A Western Reversal Since the Neolithic?
The Long-Run Impact of Early Agriculture
Ola Olssony
University of Gothenburg
Christopher Paikz
NYU Abu Dhabi
July 2, 2014
Abstract
In this paper, we document a reversal of fortune within the Western 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 some recent in‡uential work emphasizing the bene…cial role of an early transition. Using data from
a large number of carbon-dated Neolithic sites throughout the Western agricultural
core area, we determine an approximate date of transition for about 60 countries and
1400 regions. Our empirical analysis shows that there is a robust negative correlation
between years since transition to agriculture and contemporary levels of income on the
national level whereas the regional result depends on the inclusion of country …xed
e¤ects. Our results further indicate that the reversal had started to emerge already
before the era of European colonization. Countries that made an early transition
to agriculture experienced a delayed adoption of democracy, which appears to be an
important intermediate channel.
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
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
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
of the seminal works in the literature proposing a positive relationship between an early
agricultural transition in history 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.
By creating a comprehensive data set on the Neolithic transition for all Western areas,
we con…rm that regions that 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. 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. On a …ner
regional level, using data on 1,409 European NUTS3-regions, there is likewise a very
strong overall negative relationship, although the relationship vanishes when we control
for country …xed e¤ects. Most of the action thus derives from the cross-country level.
But even within several large countries such as Italy, Turkey, and Spain, there is a clear
negative association between time since agriculture and current economic performance.
What is the explanation to this reversal? Why did an early transition to agriculture
and civilization become a curse during the last half millennium? We argue that a key
intermediate variable for understanding this process is democracy. After 1500 CE, democratic institutions such as constraints against the executive and a more inclusive sharing
of power gradually won territory in several European countries. We propose that the
adoption of democratic institutions was more di¢ cult in countries that made an early
transition to agriculture. These countries had for millennia been characterized by collectivist, autocratic societies where powerful elites had extracted rents from their people
since the …rst emergence of states. In the initially backward northern periphery, on the
other hand, the more egalitarian and individualist-minded populations found it easier to
accept democratic institutions.2 Once these were in place, they proved to be a superior
mode of social organization which greatly boosted countries’…scal capacity, their ability
to conscript soldiers, and individual incentives for innovation and investment. Within a
few centuries, the countries in the periphery had overtaken the old civilizations in terms
of prosperity.
Our empirical results con…rm that the time since Neolithic transition has a strong
negative impact on our measures of historical levels of democracy. Furthermore, the
1
2
BP refers to Before Present, where the present period is set to be the year 2000.
See Lagerlöf (2014) for a theoretical model which makes the same basic point.
2
association between time since transition and current levels of income is substantially
weakened when democracy is included, suggesting that it was indeed a very important
intermediate channel of causation, though perhaps not the only one.
Our paper is related to a vast literature on Western economic and social history.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
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
3
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).
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).
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.
3
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.
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 speeded 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. Voigtländer and Voth (2012) use model simulations and their focus
is on explaining how Europe as a whole became richer rather than how some countries
overtook others.7
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
a more bene…cial economic development.
We believe our paper makes the following important contributions to the literature.
First, we provide strong evidence of a distinctive reversal of income and development levels
7
In our empirical analysis, we address whether the Black Death could be a potential intermediate
mechanism. However, any assessment of the impact of the Black Death is hampered by the notoriously
unreliable quality of mortality data from the period.
4
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, we point to one likely intermediate factor behind the reversal; the development of
democracy which diverged dramatically starting in 1500 CE.
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 provide an overview of the relevant literature and present the basic
theoretical framework behind 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) outlines a theory arguing that societies that 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.8 When the transition to sedentary agriculture was complete, societies typically embarked on a path towards civilization that eventually resulted in codi…ed language,
large public monuments, organized religion, 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 clear 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.
8
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
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 relatively 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 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 north European neighbors) is now 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.
2.2
Our model
In an extremely simpli…ed sense, our basic model of the causal relationship between the
time of agricultural transition and current income per capita in the Western core is shown
in Figure 2. The …gure shows historical time on the horizontal axis combined with a
proposed causal chain of time-invariant and historical variables.
Figure 2
Exogenous factors Xi such as geography, biogeography and climate are shown in the
top of the graph because they are not generally functions of time but still a¤ect historical developments in various ways. The …rst link in the …gure posit 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 where it is argued that speci…c
geographical and biogeographical characteristics implied that the …rst agricultural transition happened in the Fertile Crescent. In this paper, we will not provide an exhaustive
analysis of why the original domestication of plants and animals happened where it did.9
9
See for instance Smith (1995) or Bellwood (2005) for general treatments on the rise of agriculture.
6
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 shown in
Figure 2. 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 3 and in Appendix A of this paper, we document for instance that the
time since transition to agriculture for country i or region j is negatively associated with
distance to the Fertile Crescent. Time since transition further 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
northwesterly direction.
The horizontal set of arrows in Figure 2 contain our main hypothesis: That the time
since the Neolithic transition to agriculture had a lasting causal in‡uence on current
levels of income. Most likely, this e¤ect was indirect and came about through one or
more observable or unobservable intermediate historical factors (referred to as Zt in the
…gure). It should be noted however that the interpretations of these main relationships
are confounded by the fact that exogenous factors such as climate very likely also have a
direct e¤ect on the 2005 levels of income per capita, as displayed in Figure 2, apart from
in‡uencing the time of the transition.
In this article, our focus will be on documenting a reduced-form relationship between
time since agricultural transition and current income, controlling for exogenous factors, on
cross-country, cross-regional, and within-country levels of analysis. This Western reversal
is a striking fact in itself that has not been su¢ ciently recognized in the literature. We will
also pay some attention to the issue of when the reversal actually happened. For instance,
by 500 CE, most indicators for social progress would certainly put the Byzantine world far
ahead of the hunter-gatherer societies of Scandinavia, but 1000 years later, that was not
obviously the case any more. At some point during the last millennia, a long history of
agricultural production ceased to be an economic advantage for the old civilizations and
instead became a serious disadvantage to economic development. What is the intermediate
historical mechanism that made a long history of agriculture a disadvantage?
We argue that the most likely mediating factor between the agricultural transition
and current income is the old civilizations’ delayed adoption of democracy during the
Modern Era (1500 CE -). More generally, we hypothesize that a cluster of institutions
and norms oriented towards individualism rather than collectivism, constitute the main
intermediate channel of causation between the agricultural transition and current income
levels. Furthermore, democracy only became a key competitive factor in the Western core
during the last half millennium.
7
Our emphasis on democracy 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 (2005, 2008, 2012) have
in a series of contributions shown that e¤ective constraints on rulers and freedom from
expropriation have been crucial for the willingness to invest and to pursue productive
activities. Extractive institutions, favoring a narrow elite in power, can be associated with
temporary spells of economic progress but will not be sustainable in the long run. Only
in countries with inclusive, democratic institutions will the incentives for broad segments
of society be conducive to economic growth based on innovation and long-term contracts.
Other authors have shown that states with a greater degree of democracy were also more
capable of raising sovereign debt (Stasavage, 2011) and to conscript soldiers for successful
wars (Tilly, 1990). These works, combined with the argument in the literature for the
long-run persistence of early institutions, suggest that modern economic outcomes can be
determined by which type of (pre)historical institutions were established.
But why did early agricultural transitions lead to autocratic forms of government
whereas later transitions implied an easier path towards inclusive institutions? In a famous treatise, Wittfogel (1957) argued that it was the type of irrigation practiced in the
river plains of Euphrates, Tigris, and the Nile, with its particular need for centralized,
coordination collective action, that provided the fundament for the emergence of despotic
states in Mesopotamia and Egypt. In a cross-country study, Bentzen et al (2012) demonstrate that irrigation potential in countries, proxied by potential yield gain from irrigation,
is strongly correlated with autocratic forms of government. Their analysis suggests that
although irrigation might not necessarily give rise to states, countries with high irrigation
potential tend to be more autocratic than countries that rely on rain-fed agriculture.10
In a recent paper, Lagerlöf (2014) presents a model in which ruling groups in the old
agricultural civilizations invest in public goods like cities and infrastructure but also in
an "extractive capacity" that allows them to collect rents for themselves. These rents are
socially wasteful and mean that autocratic states do not reach their full growth potential.
When the idea of democracy is introduced as an exogenous shock, the heavy investments in
extractive capacity in the old civilizations imply that they are more resistant to democracy
than younger states. Democracies, however, have a more e¢ cient provision of public goods,
which eventually makes them overtake the older civilizations.
In a companion paper to this one, Olsson and Paik (2013) argue that it was probably
the most non-conformist, individualistic farmers from the agricultural center who migrated
and colonized new lands. After several successive waves of colonizations and migrations,
our model predicts that the people who eventually settled in the periphery of the Western
zone were culturally much more individualistic-minded than those who remained in the
founder societies in the Fertile Crescent.
Also other reasons most likely played a role for the easier transition to democracy in the
late transition societies. Not only did the old civilizations spend more on an extractive ca10
See also Jones (1981) who makes a similar point regarding the advantages of rain-fed agriculture.
8
pacity visavi their own people; they would soon also have to devote substantial resources to
defend themselves against the recurring invasions by barbarian peoples who were strongly
attracted by the accumulated wealth in the densely populated farming areas. Such circumstances were not conducive for the emergence of democracy or a culture of trust, if
one considers the likely centralized wealth accumulation in these early states . Given the
autocrat’s control over the wealth, state-building spurred on by external threats would
have led to more investment in maintaining the hierarchy, and less enfranchisement of the
people. In the periphery, on the other hand, accumulated agricultural wealth was negligible and roving bandit armies rarely found it worthwhile to attempt predatory invasions.
In these relatively egalitarian backwaters of Western civilization, democratic institutions
would take root more easily.11
In conclusion, our argument is that time since the transition to agriculture should have
a negative, reduced-form impact on current levels of development and that this relationship
should mainly be intermediated through levels of democracy and inclusive institutions.
3
Data
The key explanatory variable in the empirical analysis is the time since the Neolithic
transition.12 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
11
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. See also the working paper version Olsson and Paik (2012) for a more extensive account of
the di¤erences between regions in the center and periphery of the Western core area.
12
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.
9
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.13 Figure
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 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,14
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.
13
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.
14
In our study, N is set to 15.
10
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 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.15 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
as the average of all the estimated adoption date of location cells within that subnational
unit.1617
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 crossregional sample is 7,055 years with a range from 5,243 to 11,320. This translates into a
mean adoption date of 5,055 BCE and a …rst adoption date of 9,320 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).18 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.
15
As explained in Johnston (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.
16
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.
17
Figure A3 in the Appendix shows an example of IDW methodology applied to the 78 Neolithic sites
within Italy.
18
We average over the calculated scores for all the cells within each country to get the country score.
11
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 geodesic distance from Jericho
to each archaeological site in our sample.19 More 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 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 3 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 3
As the …tted line in Figure 3 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.20
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 and 0.74 in the regional
sample.
19
In their investigation of the most likely centers of original domestication, Pinhasi et al (2005) …nd that
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.
20
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.
12
4
Empirical strategy
The basic theoretical framework that guides our empirical strategy was shown in Figure
2. 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.
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 B2.21
Is it possible that there were pre-existing institutional 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.22
Although we recognize this possibility of an unobservable, omitted variable problem in
equation (2), we …nd no reason to believe that such variations were systematically related
to either Di ; Xi or Yi 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
+ "i
(3)
21
A noteworthy feature of the results in this table is that Distance to Jericho changes 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.
22
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.
13
where Yi is income per capita in country i in our Western sample 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
1
+ Xi0 (
1 2
+
2)
+
1 i
+ "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
might be noted that the sum of the indirect and direct e¤ects of Xi is
1 2
+
23
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.24 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.25 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 in Di which makes it possible to specify
a 2SLS-regression framework where equation (2) is the …rst stage and equation (3) is the
second stage, where Di serves as an instrumental variable. The exclusion restriction is
as always that Cov ("i ; Di ) = 0, i.e. Di should only a¤ect the outcome variable through
Ti . We do not see any reason 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 2, our hypothesis is that time since transition to agriculture
should mainly a¤ect current economic performance through inclusive institutions and
23
The indirect e¤ect will not be captured 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 .
24
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.
25
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.
14
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
+
(5)
i
< 0 implies that an early transition to agriculture retards the adoption of democ-
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 that 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
+ 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, 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 . As will be shown below, the choice whether to include this …xed
e¤ect or not in the regional analysis will have important consequences for the results. The
key parameter of interest is also here
5
1,
which we expect to be negative.
Results
5.1
Cross-country analysis
Table 1 shows the results associated with our 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 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 have consistently 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
15
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. 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.
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 large
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 4 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
would have resulted in 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 4
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 a 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).
16
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
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 B3 of Appendix B, 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 esimates in Table
1, Columns (1) and (3).
In Table 3, we check the robustness of our main results to inclusion of a list of other
historical variables Zi that have been proposed/used in the literature, while also controlling
17
for geography. In terms of Figure 2, what we are checking for is alternative intermediate
mechanisms apart from democracy, which will be analyzed in a later section. 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.26 The estimates of the historical variables
are shown in the second column. For brevity, we have omitted to report stardard errors.
Table 3
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.27 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. Trust in Row (8), capturing the strength of generalized
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
26
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.
27
The latter Mongol variable, we have coded ourselves as described in Appendix F.
18
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.28 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 seem 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
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; Voigländer 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.29
In summary, Table 3 rules out a number of alternative historical hypotheses in the
literature and shows that our main variable is robust to most of them. However, inclusion
of Muslim and Protestant populations and 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
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
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
28
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.
29
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, were magni…ed and pushed northwestern Europe into a take-o¤.
19
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 good indicator
for the level of economic development in a country (Ashraf and Galor, 2011). In Table 4,
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 4
In Columns (4)-(9), we instead use Log GDP per capita from 1, 1000, 1500, and 1820
CE as the dependent variables.30 In (7) and (9), we also include our geographical controls.
The sample is now limited to only 20-28 countries but a quite similar pattern emerges in
these regressions. The time since agricultural transition even has a positive and signi…cant
in‡uence on income in year 1000 CE, whereas the reversal becomes visible by 1500 CE,
although the result for that century is not robust when controls are included. From 1820,
the relationship is unambiguously negative.
When we use income per capita data also for 1600, 1700, 1913, 1980, and 2005 and run
the same bivariate regressions as in Table 4, we can track more in detail how the regression
coe¢ cient of Average time since agricultural transition develops over time. Figure 5 shows
the patterns, with and without controls, suggesting a stronger and stronger tendency for
a reversal after 1500.
Figure 5
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. Below, we will brie‡y analyze how democracy seems to have been an intermediate
variable already by the start of the Modern Era.
30
Detailed variable sources and data descriptions are available in Appendix D.
20
5.3
Cross-regional and within-country analysis
Table 5 shows the results for the cross-regional analysis among about 1,400 Western
NUTS3-regions in 30 European countries.31 In this table, we include an even broader
set of geographical control variables including size of region area, latitude, elevation, share
of arable land in 1600 CE and an Atlantic dummy. The dependent variable is Average log
GDP per capita 1997-2008.
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, when we include …xed
e¤ects in Columns (3)-(4), the sign is no longer signi…cant. The same basic pattern applies
in Columns (5)-(6) where we instead use Predicted time since agricultural transition as
the main explanatory variable.
Table 5
In order to analyze these contrasting results further, we show the unconditional bivariate scatter plot for all 1,371 regions in Figure 6. 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.32 For other countries in the sample, there is no signi…cant
within-country relationship in either direction.
Table 6
Figure 6
Our interpretation of these results is that most of the action in the relationship between
current income levels and time since agricultural transition comes from the cross-country
level rather than from the cross-regional level, although the strong negative correlations
within countries like Turkey, Spain and Italy are interesting in themselves. One potential
interpretation of the general tendency is that the average time since transition within a
31
Please see Appendix D2 for further details about the variables included in the regional analysis.
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.
32
21
country has had a strong in‡uence on that country’s comparative average level of development (as proxied by GDP). The government in each of the smaller countries, with a
low number of intrastate regions in it, managed to also even out the initial di¤erences
from di¤erent dates of agricultural transition, perhaps by installing common institutions
throughout the country.
5.4
Intermediate causal channel
So far, our focus has been on documenting a reduced-form relationship between time since
agricultural transition and current wealth. In this …nal section, we demonstrate that an
important intermediate channel between the timing of Neolithic agriculture and current
levels of economic development is levels of inclusive institutions such as democracy during
the Modern Era.
When we studied the timing of the reversal in section 5.2, the results suggested that
the negative impact of time since agriculture seemed to have emerged sometime during
the last 500 years. In Table 7a, 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.33
Table 7a
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 7a is that they are consistent with the hypothesis that a long
agricultural history provided an obstacle to institutional development already in 1500 and
that constraints 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 think the results from Table 7a should be interpreted with
caution. In order to obtain a more reliable estimate of historical democratic institutions,
we introduce Democracy stock 1900-2000 from Gerring et al (2005). As described in
33
See Appendix D1 for a description of the historical variables used in this section.
22
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.34 Given that our main dependent variable below is
GDP per capita in 2005, the endogeneity problem should be alleviated.35
Table 7b shows the results. In Columns (1)-(2), Democracy stock 1900-2000 is the
dependent variable. In line with our hypothesis of democracy being an 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 7 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 7b
Figure 7
When we include both Average time since agricultural transition and Democracy stock
1900-2000 as regressors in Column (4), Democracy stock is still highly signi…cant and
positive whereas Average time becomes insigni…cant. The conditional scatter plots for
this regression are shown in Figure C8 in Appendix C.
In Column (5), 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 3. However, when we employ
them here alongside Democracy stock 1900-2000 and the standard geographical control
variables in Column (5), neither of their coe¢ cients are signi…cant whereas the democracy estimate is. The same tendency remains when we also include Trust in Column (6).
Lastly, in Columns (7)-(8), we include instead Predicted time since agricultural transition
which turns out to be insigni…cant whereas the coe¢ cient for Democracy is still measured
with precision.36
34
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.
35
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 think that
the correlations reported in Table 7b are still useful for a deeper understanding of the historical linkages.
36
Additional evidence on individual level of a long-run impact of the agricultural transition on democratic
23
In summary, our interpretation of these results is that democracy is an important
intermediate variable between time since agricultural transition and current prosperity.
Given the mixed levels of signi…cance for our agricultural transition variable, we do not
claim, however, that democracy is necessarily the only channel. 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 among European
regions, although this result is not robust to introducing country …xed e¤ects that can
capture the level of democracy at the state level. 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 a key intermediate channel
between agricultural transition and current income levels is inclusive institutions such
as democracy. 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 leave it for future research to probe deeper into the exact mechanisms that
caused the Western reversal.
References
[1] Acemoglu, D., S. Johnson and J. Robinson (2002) "Reversal of Fortune: Geography and
Institutions in the Making of the Modern World Income Distribution" Quarterly Journal
of Economics 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.
norms, are provided by Olsson and Paik (2013). The paper shows that people in regions that transited
early to agriculture value obedience more and feel less in control of their own lives than people in regions
that made a late transition.
24
[4] Adams-Faure (1997) "Preliminary Vegetation Maps of the World since the Last Glacial Maximum: An Aid to Archaeological Understanding" Journal of Archaeological Science 24,
623-624.
[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 (2011) "Dynamics and Stagnation in the Malthusian Epoch" American Economic Review 101(5), 2003-2041.
[8] 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.
[9] Ashraf, Q. and S. Michalopoulos (2011) "The Climatic Origins of the Neolithic Revolution:
Theory and Evidence" working paper.
[10] Bellwood, P. (2005) First Farmers: The Origins of Agricultural Societies, Oxford: Blackwell
Publishers.
[11] Bentzen, J.S., N. Kaarsen, and A.M. Wingender (2012) "Irrigation and Autocracy" University
of Copenhagen, working paper.
[12] 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),
[13] Bleaney, M. and A. Dimico (2011) "Biogeographical Conditions, the Transition to Agriculture,
and Long-Run Growth" European Economic Review 55(7), 943-954.
[14] Borsch, S. (2005) The Black Death in Egypt and England: A Comparative Study, University
of Texas Press.
[15] 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.
[16] Bowles, S. and J-K Choi (2013) "The Origin of Private Property" working paper, Santa Fe
Institute.
[17] 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.
[18] Clark, G. (2008) A Farewell to Alms: A Brief Economic History of the World, Princeton
University Press.
[19] Conley, T. (1999) "GMM Estimation with Cross-Sectional Dependence," Journal of Econometrics 92, 1-45.
[20] Diamond, J. (1997) Guns, Germs and Steel: The Fates of Human Societies. New York: Norton.
[21] Euratlas
(2012)
"History
of
Europe",
http://www.euratlas.net/history/europe/1/index.html.
online
resource,
[22] Eurostat (2012) http://epp.eurostat.ec.europa.eu/portal/page/portal/region_cities/introduction.
[23] Gerring, J., P. Bond, W. Barndt, and C. Moreno (2005) "Democracy and Economic Growth:
A Historical Perspective" World Politics 57, 323-364.
25
[24] Hansen, C.W., P.S. Jensen, C. Skovsgaard (2013) "Modern Gender Roles and Agricultural
History: The Neolithic Inheritance" University of Southern Denmark, working paper.
[25] Harlan, J.R. (1995) The Living Fields: Our Agricultural Heritage, Cambridge University
Press.
[26] 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.
[27] Johnston, K., J.M. Ver Hoef, K. Krivoruchko, and N. Lucas (2003) ArcGIS: Using ArcGIS
Geostatistical Analyst. Redlands, CA: ESRI.
[28] Jones, E.L. (1981) The European Miracle: Environments, Economies and Geopolitics in the
History of Europe and Asia. Cambridge: Cambridge University Press.
[29] 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.
[30] Kuran, T. (2011) The Long Divergence: How Islamic Law Held Back the Middle East, Princeton University Press.
[31] Lagerlöf, N. (2014) "Statehood, Democracy and Preindustrial Development" working paper.
[32] Landes, D. (1998) The Wealth and Poverty of Nations: Why Some Are So Rich and Others
So Poor, New York: W.W. Norton.
[33] Maddison, A. (2013) "Statistics on World Population, GDP and Per Capita GDP, 1-2008
AD" www.ggdc.net/maddison/Maddison.htm.
[34] Michalopoulos, S. (2012) "The Origins of Ethnolinguistic Diversity" American Economic Review 102(4), 1508-1539.
[35] Mokyr, J. (1990) The Lever of Riches: Technological Creativity and Economic Progress. Oxford: Oxford University Press.
[36] Morris, I. (2010) Why the West Rules - For Now: The Patterns of History and What They
Reveal About the Future, London: Pro…le Books.
[37] North, D. (1990) Institutions, Institutional Change and Economic Performance. Cambridge
University Press.
[38] Olsson, O. and Hibbs, D.A. (2005) "Biogeography and Long-Run Economic Development"
European Economic Review 49(4), 909-938.
[39] Olsson, O. and C. Paik (2012) "A Western Reversal Since the Neolithic? The Long-Run
Impact of Early Agriculture" SSRN Working Paper http://ssrn.com/abstract=2206198
or http://dx.doi.org/10.2139/ssrn.2206198.
[40] Olsson, O. and C. Paik (2013) "Long-Run Cultural Divergence: Evidence from the Neolithic
Revolution" working paper.
[41] Pinhasi, R., J. Fort, and A. Ammerman (2005) "Tracing the Origins and Spread of Agriculture
in Europe" PLOS Biology 3(12), 2220-2228.
[42] Pomeranz, K. (2000) The Great Divergence: China, Europe, and the Making of the Modern
World Economy. Princeton: Princeton University Press.
[43] Putterman,
L. (2006) "Agricultural Transition Data Set" online resource,
http://www.econ.brown.edu/fac/louis_putterman/agricultural%20data%20page.htm.
[44] Putterman, L. (2008) "Agriculture, Di¤usion and Development: Ripple E¤ects of the Neolithic Revolution" Economica 75(300): 729-748.
26
[45] 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.
[46] Smith, B.D. (1998) The Emergence of Agriculture, New York: Scienti…c American Library.
[47] Spolaore, E. and R. Wacziarg (2009) "The Di¤usion of Development" Quarterly Journal of
Economics 124(2), 469-529.
[48] Stasavage, J. (2011) States of Credit: Size, Power and the Development of European Polities,
Princeton University Press.
[49] Tilly, C. (1990) Coercion, Capital and European States: AD 990-1990, London: Blackwell.
[50] 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).
[51] Weisdorf, J. (2005) "From Foraging to Farming: Explaining the Neolithic Revolution", Journal of Economic Surveys 19(4), 561-586.
[52] Wittfogel, K. (1957) Oriental Despotism: A Comparative Study of Total Power. New Haven:
Yale University Press.
27
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.
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: Modeling of causal relationships
Distance to
agricultural origin
- Di
Agricultural
transition - Ti
Exogenous factors:
Geography, biogeography, climate Xi
Intermediate historical
mechanisms - Zi
9700 BP – 5600 BP
Economic
development - Yi
2005 CE
Time
Figure 3: 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 4: 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.
Figure 5: Historical evolution of relationship between log GDP per capita and average time since
agricultural transition among Western countries
Note: The figure shows the regression coefficients from 18 separate regressions for different time periods, with and
without control variables. The specification with no controls is given by Log GDP per capita (in year t) = β0 +
β1*Average time since agricultural transition (in year t) + ε (in year t) where β0 is a constant, β1 is the regression coefficient of
interest, and ε is a random error term. The 9 β1-estimates from these regressions are shown as grey circles where the
dashed lines display the 95% confidence interval. The specification with controls 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) where
everything is the same as above except for the included standard geographical controls. The 9 β1-estimates from
these regressions are shown as blue triangles where the dotted lines represent the associated 95% confidence interval.
The included years t (in the CE period with number of Western country observations in parenthesis) are 1 (23), 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 without controls for 1,
1000, 1500, and 1820 and with controls for 1500 and 1820 are shown in Table 4 and for 2005 in Table 1. Results
from the regressions for 1600, 1700, 1913, and 1980 are available upon request.
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.
Note: The figure shows the scatter plot for the 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 4. The regression coefficient for the whole sample is
shown in Table 3, Column 1.
Figure 7: 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 7b, column (1).
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)
[0.189]
[0.194]
1.733**
(0.777)
0.078
(0.188)
-0.632***
(0.223)
(5)
(6)
Dependent variable:
Log GDP per capita in 2005
-0.308**
(0.149)
-0.347**
(0.150)
(7)
(9)
-0.396**
(0.149)
[0.194]
-0.338***
(0.117)
[0.136]
-0.313**
(0.136)
[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)
-0.033
(0.027)
0.006
(0.007)
(8)
-0.326***
(0.107)
-0.757***
(0.109)
0.211
(0.158)
-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)
53
0.56
59
0.55
(7)
(8)
(9)
(10)
All
Sites>0
All
All
-0.219*
(0.115)
-0.449***
(0.118)
-0.436***
(0.127)
1.010***
(0.372)
0.367
(0.288)
Southwest Asia dummy
Fertile Crescent
dummy
(4)
-0.223**
(0.095)
-0.139**
(0.066)
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
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 include Log arable land area and Log distance to coast or river.
Table 3: 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
historical control
agricultural transition
Obs
(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
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 4: Historical evolution of relationship
(1)
(2)
(3)
Log Population density in:
1 CE
1000 CE 1500 CE
Average time since
agricultural transition
- in 1 CE
0.215
(0.142)
- in 1000 CE
(4)
(5)
(6)
(7)
Dependent variable:
Log GDP per capita in:
1 CE
1000 CE
1500 CE
(8)
(9)
1820 CE
0.076***
(0.015)
-0.001
(0.125)
- in 1500 CE
0.105***
(0.022)
-0.231*
(0.137)
-0.050*
(0.024)
-0.034
(0.049)
- in 1820 CE
Geographical controls
No
No
No
No
No
No
Yes
-0.180***
(0.033)
No
Observations
R-squared
54
0.04
60
0.00
61
0.04
23
0.30
20
0.69
21
0.07
21
0.26
28
0.36
-0.121***
(0.030)
Yes
28
0.60
Note: The estimator is OLS in all specifications. 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 include Log arable land area and Log distance to
coast or river.
Table 5: Cross-regional analysis among 1,409 European NUTS3-regions
(1)
Average time since
agricultural transition
Predicted time since
agricultural transition
Conley SE
(2)
(3)
(4)
(5)
Dependent variable:
Average log GDP per capita 1997-2008
OLS
FE
OLS
-0.484*** -0.451***
-0.063
-0.081
(0.021)
(0.034)
(0.136)
(0.096)
(6)
FE
[0.104]
[0.137]
[0.102]
[0.077]
-0.582***
(0.036)
[0.132]
-0.041
(0.120)
[0.092]
Geographical controls
No
Yes
No
Yes
Yes
Yes
Observations
R-squared
Number of Countries
1,371
0.23
1,371
0.42
1,371
0.01
29
1,371
0.17
29
1,409
0.53
1,409
0.15
30
Note: The estimator is OLS in columns 1, 2 and 5 and fixed effects (FE) with countries as the group variable in columns
3, 4 and 6. The sample is all Western regions with available data on NUTS3-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 and Predicted time since agricultural transition, we assume that spatial autocorrelation
exists among observations which are within ten degrees of each other. The geographical controls include Log area,
Latitude, Altitude, Arable land in 1600 CE, as well as an Atlantic dummy.
Table 6: Within-country relationships between average GDP per capita and time since agricultural transition
for NUTS3-regions in five large countries
Country
France
Germany
Italy
Spain
Turkey
OLS point estimate for
Average time since agricultural
transition
No
With
controls
controls
-0.197**
(0.075)
{0.098}
0.158***
(0.037)
{0.040}
-0.565***
(0.106)
{0.208}
-0.807***
(0.175)
{0.234}
-0.405***
(0.047)
{0.407}
-0.084
(0.063)
{0.444}
-0.079*
(0.040)
{0.396}
-0.251***
(0.062)
{0.802}
-0.508***
(0.251)
{0.439}
-0.241***
(0.053)
{0.531}
Observations
Pinhasi
sites
96
108
429
57
107
78
51
8
81
30
Note: The table shows estimated coefficients for the 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, R2-values for the regression 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 7a: Democracy 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
-0.260***
(0.055)
(3)
(4)
(5)
Dependent variable:
Log GDP per capita
in 1500
(6)
-0.050*
(0.024)
-0.020
(0.033)
0.222***
(0.071)
Yes
21
0.58
-0.244***
(0.064)
No
No
0.222***
(0.051)
No
47
0.23
21
0.17
21
0.51
No
0.004
(0.027)
0.225***
(0.060)
No
21
0.07
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 7b: Democracy stock 1900-2000 as a causal channel
(1)
(2)
(3)
Dem. stock 1900-2000
Average time since
agricultural transition
Predicted time since
agricultural transition
Democracy stock
1900-2000
Protestant population
-155.4***
(39.05)
-119.7***
(48.27)
-0.492***
(0.133)
(4)
(5)
(6)
Dependent variable:
Log GDP per capita in 2005
-0.114
(0.078)
-0.160
(0.096)
0.243***
(0.035)
0. 219***
(0.034)
0.001
(0.003)
-0.002
(0.003)
No
64
0.77
Yes
58
0.78
Muslim population
Trust
Geographical controls
Observations
R-squared
No
64
0.22
Yes
59
0.28
No
64
0.157
(7)
(8)
0.001
(0.108)
0.262***
(0.030)
-0.030
(0.099)
0.222***
(0.028)
No
65
0.48
Yes
60
0.72
-0.176
(0.124)
0.245***
(0.040)
0.004
(0.004)
-0.004
(0.005)
-1.306
(0.839)
Yes
46
0.83
Notes: The dependent variable is Democracy stock 1900-2000 in columns (1)-(2) and Log GDP per capita in 2005 in columns (3)-(8). Robust standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1. In columns (3)-(8), 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 include Log arable land area and Log distance to coast or river.
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
Calibrated
date (yrs BP)
St dev
Oldest 5 sites
M’lefaat
Hallan Cemi Tepsi
Cayönü
Mureybet
Wadi Faynan 16
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
(3)
Dependent variable:
Average time since
agricultural transition
-0.797***
(0.028)
(2)
9,255.381***
(81.295)
11,235.225***
(162.407)
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)
1,404
0.54
1,404
0.72
1,404
0.74
Latitude
Longitude
0.394***
(0.098)
-123.587***
(5.378)
56.772***
(7.007)
Log region area
Mean elevation
Arable land in 1600
CE
Atlantic dummy
Constant
Observations
R-squared
(4)
(5)
Predicted time since
agricultural transition
-54.545***
(2.701)
63.332***
(1.320)
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)
1,457
0.93
1,457
0.94
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
Table B3: 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
(3)
(4)
Dependent variable:
Log GDP per capita
in 2005
-0.468**
-0.558***
(0.190)
(0.157)
Average time since
agricultural transition
(IV – Dist. to Jericho)
Log arable land
Log distance to coast or
river
Observations
64
viii
-0.501***
(0.112)
-0.643***
(0.133)
59
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).
ix
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).
x
Figure C3: Average time since agricultural transition among 107 Italian regions
xi
Figure C4: Average GDP per capita 1997-2008 among 107 Italian regions
xii
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.
xiii
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.
xiv
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 7a, column (5). Each country is
identified by a three-letter isocode.
xv
Figure C8: 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 7b, column (4). Each country is
identified by a three-letter isocode.
xvi
Figure C9: Biogeography in the Western core 10,000 years ago
Notes: The figure shows the geographical distribution of different types of biogeography 10,000 years BP. The data from
this map is used in Table B1 where we analyze whether certain types of biogeography were associated with an earlier
transition to agriculture.
xvii
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).
xviii
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).
xix
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.
Democracy stock 1900-2000. Index capturing average levels of democracy during 1900-2000 according to the
Polity2 measure. The score for country i is calculated as Democracy stock 1900-2000 (i)
∑
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).
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.
D2: Variable definitions and sources for the 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:
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.
Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat, R. Wacziarg (2003) ”Fractionalization” Journal of
Economic Growth, 8: 155-194.
xx
G-Econ (2006), available online at http://gecon.yale.edu/.
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.
CIA World Factbook (2013), CIA
D3: 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)
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
xxi
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)
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)
64
0.174
0.357
0
1
64
0.163
0.350
0
1
64
0.267
0.383
0
1
0
1
64
65
51
52
65
47
0
0.291 0.128 0.111
0.606 0.151 0.290
-7.078 364.75 -604.6
1.34
0.635
1
1
0.663
0.887
637.6
3
Table D2: Summary statistics and sources of variables used in the cross-regional study
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.
Postgate, J.N. (1995) Early Mesopotamia: Society and Economy at the Dawn of History. London: Routledge
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
Appendix F:
The Mongol Invasion in Western Eurasia: History and Coding
1. Introduction
The Mongol empire during the 13th century under Genghis Khan,6 Ogedei, Mengke, and Khubilai was one
of the largest empires in world history. The sudden appearance of Mongol influence after 1206 and its equally
sudden disintegration makes it a particularly difficult case for any quanititative assessment of its lasting legacy.
Compared to, for instance, the Roman empire with borders remaining intact for centuries, the coding of
Mongol influence in countries and regions is a lot more complicated.
The aim of this appendix is to document which countries that either formally submitted to Mongol
authority or were otherwise directly affected through military campaigns. We first present a brief historical
background of the period in order to motivate our particular coding of a Mongol Empire dummy for each
country.
It is possible to study the Mongolian invasion of Asia, Europe and the Middle East from various sources,
but this memo focuses exclusively on English-language secondary sources. In places where there are
discrepancies either in description or years, we have deferred to May (2012) as it is one of the most clear
sources on the spread of the Mongol Empire, and is relatively recent in its sources and publication. Other key
sources for the account below are Library of Congress (2013) and Saunders (1971).
2. Historical Background
2.1 The Rise of the Mongol Empire
Accounts of the rise of the Mongol Empire typically start in 1206 CE when the Mongol ruler Temujin
united all Mongolian tribes and areas and let himself be crowned as Genghis Khan. For the next two decades
he expanded his kingdom beyond the steppe. In this early period, he focused on the Jin and Western Xia
empires in northern China. After a series of campaigns, all of the Western Xia was defeated by 1227 and the
Jin empire by 1234. The stage in East Asia was thereby set for the titanic struggle between the Mongols and
the Song empire in Southern China which would last several decades.
Skirmishes between the Mongols and the troops from the large Khwarezmian empire of current Iran,
Afghanistan, and Turkmenistan led Genghis to order a massive invasion which in 1220 resulted in the defeat
of the Khwarezmian dynasty. With no other major power to stop them, the Mongols continued to make
incursions into Georgia and Russia, marching north of the Caspian Sea to win a major battle against various
forces at Kalka river (in current southeastern Ukraine) in 1223. With these western borders pacified, the
Mongols made a temporary halt to further advances.
Genghis Khan died in 1227 and was succeeded by his son Ogedei. At his death, the Mongol empire
stretched from the Pacific to the Caspian Sea.
2.2 Expansion into Europe
6
Genghis Khan is also known as Chinggis Khan in typical Western sources. However, we will use Genghis Khan
throughout.
xxx
Under Ogedei, the Mongols pursued their most extensive campaigns westward which would have
enormous consequences for the affected states and regions. Led by the Mongolian general Subetei, an army of
600,000 Mongols followed the steppe corridor north of the Caspian Sea and advanced over the Volga into
Russia, defeating Moscow in 1237, as well as other Russian principalities. After a pause, the army then turned
south into current Ukraine where Kiev fell in 1240.
A northern flank of the Mongol army under Kaidu swept through current Belarus and Lithuania into
Poland. Detachments were also left in the Baltic regions. The city of Krakow was defeated in March 18, 1241.
A combined European army then met the Mongols in a major battle at Liegnitz in suthwestern Poland on
April 9. Once again, the numerically superior European were defeated, but this time they managed to offer a
stronger resistance.
A southern flank of the Mongolian army meanwhile advanced through Transylvania in current Romania. In
April, all the Mongolian army units met in central Hungary where they soon defeated a combined HungarianCroatian army, supported by Templar knights, in the battle at Sajo river. The Hungarian kingdom was on the
brink of a major disaster. Meeting no resistance, the Mongols then raided northwards in Bohemia and in
eastern Austria. By winter 1242, the Mongols were not far from Vienna when suddenly surprised Europeans
witnessed how the Mongols started to retreat eastwards. Ogedei Khan had died in December, 1241 and all
major army units were required by custom to return to Mongolia during the election (kurultai) of a new Khan.
One of the most remarkable historical twists of fortune was a fact.
During the retreat, the Mongolian armies passed slowly through the Balkan region (current Slovenia,
Croatia, Serbia, Bosnia), destroying the kingdom in Bulgaria before crossing the Danube. Most of the East
European regions were evacuated and the Mongols would never again try to conquer these areas. The impact
of the 1236-1242 campaign would however linger for a very long time, especially in the most heavily affected
Polish and Hungarian kingdoms.
2.3 The era of Mengke and Khubilai
After some years of internal struggles, Mengke Khan emerged as the sole ruler in 1251. Mengke's rule gave
new energy to Mongolian conquest ambitions. Prince Batu once again raided Poland, Lithuania and the
Baltics, but his army was too small to consolidate their victories.
In the southwest, Mengke's brother Hulegu had Persia and adjacent lands as his dominion and soon started
to consolidate Mongol rule in the area. He first turned to the Abbasid Caliphate in current Iraq and Kuwait,
which had so far not been conquered. Its capital Baghdad was considered to be the cultural capital of the
Islamic world. The fall and subsequent destruction of Baghdad in 1258 is generally considered to be one of
the major disasters of Islamic history. Baghdad would never recover the position it had before the Mongolian
attack.
Hulegu then continued west and conquered Syria and eastern parts of modern Turkey. In 1260, an army of
Mamluks from Egypt, supported by the Christian strongholds in the area, managed to defeat the Mongols at
the battle of Ain Jalut. This marked the southern border of the Mongol Empire which therefore partially
included current Lebanon and Syria but never Jordan, Israel, and Egypt.
During the 1250s, Mengke himself and his brother Khubilai had focused on conquering the Song empire of
southern China. In 1259, Mengke died during one of these numerous campaigns before a definite defeat of
the Song had been completed. He was eventually succeeded on the throne by his brother Khubilai Khan.
After another massive Mongolian effort in the 1270s, the Song empire was finally defeated in 1279. Khubilai
then formed the Yuan dynasty of China, which would rule there until 1368.
xxxi
2.4 Disintegration of Mongol rule
During the long rule of Khubilai Khan, the Mongols mainly concentrated their conquest efforts to East
Asia. Although they had successes in Burma and Thailand, they also suffered two bitter defeats against Japan,
which was never conquered. After the defeat against the Mamluks at Ain Jalut 1260, the Mongols did not
make any serious further incursions to the west. By 1300, Mongol direct influence reached to the current
Russian border and the eastern part of Ukraine in the north. In the south, the Mongols controlled Syria and
eastern Turkey, but not the area under Byzantium and the Mamluks.
In the Western parts, the huge empire started to split up into autonomous units during Khubilai's rule. In
the northwest, the Kipchak Khanate (the Golden Horde) would rule over Russia and parts of Ukraine to the
west of the Caspian Sea. By the late 14th century, the empire fell apart into many smaller units, some of which
would survive for several hundred years.
In the southwest, the Ilkhanate (covering eastern Turkey, Syria, Iraq, Kuwait, Iran, Afghanistan, Armenia,
Georgia, and parts of Pakistan) was founded in the 1250s and would be the major power in the area until
internal struggles caused the decline of the empire from 1335. Islam became the official religion in the area in
1295. As in most of the other regions, the relative peace of the last decade of the 13th century caused a major
upswing in international trade and exhange of ideas between east and west (Pax Mongolica).
3. Data
Our dichotomous coding of a Mongol Empire dummy is based on the following criteria. Countries are
coded with 1 if either (a) some part of their current territory were parts of the Kipchak (Golden Horde) or
Ilkhan empires by 1300 according to Euratlas (2012), or (b) they had at some date during the 13th century
experienced Mongol invasions with serious consequences for the country.
The first principle (a) that uses borders in 1300 from Euratlas (2012) exploits the same basic methodology
as our coding of the maximum extension of the other empires in our data set (Roman, Byzantine, Ottoman,
etc). However, compared to these empires, the Mongol conquests and campaigns were a lot more unstable
and short-lived. Even so, they often had profound consequences for the areas affected.
Apart from the first criterion of current borders within a Mongol empire in 1300 (a), we therefore include a
second criterion (b) of having been seriously affected by invasions during the years 1200-1300. For instance,
Poland and Hungary - both invaded during the massive Mongol campaigns of the early 1240s - were never
formally included within either the Kipchak or the Ilkhanate state borders. Nonetheless, it is well known that
the Mongol invasion had enormous historical consequences for both countries.7 Similar considerations have
induced us to code other East European countries such as Romania, Bulgaria, and Croatia as having been
under strong Mongol influence.
The list below presents the present-day countries that we consider to either have been part of a Mongol
empire in 1300 or having been seriously affected by Mongols during the 13th century. We also briefly
motivate our coding decision.
Country
Afghanistan
Albania
Armenia
7
Criterion
Part of Ilkhanate in 1300
Invaded during 1236-1242
Part of Ilkhanate in 1300
According to some estimates, half of Hungary's population perished during these years.
xxxii
Austria
Azerbaijan
Belarus
Bosnia & Herzegovia
Bulgaria
Croatia
Czech Republic
Estonia
Georgia
Hungary
Iran
Iraq
Kuwait
Latvia
Lebanon
Lithuania
Moldova
Pakistan
Poland
Romania
Serbia
Slovak Republic
Slovenia
Syria
Macedonia
Turkey
Ukraine
Invaded during 1236-1242
Part of Ilkhanate in 1300
Part of Kipchak in 1300
Invaded during 1236-1242
Invaded during 1236-1242
Invaded during 1236-1242
Invaded during 1236-1242
Invaded during 1236-1242
Part of Ilkhanate in 1300
Invaded during 1236-1242
Part of Ilkhanate in 1300
Part of Ilkhanate in 1300
Part of Ilkhanate in 1300
Invaded during 1236-1242
Invaded in 1260
Invaded during 1236-1242
Invaded during 1236-1242
Part of Ilkhanate in 1300
Invaded during 1236-1242
Invaded during 1236-1242
Invaded during 1236-1242
Invaded during 1236-1242
Invaded during 1236-1242
Part of Ilkhanate in 1300
Invaded during 1236-1242
Part of Ilkhanate in 1300
Part of Kipchak in 1300
Additional references:
Library of Congress (2013) "A Country Study: Mongolia" http://lcweb2.loc.gov/frd/cs/mntoc.html.
May, T. (2012) The Mongol Conquests in World History. New York: Reaction Books
Saunders, J.J. (1971) The History of the Mongol Conquests. University of Pennsylvania Press.
xxxiii