolation, on the basis of the 765 sites from Pinhasi et al
(2005).
Time since agricultural transition (Putterman). First attested date of agricultural transition within a country. Source:
Putterman (2006).
Geographical variables
Distance to Jericho. Using information on the latitude and longitude of the country’s and Jericho’s approximate,
centroids, we calculate the aerial distance to Jericho in kilometers by employing the Great Circle Formula. The
variable does not take into account geographical obstacles such as water or mountains.
Latitude. Latitude degree of the approximate centroid of the country. Source: CIA World Factbook (2013).
Longitude. Longitude degree of the approximate centroid of the country. Source: CIA World Factbook (2013).
xvii
Land suitability for agriculture. Index of the suitability of land for agriculture based on indicators of climate and
ecological suitability for cultivation. Source: Michalopoulos (2012)
Arable land area. Arable land in 2001-2005 as percent of total land area. Source: World Development
Indicators.
Distance to coast or river. Mean distance to an ice-gree coastline or a sea-navigable river in kms. Source: Harvard
CID Research Datasets.
Mean altitude. Mean elevation of country in meters above sea level. Source: Harvard CID Research Datasets.
Area. Area of country in square kilometers. Source: CIA World Factbook.
Rougness of terrain. Average degree of terrain roughness in a country based on grid cell elevation data. Source:
Nordhaus (2006).
Temperature. Average monthly temperature of a country in degrees Celsius. Source: Nordhaus (2006).
Precipitation. Average monthly precipitation of a country in mms over the 1960-1990 period. Source: Nordhaus
(2006).
Migratory distance to Addis Abeba. Measure of migratory distance from Addis Abeba to the country in question,
assuming intermediate geographical stepping-points, as explained in Ashraf and Galor (2013).
Historical variables
Predicted genetic diversity. The expected heterozygosity (genetic diversity) of a given country as predicted by
migratory distance from Addis Abeba. Source: Ashraf and Galor (2013)
Ethnic fractionalization. Herfindahl index of ethnic fractionalization. The variable approximates the probability
that two random persons within a country do not belong to the same ethnic group. Source: Alesina et al (2003).
Muslim and Protestant population. Variables capturing the percentage of Muslims and Protestants in a country’s
population. Source: Ashraf and Galor (2013).
Roman, Byzantine, Carolingian, and Ottoman empire variables. Own assessment based on digitalized maps from
Euratlas (2012).
Mongol invasion in 1300 CE. Dummy=1 if country was strongly affected by Mongol raids around 1300 CE.
Own assessment based on principles described in Appendix F.
Legal origin variables. Dummy variables capturing the legal origin of countries. Source: La Porta et al (1990).
xviii
Generalized trust. The fraction of total respondents within a country who answered “Most people can be
trusted” when answering the survey question “Generally speaking, would you say that most people can be
trusted or that you can’t be too careful in dealing with people?” The variable combines information from five
waves of the World Values Survey during the time period 1981-2008. Source: Ashraf and Galor (2013).
State history 1-1950 CE. The variable captures the extent of a country’s state experience during 1-1950 CE. The
variable is normalized to range between 0 and 1 where 1 means maximum state experience during the period.
Source: Putterman (2010), http://www.econ.brown.edu/fac/louis_putterman/antiquity%20index.htm.
Executive contraints in 1500. This variable uses information from all available observations in Acemoglu et al
(2005) who have coded the variable independently in line with the Polity IV-codebook. The variable ranges
between 1-7 where 1 implies no constraints against the executive and 7 implies very strong constraints as in a
fully developed democracy. We have added information from countries that belonged to the Ottoman empire
during the period that all of them are coded as 1. This is well in line with the literature, for instance Hourani
(2013), where it is emphasized that the Ottoman empire was strongly autocratic in this era.
Democracy stock 1900-2000. Index capturing average levels of democracy during 1900-2000 according to the
2000
Polity2 measure. The score for country i is calculated as Democracy stock 1900-2000 (i) = �𝑠=1900 0.992000−𝑠 ∙
𝑃𝑃𝑃𝑃𝑃𝑃2𝑖,𝑠 where Polity2i,s is the score for country i at time s and where 0.99 is a time discount factor. Source:
Gerring et al (2005).
Historical pathogen prevalence. Index capturing historical levels of pathogen prevalence on a country level. The
index is created by using historical maps and measures the prevalence of nine common infectious diseases;
leishmanias, schistosomes, trypanosomes, leprosy, malaria, typhus, filariae, dengue, and tuberculosis. The
index is normalized to have a zero mean and higher values mean a high prevalence of disease and vice versa.
Source: Murray and Schaller (2010).
D2: Variable definitions and sources for the NUTS2 cross-regional analysis
Average GDP per capita in 2010. Average level of GDP per capita on NUTS2-region level in 2010 in euros.
Source: Eurostat (2012)
Average time since agricultural transition. Constructed in the same way as the equivalent variables on country level.
See D1 for further information.
Area. Area of a region in square kilometers. Source: Own assessment
Latitude. Latitude degree of the approximate centroid of the region. Source: Own assessment
Altitude. Mean elevation of region in meters above sea level. Source: Own assessment
Arable land in 1600. Fraction of cultivated land in 1600 CE. Source: Own assessment using Pongratz et al.’s
data.
xix
Access to water. Indicator for direct water access. Source: Own assessment.
D3: Variable definitions and sources for the NUTS3 cross-regional analysis
Average GDP per capita 1997-2008. Average level of GDP per capita on NUTS3-region level during 1997-2008
in euros. Source: Eurostat (2012)
Average and Predicted time since agricultural transition. Constructed in the same way as the equivalent variables on
country level. See D1 for further information.
Area. Area of a region in square kilometers. Source: Own assessment
Latitude. Latitude degree of the approximate centroid of the region. Source: Own assessment
Altitude. Mean elevation of region in meters above sea level. Source: Own assessment
Arable land in 1600. Fraction of cultivated land in 1600 CE. Source: Own assessment.
Additional references:
Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat, R. Wacziarg (2003) ”Fractionalization” Journal of
Economic Growth, 8: 155-194.
CIA World Factbook (2013), CIA
G-Econ (2006), available online at http://gecon.yale.edu/.
Maddison, A. (2013) “Statistics on World Population, GDP and per capita GDP, 1-2008 AD”
www.ggdc.net/maddison/Maddison.htm.
Hourani, A. (2013) A History of the Arab Peoples, London: Faber and Faber.
Harvard CID Research Data Sets, Center for International Development, Harvard University, available online.
La Porta, R., F. Lopez-de-Silanes, A. Schleifer, and R. Vishny (1999) "The Quality of Government" Journal of
Law, Economics & Organization, 15(1):222-79.
xx
D4: Descriptive statistics
Table D1: Summary statistics and sources of variables for the cross-country analysis in tables 1-2
Variables
Dependent variables
Log GDP per capita in 2005 (constant 2000
USD)
Log Population density in 1 CE
Log Population density in 1000 CE
Log Population density in 1500 CE
Log GDP per capita in 1 CE
Log GDP per capita in 1000 CE
Log GDP per capita in 1500 CE
Log GDP per capita in 1820 CE
Independent variables
Average time since agricultural transition (in
1000 yrs from 2000 CE)
Predicted time since agricultural transition (in
1000 yrs from 2000 CE)
Earliest date of transition (for any region
country, in 1000 yrs from 2000 CE)
Time since agricultural transition (Putterman)
Geographical controls
Log latitude (degrees)
Log land suitability for agriculture (index)
Log arable land area (percent)
Log area (sq. kms)
Rougness of terrain
Temperature (degrees Celsius)
Precipitation (mms)
Migratory distance to Addis Abeba (in 1000
kms)
Longitude (degrees)
Europe, Southwest Asia and Fertile Crescent
dummies
Historical variables
Predicted genetic diversity
Ethnic fractionalization (Herfindahl index)
Muslim population (percent)
Protestant population
Roman empire (fraction of country part of
empire in 200 CE)
Byzantine empire (fraction of country part of
empire in 500 CE)
Carolingian empire (fraction of country part
of empire in 800 CE)
Ottoman empire (fraction of country part of
empire in 1600 CE)
Mongol empire (dummy=1 if country was
overrun by Mongols in 13-14th centuries
CE)
N
Mean
SD
Min
Max
64
8.688
1.369
6.224
10.859
54
60
61
24
21
22
29
1.024
1.318
1.733
6.144
6.097
6.432
6.693
1.266
1.136
1.296
0.164
0.156
0.231
0.396
-1.481
-1.258
-1.258
5.991
5.991
6.064
6.064
3.170
3.442
4.135
6.696
6.477
7.003
7.516
64
7.611
1.100
5.608
9.743
65
7.430
1.122
4.393
9.737
60
8.446
1.593
5.673
12.408
62
7.003
1.850
3.5
10.5
64
-1.117
1.032
-4.198
0.588
64
64
63
63
63
65
2.575
4.501
0.189
13.26
50.60
4.945
1.250
2.131
0.145
7.163
28.91
1.050
-2.106
-2.797
0.017
1.026
2.911
2.462
4.028
8.098
0.569
27.36
109.1
7.821
65
64
24.71
20.41
-21.97
0
77.22
1
65
64
64
63
64
0.737
0.346
33.40
11.87
0.469
0.008
0.221
43.33
26.56
0.455
0.715
0.034
0
0
0
0.756
0.792
99.5
97.8
1
64
0.174
0.357
0
1
64
0.163
0.350
0
1
64
0.267
0.383
0
1
0
1
64
xxi
Legal origin UK, France, Scandinavia
Trust (fraction of population)
State history 1-1950 CE (index)
Democracy stock 1900-2000 (index)
Executive constraints in 1500 (index)
Historical pathogen prevalence
65
51
52
65
47
60
0
0.291 0.128 0.111
0.606 0.151 0.290
-7.078 364.75 -604.6
1.34
0.635
1
-0.273 0.554 -1.19
1
0.663
0.887
637.6
3
0.94
Table D2: Summary statistics and sources of variables used in the regional analysis (NUTS2)
Variables
Dependent variable
Log Average GDP per capita 2010 (in €)
Independent variables
Average time since agricultural transition (in 1000
yrs from 2000 CE)
Geographical controls
Log Region area
Altitude (meters above sea level)
Latitude (degrees)
Arable land in 1600 CE (as fraction of total land
area)
Indicator for direct water access
N
Mean
SD
Min
Max
306
9.932
0.441
8.537
11.27
294
7.050
0.957
5.598
10.29
316
316
316
316
10.02
196.9
47.43
0.220
1.345 2.815
219.4 -0.62
8.416 -21.13
0.110
0
14.11
1387
68.85
0.471
316
0.380
0.486
0
1
Table D3: Summary statistics and sources of variables used in the regional analysis
Variables
Dependent variable
Log Average GDP per capita 1997-2008 (in €)
Independent variables
Average time since agricultural transition (in 1000
yrs from 2000 CE)
Predicted time since agricultural transition (in 1000
yrs from 2000 CE)
Geographical controls
Log Region area
Altitude (meters above sea level)
Latitude (degrees)
Arable land in 1600 CE (fraction of
N
Mean
SD
Min
Max
1,371
9.580
0.846
6.802
11.83
1,371
7.055
0.847
5.243
11.32
1,409
6.924
0.821
4.034
9.368
1,371
1,371
1,371
1.371
7.382
347.9
47.97
0.154
1.384 2.962
362.7 -2.639
5.169 35.05
0.079
0
10.58
2328
59.77
0.371
xxii
Appendix E:
A comparative country study of Iraq and Sweden
In order to illustrate the logic of our model further, we will in this section make a more detailed
comparative country analysis, using Iraq and Sweden as examples. Iraq and Sweden are two extreme
observations in Figure 1. Iraq was one of the very earliest countries to experience sedentary agriculture and
statehood and is currently one of the poorest and institutionally least developed countries in the western
world. For Sweden, the situation is right the opposite. Our analysis aims to demonstrate that Iraq's long
history has been characterized by extractive institutions in the form of autocratic, unstable state formations, a
weak rule of law, a persistent social and economic inequality, and intense rent seeking, whereas Sweden's late
transition to agriculture and much shorter history is characterized by more stable and broadly inclusive
governments, an egalitarian and democratic society, and an absence of major internal or external struggles
over the distribution of resources.
Iraq
The northern parts of current Iraq, as well as the eastern parts on both sides of the Iranian border, were
part of the Fertile Crescent where the very earliest signs of domesticated plants and animals in the world have
been found. In Neolithic times, the area was characterized by open wood and grasslands and hosted a large
number of heavy-seeded wild grasses including the wild progenitors of wheat and barley. After a temporary
downturn in temperature levels (referred to as the Younger Dryas), evidence of domesticated plants started to
appear after 9500 BCE in Syrian sites such as Abu Hureyra and soon spread to the whole region (Diamond,
1997; Bellwood, 2005). Animals such as goats and pigs were domesticated somewhat later, presumably on the
flanks of the Zagros Mountains (Smith, 1995; Diamond, 1997). A few thousand years later, they would be
complemented with cattle and horses.
One of the oldest archaeological sites in northern Iraq is Qermez Dere, with an agricultural settlement from
about 9000 BCE (Pinhasi et al, 2005). One of the most notable findings on this site was three circular,
underground chambers filled with tons of soil, supported by heavy plastered stone pillars and containing six
buried jawless skulls. Even more impressive monuments have been found in other parts of the Fertile
Crescent, such as a cluster of spacious underground chambers with pillars, each weighing 8 tons or more in
Göbekli Tepe, and the massive walls and nearly 8 meter high defensive tower of Jericho, both from about
9000 BCE (Morris, 2010). Clearly, monuments like these were the result of closely coordinated work efforts
by a village community.
Similar findings from neighboring sites from the same period provide a picture of a sedentary population,
living in tight village settlements with houses in uniform style, separate community buildings, shrines and
sometimes defensive fortifications, and a worshipping of ancestors manifested in the keeping of plastered
skulls, presumably of ancestors. Several instances of headless burials could potentially be interpreted as human
sacrifice. Some of the largest settlements, such as Abu Hureyra in Syria, hosted around 5000 people already by
8000 BCE (Bellwood, 2005). Houses with own storerooms and kitchens would soon emerge. Together with
the almost obsessive reverence of ancestors, these social changes indicate a society with important public
goods but also one where private property started to matter and where land was inherited from one
generation to the next (Morris, 2010). As we know from the work of Mulder et al (2009), such a pattern of
inheritance in agricultural communities typically leads to the emergence of income inequality and a hierarchical
society.
xxiii
The heartland of historical Mesopotamia, i.e. the land between Euphrates and Tigris, was not settled by
farmers until about 5000 BCE. Despite the rivers, the area was generally drier than in the grasslands of the
nearby Fertile Crescent, further to the east. Productive farming could still be carried out with the use of
irrigation technology that diverted water from the flooding rivers. This technique required substantial
collective activities such as the construction of canal regulators, digging out and dredging canals and
reservoirs, and digging earthen dykes that protected fields from flooding water. In order to be efficient, the
structure of the irrigation system had to be planned and work carefully coordinated. A failure at any link in
this highly sensitive system might have disastrous consequences for the village. Workers further had to be fed
and provided with effective tools and the structure of fields and cultivation had to be carefully planned
(Postgate, 1995).
All these characteristics of the new production technology implied pressures towards creating a strong
centralized power. In the rural villages, irrigation management was coordinated by an official referred to as
gugallum (Postgate, 1995, p 178). It is certainly straightforward to think of this kind of social organization as
being the driver of the further development towards more centralized power in the villages which eventually
resulted in full-blown city-states. The famous hydraulic hypothesis of Wittfogel (1957) proposes that the
nature of water management in Mesopotamia, Egypt, and China was responsible for the rise of highly
centralized states in these regions. 1 Exactly how causality worked between collective water management and
centralized rule in Mesopotamia remains to be established, but the association is in any case clear.
A climate downturn around 3800 BCE with even less rain and less reliable flooding of the rivers, put a great
strain on the already fragile ecological system of Mesopotamia. It has been suggested that this downturn
provided a further impetus towards a more efficient organization and use of resources (Morris, 2010). Only
the cities that succeeded in this endeavour survived and most likely experienced an inflow of people from
collapsed communities. 2 One of the oldest cities in the area was Eridu where the earliest settlement dates back
to 5400 BCE. Archaeological excavations have revealed that the city hosted a succession of temple structures,
built on top of each other, with the earliest small shrine dating to 5000 BCE (Postgate, 1995).
During this period, Uruk emerged as a leader among the growing cities in the area and soon became very
influential culturally, not only in Mesopotamia but also in neighboring lands. Around 3500 BCE, the first
statehood arose. This event is generally viewed as the onset of what is usually referred to as civilization. The
Sumerians in Uruk invented cuneiform writing around 3300 BCE, inscribed on clay tables, and used bronze
on a large scale by 3000 BCE. The massive temple area was clearly the home of a very powerful "priest-king",
managing a temple economy that controlled substantial areas of agricultural land as well as farm labor to work
its fields. The clay tablets that have been recovered often record temple production quotas. A highly
specialized labor force worked in the temples apart from the priests, including accountants, guards, watercarriers, weavers, boatmen, barbers, and musicians, as well as regular slaves (Postgate, 1995). The temple most
likely also organized the long-distance trade that brought foreign luxury goods to Mesopotamia and
contributed to the more efficient mining of metals like bronze and copper. Humble citizens presented regular
offerings to the gods (and food rations to the temple staff) and were often sent to war on neighboring citystates in large military campaigns. The city of Uruk is believed to have covered around 400 hectares with a 9
km town wall (Postgate, 1995) and presumably hosted around 20,000 individuals, packed into carefully
planned mud brick dwellings (Morris, 2010).
1
The hypothesis has been criticized on many grounds, for instance that the major Chinese efforts at centralized water
management were made after the rise of centralized rule.
2
Yoffee (1995) even refers to this sudden urbanization process as a "demographic implosion" which resulted in a relative
depopulation of the countryside.
xxiv
Uruk's eminence in Mesopotamia ended early in the 3rd millennium BCE for unclear reasons but their
power had anyway always been more cultural than political. The southern part of Mesopotamia remained for
several centuries a land united by a common Sumerian culture but where rivalling city-states often fought each
other and remained largely independent. In this sense, Mesopotamia after Uruk was more similar to the highly
fractionalized pre-Hellenic Greek system of city states than to neighboring Egypt, which was led by a single
ruler. Sargon of Akkad was the first Mesopotamian ruler to unite the whole area under one reign around 2350
BCE, but the Akkadian domination turned out to be short-lived. By 2200, both the Egyptian and the
Mesopotamian states collapsed and foreign rulers (Gutians) assumed power in the Sumerian heartland.
This was perhaps the first of many state collapses in which invading peoples from neighboring areas played
an important role. By 2030, the domination of the highly bureaucratic regime in Ur was ended by invading
Amorites from Iran. The Old Babylonian empire that emerged after Ur was in turn conquered by yet another
eastern people, the Kassites, in 1595. An even greater disaster for the Western core would happen around
1200 BCE when more or less all the established states of the period collapsed, including Egypt, the Mitanni
and Hittite empires, and the Mycaenean Greek state. In Mesopotamia, Assyria initially could expand but soon
fell into the general decline. The city of Babylon was sacked by Elamites from Iran.
Exactly what happened during this Dark Age has much debated and explanations range from natural
disasters to massive migrations. The outcome was basically the same everywhere; burned and destroyed cities,
depopulation, a protracted break in written records, collapse of long-distance trade, and an abandonment of
agricultural land (Morris, 2010). Not until the brief rise of the neo-Assyrian empire around 750 BCE would
the current territory of Iraq be under a single ruler again, and by this time the political and technological
leadership of the western world had shifted westwards towards the Mediterranean.
In 539 BCE, the Persians under Cyrus II plundered Babylon and initiated a period of more than a thousand
years when Mesopotamia was completely under the domination of foreign powers; most often under Persian
rule but also briefly under the Seleucid, Roman and Sassanid empires (see figure 6). Almost all of the major
autocratic empires of the Mediterranean seemed to want to conquer Mesopotamia during some part of their
expansion. The time of independent city-states was long gone. 3
The long era of Arab Muslim rule started in 638 CE when the Sassanids were defeated. A mass immigration
of Arabs took place in this period. The following centuries witnessed a revival of social development and the
newly founded capital of the Abbasid caliphate, Baghdad, became a center of the Muslim intellectual world.
The caliphate reached a peak the period 750-800 CE in terms of the extension of statehood and it is believed
that the city hosted nearly a million people.
However, once again, a foreign aggressor would invade Mesopotamia, as had happened so many times
before. The Mongols lay Baghdad under siege in 1258 and eventually conquered and looted the city and
murdered a large share of its population. Some sources claim that at least 100,000 people, perhaps much more
than so, were slaughtered and palaces and libraries were destroyed. The Mongol destruction initiated a new
period of depopulation and agricultural decline, perhaps due to irreparable damages to the sensitive irrigation
system in the area. The sacking of Baghdad was a major blow not only to Arab civilization in Iraq but also to
the Muslim world as a whole.
From 1533, the territory of modern Iraq would fall under the domination of yet another foreign power, the
Ottomans, who would rule the area more or less effectively until 1918. During this period, Iraq was divided
into three provinces (vilayets) - Mosul, Baghdad, and Basra - a division which would play an important role
since then. At its peak, the Ottoman empire was the largest in the western world since Rome. The Ottomans
put the Iraqi provinces under direct rule shortly after their conquest, but during the protracted disintegration
of the Ottoman empire after its peak around 1600, local rulers would frequently struggle for power against
3
The rest of the section relies on Britannica (2012) for general dates and facts.
xxv
each other and against foreign powers. The British gradually asserted their influence during the 19th century in
the area. When the Ottoman empire collapsed after the World War I, Britain administered Iraq on a mandate
from the League of Nations until Iraqi independence in 1932.
Independence did not create political stability in Iraq. Frequent power struggles and a new intervention by
Great Britain during World War II followed, and military coups took place both in 1958 and in 1968. The
latter brought the Baath-party of Saddam Hussein to power. Iraq established closer ties to the Soviet union
and oil started to generate substantial revenues to the government. After 1980, three successive wars in the
Gulf area (the Iran-Iraq war, 1980-1988, and the Persian Gulf Wars in 1991 and 2003) would however lead to
an economic deterioration of the country and to the dissolution of Saddam Hussein's regime. From 2003 to
2011, the country was occupied by yet another foreign power on Mesopotamian soil; the United States.
In summary, we believe the Iraqi example has the following key components in relation to our model: It
describes an area that was one of the first in the world to develop sedentary agriculture by 9000 BCE and
statehood and civilization already by 3500 BCE. Although Mesopotamia started off with fairly independent
city-states, government was highly autocratic, inequality was high and wealth was highly concentrated in
temples and palaces. The rich Mesopotamian cities soon became the object of predation by numerous
neighboring tribes and foreign armies with subsequent state collapses. By medieval times, the polities in the
Iraqi territories were clearly stagnant. Throughout the 20th century, internal and external rent seeking have
been the norm. Although a kind of parliamentary democracy is now in place, Iraq as of 2012 is by many
regarded as one of the failed states of the world.
Sweden
At about the same time that a full-blown civilization with a socially stratified state, codified writing, massive
public monuments, and a centralized "palace economy", administered by bureaucrats, was emerging in
Mesopotamia (i.e. middle 4th millennium BCE), the first small farming communities established themselves
along the coasts of what is now Sweden, employing the agricultural food package from the Fertile Crescent;
wheat, barley, pigs, goats, and cattle. Recent evidence strongly indicate that it was migrants from southern
parts that introduced agriculture and that these migrants later blended with the local hunter-gatherer
population (Skoglund et al, 2012). There are clear links from this early farming culture with other communities
in northern Europe at the time. 4
Unlike the riverine type of farming practiced in Mesopotamia, agriculture in Sweden was always rain fed,
implying a weaker need for a centralized organization of the economy. As elsewhere in the Atlantic parts of
Northern Europe, the construction of megalithic tombs were a common feature of early farming
communities, suggesting an increased reverence for the dead and an emerging organization for the provision
of public goods. Copper was used by these early farmers but the metal was not widespread and presumably
imported. Bronze was not widely used until about 1500 BCE, a change which most likely revolutionized
warfare and social organization. During the latter part of the period, very large burial mounds suggest an
increase in social stratification although not beyond tribal level (Welinder, 2009).
The use of iron became widespread in Sweden around 500 BCE. The Scandinavian countries did not
become part of the Roman empire but were nonetheless affected by developments on the continent. In
particular, the collapse of the Western Roman empire in the 5th century CE and the massive migrations
during the period, resulted in a period of social unrest in Sweden, manifested in the construction of multiple
defensive fortifications.
4
The culture is referred to as TRB and shares many features with the so called "Ertebölle" culture of Denmark from the
same period (Bellwood, 2005).
xxvi
By 550 CE, a more peaceful period referred to as "Vendel" started in Sweden when tribal chiefs in various
parts of the country could establish rule over extended regions and were buried in impressive burial mounds.
The greater concentration of wealth was due to a centralized control over iron mining and the establishment
of trading centers which greatly increased imports and exports. The chiefs controlled forces of mounted elite
warriors with costly weaponry, offered pagan ceremonies in wooden halls, and oversaw the production of
jewelry in gold (Welinder, 2009). Primitive writing in the form of runes were carved into stone.
In the latter part of the 700s, the trading post Birka was founded in Lake Mälaren by a local unidentified
chief and quickly emerged into the first urban settlement of Sweden. However, even at its peak, the size of the
population still did not exceed 1000 individuals. Traded goods from many parts of the world have been
excavated, including Arabic countries. Christian missionaries appeared in the town in the 830s but failed to
establish Christendom in the area.
By 800 CE, Sweden was still very much a periphery in the Western world; culturally, economically and
politically. However, population growth had been rapid since the great migrations ended and local chiefs had
accumulated the power to equip troops with expensive weaponry and horses and to build long ships that were
capable of making long sea voyages. In 793, a fleet of such ships reached the monastery of Lindisfarne in
Northern England, which was subsequently looted and the local population terrorized. This attack is usually
regarded as marking the beginning of the Viking Age in the Scandinavian countries.
The Viking attacks on the more developed parts of Europe that started around 800 basically followed the
same pattern as so many other attacks by primitive peoples on more advanced core regions. The wealthy cities
of Mesopotamia were overrun by neighboring tribes who had adopted the basic military technology of the
center but were still not integrated into its urban civilization. In the same manner, the Vikings looted
numerous towns and monasteries in Atlantic Europe, around the Baltic Sea, as well as in the Mediterranean.
Vikings formed new kingdoms in England and Normandy, settled Iceland and Greenland, served as
mercenaries to the emperor in Constantinople, and even formed a temporary colony on New Foundland.
Eventually, the Vikings outside Scandinavia blended with the local populations, settled down and were
Christianized.
When Viking energy subsided in the 11th century, there were kingdoms and states in Norway and Denmark
but still not in Sweden. The two core areas, Svealand around Lake Mälaren and Götaland in the southwest,
were not easily united and were divided by deep forests. The country was eventually Christianized by 1100 and
not until around 1150 was the country united under a king. From this period onwards, Sweden is considered
to have had a government above tribal level but with a very weak state and a king who had to struggle to
maintain jurisdiction over local chiefs and independent bishops. Not until the era of Birger Jarl, a strongman
from Götaland whose son inherited the throne, was the Swedish state consolidated around 1250, almost five
thousand years after the initial emergence of agriculture.
By about 1250, one might argue that Sweden started to overtake Iraq in terms of state organization and
stability. In 1258, Iraq was overrun by yet another foreign people, the Mongols, an event which ended the
golden era of Arab culture in the region and possibly led to permanent damage to the sensitive irrigation
systems. In peripheral Sweden, in contrast, no foreign peoples ever sought to invade the sparsely populated
and culturally backward country, lacking proper cities and fertile farming lands. Traders from the German
towns by the Baltic Sea dominated economic life in Sweden throughout the Middle Ages but were never
interested in the land as such. As elsewhere in Europe, feudalism was the main economic organization of the
countryside, but peasants were generally a lot more independent than in for instance Eastern Europe. Unlike
any other country in Europe at the time, peasants were represented in parliament and often initiated forceful
rebellions when the burden of taxation was considered too high. In 1393, Sweden entered a union with its
more powerful neighbor Denmark which would remain until 1523, when the founder of the modern Swedish
xxvii
state, Gustav Wasa, defeated the Danes, renegotiated debts with the German trading towns, suppressed local
rebellions, and finally established Sweden as a fully sovereign state.
In the early 17th century, Sweden adopted Protestantism and rose in the Thirty Years' War (1618-1648) to
become a major military power in northern Europe, mainly due to its superior organization of taxation and
military service. Once again, Swedish armies pillaged continental Europe and contributed to the long decline
of the Holy German Empire. Also Poland was badly ravaged during the rule of the constantly campaigning
Charles X. New territories were conquered from Denmark in the south and substantial amounts of war booty
enriched a nobility from which most of the military commanders had been recruited. Inspired by the
autocratic rule of Louis XIV in France, the Swedish kings Charles XI and Charles XII ruled without taking
much advice from parliament. After the decisive Swedish defeat against Russia in 1709, Sweden's period as a
north European superpower was effectively over.
When Charles XII died in 1718, a new era started when the king had to secede substantial powers to the
parliament. Not even the attempts by Gustav III to restore autocratic rule were successful and the king was
even murdered in 1792. Sweden was clearly not characterized by social equality in the modern sense during
this period, but the farming population was still relatively free and protected by law. After its involvement in
the Napoleonic wars in the early 1800s, Sweden adopted a principle of neutrality and has not been engaged in
any war since.
Free from military threats domestically and campaigns abroad, the country entered a road towards an even
stronger parliament and a widened suffrage. A massive emigration of poor Swedish peasants to the United
States in the decades after 1850, eased a growing problem of overpopulation in the countryside. During the
strong social tensions after World War I, which swept away the old regimes in so many European countries,
universal suffrage of all men and women was introduced in 1921 and the Social Democratic Party was allowed
to dominate Swedish politics from 1932 onwards. Since the country was spared war damages in World War II
due to its neutrality, the speed of industrialization picked up even further when the demand for Swedish
goods surged. Extensive social reforms and progressive taxation turned Sweden into one of the most equal
societies in Europe. Not even the very serious economic crisis of the early 1990s altered the general trend of
rising prosperity.
By 2014, Sweden is one of the richest and least corrupt countries in the world and has the strongest public
finances in the European Union. 5 Like in the neighboring Scandinavian countries, there are no signs of social
unrest, even in the wake of the financial crisis.
In summary, Sweden's late transition to agriculture and to statehood, in combination with its geographical
peripherality and relatively unproductive farming lands, implied that the country was not a natural target of
predation by neighboring peoples or by rent seeking local lords. Wealth and power was relatively evenly
distributed through most of history and not even the attempts by the militarily powerful kings of the 1600s
succeeded in permanently establishing autocratic rule. Due to the great degree of social cohesion, the social
reforms after World War II could be peacefully installed without alienating capitalists and the upper classes. In
so many ways, Sweden's history was the direct opposite of that of Iraq.
Additional references:
Britannica (2012) "Iraq." Encyclopædia Britannica Online Academic Edition. Encyclopædia Britannica Inc.,
2012. Web. 21 Jun. 2012.
5
In 2005, real GDP per capita was almost 45 times higher in Sweden than in Iraq ($31,271 versus $690).
xxviii
Skoglund, P. et al (2012) "Origins and Genetic Legacy of Neolithic Farmers and Hunter-Gatherers in Europe"
Science, 336, 27 April, 466-469.
Welinder, S. (2009) Sveriges historia: 13000 f Kr till 600 e Kr, Stockholm: Norstedts Förlag.
Yoffee, N. "Political Economy in Early Mesopotamian States" Annual Review of Anthropology 24, 281-311.
xxix