UC Riverside

Cliodynamics

Title
Toward Cliodynamics – an Analytical, Predictive Science of History

Permalink
https://escholarship.org/uc/item/82s3p5hj

Journal
Cliodynamics, 2(1)

Author
Turchin, Peter

Publication Date
2011

DOI
10.21237/C7clio21210

Copyright Information
Copyright 2011 by the author(s). All rights reserved unless otherwise
indicated. Contact the author(s) for any necessary permissions. Learn
more at https://escholarship.org/terms

Peer reviewed

eScholarship.org Powered by the California Digital Library

University of California
 Cliodynamics: the Journal of Theoretical and Mathematical History

Toward Cliodynamics – an Analytical,

Predictive Science of History
Peter Turchin
University of Connecticut

This article responds to those who think that a science of history is in

principle impossible. First, I tackle the issue of prediction and point out
that it is not limited to forecasting the future. Scientific prediction is also
(an much more usefully) employed in empirical tests of scientific
theories. Next, I switch from conceptual to empirical issues, and review
evidence for general empirical regularities. I also discuss some recent
examples of using scientific prediction in testing theories about historical
dynamics. I conclude by pointing out that we now have the right
quantitative tools and, even more important, a growing corpus of
historical data for testing theories. An analytical, predictive history, or
cliodynamics, is eminently possible.

Introduction
Philosophers have long debated whether history can be a science in the same
sense that physics and biology are sciences. At the heart of the debate are two
opposing views of history. Nineteenth century thinkers, such as Leo Tolstoy
and Carl von Clausewitz (see Gaddis, this volume), believed that historical
process was governed by some kind of general laws. Many French and English
historians of the nineteenth century viewed history as a science [42].
Twentieth century historians such as Toynbee [31] proposed grand schemes to
account for the rise, the flowering, and the decline of civilizations. A less
ambitious (but in the long run more influential) effort by McNeill [17] is
another example of an attempt to discern patterns in history.
During the second half of the twentieth century, however, the general
opinion among philosophers and historians swung against the possibility of
scientific history. For example, Karl Popper [18] argued that there is a
qualitative difference between history and natural sciences. Historical
processes are too complex and different in nature from physical or biological
processes. Most tellingly, people have free will, while atoms do not.
Among the historians, research paradigms that modeled themselves on
natural sciences were still popular in the 1960s and 1970s [43]. Perhaps the
most influential of such research programs was the French Annales school of
history. During these decades the new economic history, or cliometrics, briefly
flowered in the United States [44]. However, in the 1980s historians

Corresponding author’s e-mail: peter.turchin@uconn.edu

Citation: Turchin, Peter. 2011. Toward Cliodynamics – an Analytical, Predictive Science
of History. Cliodynamics 2: 167–186.
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

repudiated these approaches. As one reviewer of an early version of this article

wrote, “cliometrics went under by the early 1990s; our own quantitative
historian was denied tenure in approximately 1996 (by that point, no one really
cared about the subject or its application in history).” Instead history
experienced its “linguistic turn” or “cultural turn” [44, 45].
The view that history is fundamentally different from natural sciences is the
one widely held today by philosophers and the lay public alike. With the
exception of a tiny minority (several of whom are represented in this volume)
the historical profession has largely abandoned the search for general laws in
history. This philosophical stance is very apparent in the views of prominent
historians, for example, in “President’s Columns”, published by of Perspectives
on History, where presidents of the American Historical Association express
opinions on a wide variety of general topics confronting the historical
profession. During the last decade (the 1999–2008 issues of Perspectives on
History) there were at least three columns that discussed the role of general
laws or theories in history [4, 10, 22]. The following quote appears to be a fair
summary of the opinions of the three historians:

After a century of grand theory, from Marxism and Social Darwinism to

structuralism and postmodernism, most historians have abandoned the
belief in general laws. We no longer search for grand designs and
dialectics. Instead, we concentrate on the particular and sometimes
even the microscopic (microstoria, as it is known in Italy) – not
because we think we can see the universe in a grain of sand but because
we have developed an increased sensitivity to the complexities that
differentiate one society or one subculture from another. Kosovo is very
different from the rest of Yugoslavia, to say nothing of Vietnam [4].

In my opinion, historians gave up on general theory too soon. The need for
an analytical, predictive history remains acute if we wish to address such
problems plaguing humanity as failed states and endemic civil wars [36]. On
the other hand, there is no question that the bankrupt paradigms mentioned
by Darnton, from Marxism to postmodernism, deserve to be abandoned.
However, we now have better theories and approaches, which have profited
from recent developments in nonlinear dynamics and complexity science.
It is possible that this new batch of theories will eventually end up on the
same trash heap of history as Marxism and Social Darwinism. But I don’t think
so. My argument has two parts. First, I respond to those who think that a
science of history is in principle impossible and discuss a broader notion of
prediction that is not limited to forecasting the future. Next, I switch from
conceptual to empirical issues, and review evidence for general empirical
regularities. I also discuss some recent examples of employing scientific
prediction in testing theories about historical dynamics. In the Conclusion I
168
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

point out that we now have the right quantitative tools and, even more
important, a growing corpus of historical data for testing theories. An
analytical, predictive history, or cliodynamics, as I propose we call it, is
eminently possible.

Is there a qualitative difference between history and

natural sciences?
The issue of prediction
As mentioned above, one of the most influential arguments against scientific
history was formulated by the philosopher Karl Popper. Popper’s main point
was that because the future course of human history is critically affected by the
development of knowledge, and because future scientific and technological
discoveries cannot be predicted, a predictive science of human history is in
principle impossible.
There are additional reasons for why accurate forecasts about the future are
difficult, or even impossible with real-life social systems. These reasons include
such phenomena as the self-defeating prophecy and mathematical chaos (the
latter of which was not yet appreciated when Popper wrote The Poverty of
Historicism). However, the notion of prediction in science is not limited to
forecasting the future. If it were, whole swaths of science would lose their
status as scientific disciplines. The paradigmatic example is the weather, which
cannot be forecast more than 7–10 days in the future, even though we perfectly
well understand the laws of hydrodynamics underlying weather fluctuations.
However, because the dynamical system governing weather is in a chaotic
regime and our measurements of initial conditions are not infinitely accurate,
long-term prediction of weather is impossible.
In fact, the future is in principle unpredictable. A high-school
demonstration of the motion of uniformly accelerated objects, using an
inclined plane, may go awry because an earthquake occurs during the
experiment. The chance of such an event is rather small, but it is not zero. In
social life rare events with huge consequences, the “Black Swans” of Nassim
Nicholas Taleb [28], occur with greater frequency than in purely physical
applications. The difference, however, is quantitative, not qualitative. Bridges
collapse, space shuttles explode, and hurricanes strike from seemingly blue
skies. However, we do not decide, on the basis of such prediction failures, that
there are no laws of physics.
Prediction is an inherent part of science, but not in the narrow sense of
forecasting the future. Scientific prediction (to distinguish it from the common
usage, which is closer in meaning to “prophecy”) is used in empirical tests of
scientific theories. Scientific prediction inverses the logic of forecasting:
whereas in making forecasts we assume the validity of the underlying theory
and want to know what will happen to observables, in a scientific prediction
169
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

exercise we want to use the degree of match between observables and

predictions to infer the validity of the theory (Turchin 2006b). Because no
theory makes perfect predictions, we typically want to compare the match
between predictions and data for two (or more) theories.
Scientific predictions may be, but do not have to be, about the future. In
many historical sciences, such as geology and evolutionary biology, making
predictions about the future is impractical. Strong predictions should address
“out-of-sample” data, that is, data that had not been used to develop the theory
that is tested. Thus, it is a perfectly valid exercise to make retrospective
predictions, or “retrodictions” [13]. Historical experiments (by an experiment I
mean a planned comparison between predictions derived from two or more
theories and data) may focus on making predictions about the state of a certain
variable for a certain past society, which is not known at the time when the
predictions are made. For example, Theory #1 says that the variable should be
decreasing, while Theory #2 says, no, it should be increasing. We then ask
historians to dig through the archives (or, perhaps, archaeologists to literally
dig up the data), and determine which of the theories is closer to the truth. As
more such experiments are conducted, and if one of the theories consistently
yields predictions that are in better agreement with empirical patterns than the
other(s), our degree of belief into the better performing theory is consequently
enhanced. I will discuss in later section some examples of such experiments in
historical applications.

History and biology

Karl Popper held strong views about what constituted science. In addition to
history, he also criticized evolutionary biology, which, in his view was not a real
science, but at best “a metaphysical research program.” Ultimately, his
rejection of history and evolutionary biology was not due to logic or empirical
evidence, but to ideology [2]. His personal experiences (he emigrated from his
native Vienna in 1937 just in time to escape the Anschluss) made him into a
life-ling opponent of totalitarian ideologies, such as Nazism and Marxism. The
real targets of Popper critique were Historical Materialism and Social
Darwinism, but somehow he ended up condemning whole fields of scientific
enquiry.
Evolutionary theory, contrary to Popper, is today an established scientific
discipline, and, in my opinion, the same will eventually happen to history.
Actually, there are some interesting parallels between the state of history now
and the state of biology in the nineteenth century, before the scientific
triumphs of Charles Darwin and Louis Pasteur. The reigning theory in biology
at that time was vitalism, a doctrine that the processes of life were not
explicable by the laws of physics and chemistry alone. It was believed that
biological entities contained a “vital spark” or “élan vital,” which could not be
studied with the methods of physics and chemistry.
170
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

Vitalism is now thoroughly discredited, but this does not mean that it was a
silly theory for its times. Early scientists noted that substances seemingly fell in
two general classes. An inorganic substance, such as a lump of gold, could be
heated to the point where it changed its state (melted), but on cooling it
returned to its original form. Organic substances, when heated, changed
irrevocably. The process of heating seemingly expelled the vital force from
such substances. The destructive effect of heat on the vital force was the reason
why Pasteur had to design the famous “col de cygne” (swan neck) bottle to
disprove the theory of spontaneous generation – his first experiments were
criticized on the grounds that by boiling broth in closed bottles he destroyed
the vital force needed for spontaneous generation of life.
Ultimately vitalism was discredited not because of critical experiments,
such as that of Pasteur, but as a result of hard, and often mundane, work by
myriads of biologists who consistently applied the scientific method to
biological questions and eventually found that there was no need of a vital
force to explain general regularities in their data. In the process biology
transformed itself from the descriptive discipline that it was in the nineteenth
century (just like history is today) to an analytical, explanatory, and predictive
science of the twentieth century. Are there lessons for those of us who would
like to achieve a similar transformation of history?

History and mathematics

One of the most important lessons is recognizing the key role of mathematics
in the transition of biology from the descriptive to explanatory science (see also
the article by Geoffrey West in this volume). It was mathematical reasoning
that almost discredited Darwin’s theory of evolution in the late nineteenth
century. The dominant theory of inheritance in Darwin’s time assumed that
the offspring’s traits were a blend of its parents’ traits. Such blending
inheritance destroyed genetic variation that was absolutely necessary for
natural selection to work on. No genetic variation meant no evolution. When
biologists discovered that the theory of blending inheritance was wrong, it was
again mathematical modelers who established the firm logical foundation for
the Neo-Darwinist Modern Synthesis during the 1930s.
One of the most striking examples of the value of mathematical models
comes from the field of population dynamics. In 1924 Charles Elton published
a paper entitled Periodic fluctuations in the number of animals: their causes
and effects. After reviewing the population data on lemmings, hares, and mice,
and considering various hypotheses that might account for periodic changes in
their numbers, Elton concluded that these fluctuations must be due to climatic
variations. What is remarkable is that Elton never considered the cause that we
now know is one of the most common drivers of population cycles – the
population interaction between predators and prey [32]. The reason is that it

171
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

never occurred to him. In Modeling Nature the historian of science Sharon

Kingsland [12] relates how two years later Julian Huxley walked into Elton’s
office and showed him an article by the Italian mathematician Vito Volterra
that was just published in Nature. The article presented a simple mathematical
model of predator-prey interaction, and showed that the outcome is
population cycles of both species. Huxley, one of the founders of modern
evolutionary biology, and Elton, often considered as the father of animal
ecology, were very intelligent people. But it took a paper written by a
mathematician, who knew nothing about real animals, to open their eyes to the
possibility of predator-prey cycles.
A common objection to employing mathematical models in the study of
historical dynamics is that social systems are so complex that any
mathematical model would be a hopeless oversimplification without any
chance of telling us interesting things about these systems. This argument gets
it exactly wrong – it is because social systems are so complex that we need
mathematical models. “Naked” human brain is not a bad tool for extrapolating
linear trends, but it fails abysmally when confronted with systems of multiple
parts interconnected with nonlinear feedback loops. This is probably why it
took a mathematical model to point out that cycles are inherent in the
interaction between predators and prey (and this is a very simple system, with
just two interacting components). We need mathematical formalism to express
our ideas unambiguously, and both analytical methods and fast computers to
determine the implications of the assumptions we made.

Complexity: social and biological

It is undeniable that social systems are very complex, and have little
resemblance to such paradigmatic success stories in physics as Newton’s
planetary motions. However, many objects in natural sciences are no less
complex than human societies. Consider, for example, a temperate forest
ecosystem. There is likely to be at least a dozen species of trees and shrubs and
a hundred or more of herbs, forbs, and other smaller plants. There will be
innumerable species of insects, mites, lower invertebrates, fungi, protozoa, and
bacteria. All this life will be busy doing its thing around you; mice will scurry
underfoot and birds will be singing in the branches. It is a horrible mess (or
glorious complexity, depending on your point of view). How could it possibly
give rise to any laws of nature? Yet it does.
Over the last century ecologists identified many kinds of empirical
regularities in forest ecosystems. To continue with population cycles, almost
every forest, especially those in boreal and temperate climatic zones, has a
particularly voracious species of insect that periodically runs amok denuding
trees of their foliage, or even killing them outright. These population cycles can
be quite predictable. For example, the populations of the larch budmoth reach

172
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

a peak in the larch forests of the Swiss Alps every 8.5 years [32]. The amplitude
of these oscillations is remarkable – the population density in the trough is five
orders of magnitude (100,000 times) lower than at the peak.
Somehow large-amplitude regular oscillations arise from the mess of
nature in ecosystems. Why should the social systems be different? After all a
social system consists of only one species. Of course people are not all the same
– there are different social classes and professions, different religions and
ethnic identities, and so on. Still, when we add together the different kinds of
humans in an average historical social system (an agrarian state, for example),
I doubt that the total would come anywhere near the number of species in an
average ecosystem.

Empirical regularities
In the Second Afterword to War and Peace, Leo Tolstoy argued that in order to
find laws of history, we should focus on large masses of people and not on
individuals, no matter how important they seem (his example was Napoleon
Bonaparte). If from the microchaos of molecular motions arise the laws of
thermodynamics, and from interactions between individual lynxes and hares
arise regular predator-prey cycles, then perhaps there may be general
regularities characterizing the dynamics of human societies, even though the
behavior of each person is unpredictable. In fact, we have already found a
number of empirical regularities in historical social systems. Moreover, certain
progress has been made in identifying general principles that may underlie
these regularities. At least, it is now possible to point toward successful
examples of scientific prediction in historical dynamics. I review three such
“success stories” in this section.

A Striking Macrohistorical Pattern: Huge Empires Tend to Rise on

Steppe Frontiers
What were the social mechanisms that held together huge historical empires?
At present, we do not have a satisfactory theory accounting for the rise of such
macrostates, with territories extending across millions of squared kilometers
and populations numbering millions (or even tens and hundreds of millions).
However, there are certain empirical regularities in the spatial and temporal
distribution of “imperiogenesis” hinting that there may be general principles at
play.
In a recent publication [38], I collected as many instances, as I could find,
of historical “mega-empires” (defined as territorial states that controlled at the
peak an area greater than one million square kilometers). I found 65 such
polities for the agrarian period of human history (that is, before 1800). Over
90 percent of these empires were situated in, or next to the arid belt that runs
through Afroeurasia, from the Sahara in the West to the Gobi in the East
173
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

(Turchin [34]: figure 1). The exceptions included the only empire in the
Americas (Inca), one empire in Southeast Asia (Khmer), and three in Europe
(the Roman and Carolingian empires, and perhaps Lithuania-Poland, although
the latter expanded during the fourteenth century into steppe lands). Thus,
there is a strong statistical association between proximity to steppe and the
rise of megaempires.

1. Between the Shang era and the present, China has been unified
fourteen times (some unifications were partial). All but one of these
unifications (the Ming) originated in the North: eight from the
Northwest, and three each from the North Central and the Northeast.
In other words, with one exception all great unifying dynasties arose in
the area right on the Inner Asian frontier of China. The other side of
the frontier saw a succession of gigantic imperial confederations of
such nomadic peoples as the Xiongnu, the Turks, and the Mongols.
2. Ancient Egypt was unified by native dynasties on four occasions: Early
Dynastic (c.3100 BCE), Old Kingdom (2700 BCE), Middle Kingdom
(2040 BCE), and New Kingdom (1570 BCE). In all four cases, unifying
dynasties arose in Southern Egypt (in Hierakonpolis or Thebes).
Furthermore, 5,000 years ago Southern Egypt was surrounded not by
a lifeless desert, but by a grassy steppe inhabited by such pastoralist
peoples as Nubians and Medjay. Towards the end of the first
millennium BCE the steppe turned into desert, and from that point on
Egypt never gave a rise to a native unifying dynasty, instead being
ruled by a succession of foreign invaders. As in East Asia, the southern
frontier of Egypt saw a succession of “mirror empires.” Starting with
the Old Kingdom, and continuing even after Egypt lost its
independence, Nubia was repeatedly unified under the empires of
Kerma, Napata, Meroë, Nobadia, Makuria (Dongola), and Funj.
3. The Eurasian arid zone intrudes into South Asia from the northwest.
Out of nine South Asian unifications (most partial, as they did not
include India’s far south), five originated in the Northwest, three in the
North, and one in the West. Despite the formation of numerous
medium- and small-size states in other regions, no megaempires
originated in the Northeast, Central, or Southern India.

In summary, in all these world regions (as well as others, such as Eastern
Europe) empires originated on a steppe frontier, and only afterwards
expanded into the agrarian hinterland. Thus, steppe frontiers appear to be very
special places for imperiogenesis, places where very large territorial states are
much more likely to arise than elsewhere. The pattern of association between
steppe frontiers and mega-empire occurrence becomes particularly striking in

174
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

regions that had a steppe frontier on only one side, as in the three cases listed
above (and unlike Mesopotamia and Iran, which experienced steppe influences
from multiple directions). The connection between steppe frontiers and mega-
empires is not deterministic (because there are exceptions), but the statistical
correlation is very strong.
Strong macrohistorical regularities suggest that the rise of any particular
mega-empire was not a random result of a concatenation of unique events;
general social mechanisms must have been at work. Building on the ideas of
the fourteenth century thinker Ibn Khaldun [11], as well as contemporary
anthropologists [1, 15], I have proposed a “mirror-empire” model as one
common route to mega-empire [38]. This model postulates that antagonistic
interactions between nomadic pastoralists and settled agriculturalists result in
an autocatalytic process, which pressures both nomadic and farming polities to
scale up polity size, and thus military power. In many cases, as happened
repeatedly in China and Ancient Egypt, the end result of this process is the
simultaneous rise of an agrarian empire and a nomadic imperial confederation
on their respective sides of the steppe frontier. However, if the agrarian state
does not have a deep hinterland to expand into, it may lose the scaling-up race
to the nomadic polity, and is conquered by it. This was the typical dynamic in
the Maghreb, so admirably described by Ibn Khaldun.

Secular cycles: linked oscillations in demographic, social, and

political structures of agrarian societies
The pattern of population change is strongly affected by the scale at which it is
observed. On a very long time scale of millennia population numbers increase
at an accelerating rate, while on the time scale of years, several bad harvests in
a row can cause a temporary dip in numbers, which is made up as soon as
weather gets better. At the intermediate scale of decades and centuries the
dominant pattern appears to be secular cycles: roughly century-long periods of
sustained population growth followed by a similarly long period of population
decline and stagnation [37]. For example, in Western Europe the thirteenth
century was a period of vigorous population growth, while during the
fourteenth and the first half of the fifteenth centuries population declined. The
sixteenth century was another period of rapid growth, followed by the decline
and stagnation of the seventeenth.
One possible explanation of this pattern of long-term population
oscillations is offered by the demographic-structural theory [8, 31, 39]. First,
population growth beyond the means of subsistence leads to declining levels of
consumption and popular discontent. Second, and more important, sustained
population growth also results in increasing numbers of aspirants for elite
positions, leading to intra-elite rivalry and factionalism. A third consequence is
persistent inflation, which causes a decline in real state revenues and a

175
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

developing fiscal crisis of the state. As these trends intensify, the result is state
bankruptcy and loss of military control; spiraling conflict among elite factions;
and a combination of elite-mobilized and popular uprisings that lead to
breakdown of central authority. In turn, political instability (urban riots,
peasant uprisings, and full-scale civil war) results in population decline.
Eventually, the balance between population numbers and the means of
subsistence is restored, and another cycle can begin.
Various assumptions about dynamical feedbacks between key
demographic-structural variables, such as population growth, elite
overproduction, state strength, and political instability, have been investigated
with formal mathematical models (Turchin [31], ch. 7; 35]). A typical
dynamical pattern of association between population growth and political
instability, predicted by these models, is coupled oscillations of population
dynamics and political instability. Both variables cycle with the same period,
but are shifted in phase with respect to each other, so that instability peaks
during the periods of population decline (figure 1).

4 1.0
population
instability

0.8
3

Sociopolitical Instability
0.6
Population

0.4

1
0.2

0 0.0
500 600 700 800 900 1000

Time
Figure 1. Linked population-instability oscillations predicted by a
demographic-structural model (Turchin and Korotayev [35], eq. 8).

176
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

In real life, we do not expect to see smooth, perfectly periodic cycles.

Historical societies are characterized by a much richer, more complex web of
dynamical feedbacks than can be portrayed in mathematical models.
Multiplicity of nonlinear feedbacks increases the probability that the dynamics
will be chaotic, resulting in irregular, noisy-looking oscillations. Furthermore,
social systems are affected by exogenous variables, such as climate
fluctuations, that also generate erratic dynamics. Events at the microlevel,
including acts of individual people, may percolate up and have macro-level
consequences. Finally, and most importantly, our data on historical dynamics
is often sparse and suffers from large amounts of observation noise. These
complications must be taken into account when we look at real data. Yet,
despite all these problems, we observe the basic dynamical pattern predicted
by theory: linked oscillations with peaks of sociopolitical instability lagging
behind population peaks (figure 2). This observation suggests that the
demographic-structural model, indeed, captures an important aspect of the
functioning of historical societies.
How can we design a general and quantitative test that goes beyond an eye-
ball comparison of predictions (figure 1) to the observed patterns (e.g., figure
2)? Given the limitations of historical data and the complexity of the dynamical
pattern (variability in oscillation periods and phase shifts), we need to employ
an appropriately coarse-grained procedure (see also the articles by Murray
Gell-Mann and Geoffrey West in this volume). One possible approach works as
follows. First, we identify the population growth and decline phases. Although
quantitative details of population dynamics for historic societies are rarely
known with any precision, there is usually a consensus among demographic
historians about when the qualitative pattern of growth changed. Second, we
count instability events (peasant uprisings, separatist rebellions, civil wars,
etc) that occurred during each phase. Finally, we compare the incidence of
instability events per decade between the two phases. Theory predicts that we
should have much greater instability during population decline versus growth
phases.
First, we apply this procedure to secular cycles in China (table 1). The test is
conducted only for periods when China was unified under one dynasty. The
empirical regularity is very strong: in all cases instability is greater during the
declining, compared to growth phases (t-test: P << 0.001).
Next, we apply the approach to all seven complete cycles examined in
Turchin and Nefedov [39] (table 2). The instability data were taken from such
compilations as that of Sorokin [23], Tilly [29], and Stearns [27]. Again, the
empirical regularity is strong and statistically highly significant (in all cases
instability is greater during the declining, compared to growth phases; t-test: P
<< 0.001).

177
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

population
60 instability 30

Sociopolitical Instability
Population

40 20

20 10

0 0
-200 -100 0 100 200 300

Figure 2. Population dynamics and sociopolitical instability in China from the

Qin unification to the period of Three Kingdoms (220 BCE – 300 CE). For data
sources, see Turchin ([31], p.164)

Table 1. Secular Cycles in Europe and China During the Last Millennium
Compared to Global Economy Processes
European cycles Chinese cycles Global Economy Processes
Ottonian-Salian Northern Song Sung* Breakthrough
920–1150 960–1127 930–1190
Capetian Mongol-Yuan Nautical/Commercial Revolutions
1150–1450 1200–1368 1190–1430
Valois Ming Oceanic Trading System
1450–1660 1368–1644 1430–1640
Bourbon Qing Industrial Takeoff
1660–1870 1644–1911 (1640–1850)
*A variant spelling of Song

178
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

Table 2. Instability events per decade during the growth and decline phases
of the secular cycles surveyed in Turchin and Nefedov [39].
Growth phase Decline phase
Secular Cycle years Instability years Instability
Plantagenet 1151–1315 0.78 1316–1485 2.53
Tudor 1486–1640 0.47 1641–1730 2.44
Capetian 1216–1315 0.80 1316–1450 3.26
Valois 1451–1570 0.75 1571–1660 6.67
Republican 350–130 BCE 0.41 130–30 BCE 4.40
Principate 30 BCE–165 0.61 165–285 3.83
Muscovite 1465–1565 0.60 1565–1615 3.80
Average (±SE) 0.6 (±0.06) 3.8 (±0.5)

In summary, the dynamical pattern predicted by the demographic-

structural model is apparent in data ranging across all Eurasia and from the
third century BCE to the nineteenth century CE. Furthermore, the same
regularity is observed in Egypt from the Hellenistic through the Ottoman
periods [14]. In fact, it appears that this empirical pattern holds for all agrarian
societies whose dynamics are not unduly influenced by exogenous forces, e.g.,
large empires (such as the Roman and Chinese ones) or island states (England
and Japan).

The dynamics of religious conversion

The last example concerns testing dynamical theories about religious
conversion (Turchin [31], ch. 6). Three more-or-less explicit models for
religious conversion and ethnic assimilation have been proposed in the
literature: the noninteractive, the autocatalytic, and the threshold models. The
justification for each of the model does not concern us here (the details are in
Turchin [31], section 6:2:1); what is important is that each model predicts a
qualitatively different trajectory (the proportion converted/assimilated as a
function of time). This means that we can determine which theory better
reflects the reality if we can find data on the temporal course of conversion.
Empirical data on conversion to Islam in Iran and Spain, all strongly
supported the autocatalytic model and were nothing like trajectories predicted
by the two alternatives (figure 3a,b). What do we conclude from this result? All
models are by definition wrong, because they oversimplify the complex reality,
but the autocatalytic model is less wrong than the alternatives. It appears that
the assumptions of the conversion process built into the autocatalytic model
capture some important aspect of the reality.

179
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

(a) Conversion to Islam: Iran (b) Conversion to Islam: Spain

1 1

data
fitted
Proportion converted

Proportion converted
model

0 0
650 700 750 800 850 900 950 1000 1050 650 700 750 800 850 900 950 1000 1050 1100 1150

(c) Christians in the Roman Empire (d) The Growth of the LDS Church
1 15
fitting data (Egypt) fitting data
prediction based model
LDS Membership, mln
Proportion Christian

on Egypt data testting data

testing data
10
(office holders)

0 0
200 250 300 350 400 450 500 1840 1860 1880 1900 1920 1940 1960 1980 2000

Figure 3. The dynamics of religious conversion. Trajectories of conversion to

Islam in (a) Iran and (b) Spain. The curve is the fitted autocatalytic model (the
logistic equation). Out-of-sample predictions: (c) Christians in the Roman
Empire and (d) the growth of the Mormon (Latter-Day Saints) Church.

Note that the three conversion models that I considered were not flexible
statistical models, such as splines or neural nets. They were based on specific
assumptions about mechanisms underlying conversion, and predicted
qualitatively different shapes of trajectories. Thus, the comparison between
theoretically predicted shapes and the empirically observed ones was definitely
a step forward, because it roundly rejected two of the models in favor of one.
Nevertheless, each model had tunable parameters, and it would strengthen the
result if one model were capable of successfully predicting out-of-sample data
(“in-sample” refers to data used in model fitting, “out-of-sample” data are
those that were not used in fitting but were reserved for testing the model; or
perhaps were collected after the model was fitted).
180
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

There was an element of out-of-sample prediction in the test a third data

set, involving conversion to Christianity (Turchin [31], section 6.3.2). This case
study was based on data in the book by Rodney Stark [24] on the rise of
Christianity (see also Hopkins [9], Stark [25]). Stark used a variant of the
autocatalytic model to predict how the number of Christians in the Roman
Empire grew from the first century on. He estimated (guessed, really) that
there were roughly a thousand converts in 40 CE and that their numbers grew
at the rate of 40% per decade. Several years after he made these estimates, a
colleague attracted his attention to the reconstruction by Roger Bagnall of the
growth of Christianity in Egypt, based on data in Egyptian papyri. Since Stark
was unaware of Bagnall’s data at the time when he constructed his prediction,
we have here a true test with out-of-sample data.
This story has a sequel. Two years after I wrote the chapter on conversion in
Historical Dynamics [31], I happened on a reference to a German dissertation
that gave a list of Pagan and Christian office-holders between 324 and 455
[40]. I immediately realized that these data enable us to make another test of
the autocatalytic model [34]. The results are shown in Figure 3b. We see that
the curve fitted to the Bagnall data (showing the proportions converted before
300 CE, filled circles) does a very good job predicting the course of
Christianization in the von Haehling data (after 330 CE, hollow circles). The
coefficient of prediction (the proportion of variance of out-of-sample data
predicted by the model) is a healthy 0.57.
A similar, although less dramatic, exercise can be performed with the data
on the growth of the Mormon (Latter-Day Saints) Church [26]. The model
fitted to the data up to the outbreak of World War II does a very good job
predicting post-War trajectory (figure 3d).
Taken together, the results in Figure 3 tell a remarkable story. They suggest
that once world religions got going, they generated a kind of momentum that
allowed them to expand at approximately constant (per capita) rate. Dramatic
events – world wars, imperial collapses, and nomadic invasions – did not
derail these massive macrohistorical processes, at least in these particular
cases (of course, certain kinds of events, such as the Christian Reconquista in
Spain, are capable of reversing the tide of religious conversion).

Conclusion
The empirical studies surveyed above are each based on a powerful
macrohistorical regularity cross-cutting across world regions and historical
periods. Although Kosovo and Vietnam (to use Robert Darnton’s example)
differ in many ways, at some deeper level their economic and political
dynamics may be driven by similar mechanisms. Certainly, Ancient Rome,
Imperial China, Capetian France, and Romanov Russia are as different from

181
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

each other as Kosovo from Vietnam. Yet these states all arose on metaethnic
frontiers and experienced a sequence of secular cycles [33].
History is not “just one damn thing after another.” Strong empirical
patterns arise because the dynamics of historical societies reflect the action of
general social mechanisms. There are laws of history (in the broad sense of the
word). Furthermore, successful case studies of scientific prediction, reviewed
in this article, show that we are well on the way to identifying some of these
laws.
As I noted in the Introduction, attempts to transform history into an
analytical, mathematized science have been made before, but were largely
unsuccessful. One of the most ambitious efforts is that of Nicholas Rashevsky
[21]; a book that, unfortunately, has been largely ignored. How is the situation
different today?
Two recent developments, one theoretical and another empirical, have
dramatically changed the scientific landscape.
First, the advances in nonlinear dynamics and complexity science have
revolutionized how we do theory in science, even (especially) in such difficult
fields as history. Our theoretical approaches to complex systems are no longer
limited to verbal theories. Dynamical models, such as systems of differential
equations, allow us to handle precisely and quantitatively such issues as the
importance of contingency and dependence on initial conditions. Such hoary
issues as “chance versus necessity” can now be addressed quantitatively by
models combining deterministic and stochastic terms (Turchin [31], pp.6,14).
Agent-based computer models (Epstein and Axtell 1996) is another key tool for
investigating the effects of stochasticity and the influence of individuals on the
historical process. This approach is also custom-made for investigating how
macro-level patterns arise from micro-level interactions.
Second, the recent years saw a qualitative increase in the amount of data
available for testing theories about historical dynamics. The key development
has been the spread of computer use among the historians and the rise of the
Web. As a result, more and more datasets are now easily accessible through the
Internet. To illustrate the potential consequences of this shift, consider that
invaluable tool of a macrohistorian, the historical atlas. Traditional book
atlases are inherently limited. A typical problem is that either the region or the
period, in which one is interested, is not in the list of maps collected in the
atlas. Furthermore, traditional atlases are ill-equipped to portray dynamical
change. What we need is a computer-based dynamical atlas that allows one to
zoom in on arbitrary geographic regions and play movies to gain an
understanding of temporal changes occurring there. Such a perfect atlas does
not exist yet, but I know of several initiatives to create one. It is a matter of
years, not decades, before we have one.
I argue that we already have the necessary analytical tools for modeling
historical processes and statistically analyzing data. Naturally we need more
182
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

data, but it is clear how to increase our “empirical capital” – all it takes is more
hard work. The greatest challenge that I see is a conceptual one: how do we
construct meaningful theory? How do we define and theorize the key variables
on which our dynamical models will be based? Some variables are conceptually
easy. To study the demographic and economic aspects of historical societies all
we need to do is to bring in the concepts already worked out by demographers
and economists. That is clearly why demographic history and cliometrics [6, 7]
were the first fields of history where the scientific method was systematically
applied.
Other variables are much more difficult to wrap one’s mind around. As an
example, take social cohesion, or the capacity of a group for collective action,
for which, I have proposed [31, 33], we could use Ibn Khaldun’s term asabiya.
It is clear that the Romans of the third century BCE, during the Punic Wars,
possessed much greater asabiya than the Italians of the fifth century CE, when
the Roman Empire in the West was in the process of disintegration. But how
do we define and measure this change? (One thing is certain, if we can figure
out how to measure asabiya, its units will be called khalduns.) It seems to be a
nebulous, hard-to-pin-down quality. Yet recently there has been some progress
in measuring it. I am thinking of the concept of social capital as proposed and
used by such political scientists as Robert Putnam (and not to be confused with
the social capital of Bourdieu [3], Putnam et al., [19], Putnam [20]). As I have
argued earlier, social capital is none other than asabiya for modern societies
(Turchin [31], p.43). Putnam and coworkers proposed a variety of approaches
to measuring relative amounts of social capital among different Italian
provinces [19], as well as changes over time in the United States [20]. Thus,
although at first a concept may appear nebulous, hard work involving theory
development and empirical testing may, in the end, lead to precise definitions
and ways to obtain quantitative measures. It is important to remember that
physics, which appears to us now as a hard science, or biology, had to travel
the same route. Such difficult concepts as, for example, entropy, were not
obvious right away, and arose as a result of lengthy collective labor by many
scientists.
In this essay I have looked back at the history of natural sciences and
argued that, although at present the obstacles to developing a scientific history
appear to be formidable, we forget that natural sciences overcame similar
challenges during their infancy periods. I am convinced that historical
scientists will also solve the problems of how to conceptualize and measure key
theoretical variables in cliodynamics, how to build meaningful theory and then
test it empirically. It will take time and a lot of work. But what is encouraging
is that, as the empirical “success stories” show, we are already well on the way
toward a science of analytical, predictive history.

183
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

References
[1] Barfield, T. J. The Perilous Frontier: Nomadic Empires and China.
Oxford, UK: Blackwell, 1989.
[2] Beatty, J. "Historical Determinism, Evolution and Totalitarianism in the
Works of Hannah Arendt and Karl Popper." In Thinking about
Evolution: Historical, Philosophical and Political Perspectives, edited by
R. Singh, C. Krimbas, D. Paul, and J. Beatty, 62-67. A Festschrift for
Richard C. Lewontin. Cambridge, MA: Cambridge University Press,
2001.
[3] Bourdieu, P. "Le capital social: notes provisoires." Actes de la
Recherches en Sciences Sociales 3 (1980): 2-3.
[4] Darnton, R. "President's Column: History Lessons." Perspectives on
History 37(6) (1999): 1-3.
[5] Epstein, J. M., and R. Axtell. Growing Artificial Societies: Social Science
from the Bottom Up. Washington D C.: Brookings Institution Press,
1996.
[6] Fogel, R. W., and G. R. Elton. Which Road to the Past? Two Views of
History. New Haven: Yale University Press, 1983.
[7] Fogel, R. W. "Economic Growth, Population Theory, and Physiology:
The Bearing of Long-Term Processes on the Making of Economic
Policy." American Economic Review 84 (1994): 369-395.
[8] Goldstone, J. A. "Revolution and Rebellion in the Early Modern World."
Berkeley, CA: University of California Press, 1991.
[9] Hopkins, K. "Christian Number and Its Implications." Journal of Early
Christian Studies 6 (1998): 185-226.
[10] Hunt, L. "President's Column: Where Have All the Theories Gone?"
Perspectives on History 40(3) (2002): 1-3.
[11] Ibn Khaldun. The Muqaddimah: An Introduction to History. Translated
from the Arabic by Franz Rosenthal. New York, Pantheon Books, 1958.
[12] Kingsland, S. Modeling Nature: Episodes in the History of Population
Ecology, 2nd edition. Chicago, IL: University of Chicago Press, 1995.
[13] Kiser, E., and M. Hechter. "The Role of General Theory is Comparative-
Historical Sociology." American Journal of Sociology 97 (1991): 1-30.
[14] Korotayev, A., and D. Khaltourina. Introduction to Social
Macrodynamics: Secular Cycles and Millennial Trends in Africa. URSS,
Moscow, 2006.
[15] Kradin, N. N. "Nomads, World-Empires, and Social Evolution." In
Alternative routes to civilization (in Russian), edited by N. N. Kradin, A.
V. Korotayev, D. M. Bondarenko, and V. A. Lynshi, 314-336. Logos,
Moscow, 2000.
[16] Lee, J. S. "The Periodic Recurrence of Intrnecine Wars in China." The
China Journal March-April (1931): 111-163.

184
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

[17] McNeill, W. H. The Rise of the West. New York: New American Library,
1963.
[18] Popper, K. R. The Poverty of Historicism. London: Routledge and Kegan
Paul, 1957.
[19] Putnam, R. D., R. Leonardi, and R. Y. Nanetti. Making Democracy Work:
Civic Traditions in Modern Italy. Princeton, NJ: Princeton University
Press, 1993.
[20] Putnam, R. D. Bowling Alone: The Collapse and Revival of American
Community. New York: Simon and Schuster, 2000.
[21] Rashevsky, N. Looking at History through Mathematics. Cambridge,
MA: MIT Press, 1968.
[22] Sheehan, J. J. "President's Column: How Do We Learn from History?"
Perspectives on History 43(1) (2005): 1-3.
[23] Sorokin, P. A. Social and Cultural Dynamics. Vol. III. Fluctuations of
Social Relationships, War, and Revolution. New York: American Book
Company, 1937.
[24] Stark, R. The Rise of Christianity: A Sociologist Reconsiders History.
Princeton, NJ: Princeton University Press, 1996.
[25] Stark, R. "E contrario." Journal of Early Christian Studies 6 (1998): 259-
267.
[26] Stark, R. The Rise of the Mormonism. New York: Columbia University
Press, 2005.
[27] Stearns, P. N. The Encyclopedia of World History, 6th edition. Boston,
MA: Houghton Mifflin, 2001.
[28] Taleb, N. N. The Black Swan: The Impact of the Highly Improbable. New
York: Random House, 2007.
[29] Tilly, C. European Revolutions: 1492-1992. Oxford, UK: Blackwell, 1993.
[30] Toynbee, A. J. A Study of History. London: Oxford University Press,
1956.
[31] Turchin, P. Historical Dynamics: Why States Rise and Fall. Princeton,
NJ: Princeton University Press, 2003a.
[32] Turchin, P. Complex Population Dynamics: A Theoretical/Empirical
Synthesis. Princeton, NJ: Princeton University Press, 2003b.
[33] Turchin, P. War and Peace and War: The Life Cycles of Imperial Nations.
New York: Pi Press, 2006a.
[34] Turchin, P. "Scientific Prediction in Historical Sociology: Ibn Khaldun
meets Al Saud." In History and Mathematics: Historical Dynamics and
Development of Complex Societies, edited by P. Turchin, L. Grinin, A.
Korotayev, and V. C. de Munck. URSS, Moscow, 2006b.
[35] Turchin, P., and A. Korotayev. "Population Dynamics and Internal
Warfare: A Reconsideration." Social Science and History 5(2) (2006):
121-158.
[36] Turchin, P. "Arise 'cliodynamics'." Nature 454 (2008): 34-35.
185
 Turchin: Toward Cliodynamics. Cliodynamics (2011) Vol. 2, Iss. 1

[37] Turchin, P. 2009a. "Long-Term Population Cycles in Human Societies."

In The Year in Ecology and Conservation Biology, edited by R. S. Ostfeld
and W. H. Schlesinger, 1-17. 2009. Ann. N. Y. Acad. Sci. 1162.
[38] Turchin, P. "A Theory for Formation of Large Empires." Journal of
Global History 4 (2009b): 191-207.
[39] Turchin, P., and S. Nefedov. Secular Cycles. Princeton, NJ: Princeton
University Press, 2009.
[40] von Haehling, R. "Die Religionszugehörigkeit der hohen Amtsträger des
Römischen Reiches seit Constantins I. Alleinherrschaft bis zum Ende der
Theodosianischen Dynastie." Ph.D. dissertation, Bonn, 1978.
[41] Zhao, W., and S. Z. Xie. Zhongguo ren kou shi: China Population History
(in Chinese). Peking: People's Publisher, 1988.
[42] Carneiro, R. L. 1997. Human History: A Domain of Competing
Perspectives. Europaea 3:9-32.
[43] Bonnell, V. E., and L. Hunt. 1999. Introduction. Pages 1-32 in V. E.
Bonnell and L. Hunt, editors. Beyond the Cultural Turn: New Directions
in the Study of Society and Culture. University of California Press,
Berkeley.
[44] Williamson, S. H. 1991. The history of cliometrics. Research in Economic
History, Supplement 6:15-31.
[45] Hunt, L. 1989. The New Cultural History. University of California Press,
Berkeley.
[45] Sewell, W. H. 2005. Logics of History: Social Theory and Social
Transformation. University of Chicago Press, Chicago.

186