1. society

1
Stones, Bones, Cities, and States:
A New Approach to the Neolithic Revolution
By
Justin Burkett
Economics Department
Wake Forest University
Winston-Salem, NC 27106
and
Richard H. Steckel
Economics Department
Ohio State University
Columbus, OH 43210
and
John Wallis
Economics Department
University of Maryland
College Park, MD 20782
April 2015
This is a preliminary draft, please do not cite, quote, or post on a web site with the authors’s
permission.
2
Abstract
The rise of agriculture and the emergence of towns and cities marked the beginning of larger
human society. Social scientists have constructed various explanations of why this happened which
can be placed into exogenous and endogenous categories such as climate change and over-hunting of
a common property resource. We are less concerned with why the shift occurred than with
understanding how larger scale societies evolved. We are, particularly concerned with how larger
societies solved the problem of violence. The answer to the how question is potentially much more
important than the answer to the why question. The fact that large societies tend to exhibit common
organizational features might be the result of a common external stimulus, but it seems much more
plausible that external stimulus varied and institutional isomorphism resulted from the solution to a
common set of organizational problems facing all large societies. We develop a network model that
illuminates how societies plagued by the problem of violence can develop larger groups with
specific internal and external relationships that limit violence. The logic leads to several empirical
implications that we test with new evidence consistent. The evidence shows that urban living was
less healthy but also considerably less violent than found among hunter-gatherers. Drawing upon
the theory of the natural state, in which the political system manipulates economic privilege to create
social order, our explanation is consistent with evidence that new methods of social organization
accompanied the rise of agriculture and urbanization. We argue that Neolithic societies preferred
urban living built on farming despite worse health outcomes because new methods of organization
created social order, enforced property rights, and reduced violence.
3
The invention of agriculture, the wide spread shift to sedentary lifestyles, and the growth
of large population centers began between 10,000 to 5,000 years ago, depending upon location,
in what we now call the Neolithic revolution. This profound change in human activity marks the
beginning of larger human societies and has long been of interest to economists, anthropologists,
and social scientists in general who have longed wondered about why the shift occurred. We ask
a different question about the rise of large societies: how did societies solve the problem of
limiting and controlling violence in populations large enough that most people could not have
known each other personally?
Living in larger sedentary groups required new social arrangements controlling the
endemic violence characteristic of hunter-gatherer societies. As small groups morphed into
larger units and the probability of face to face contact declined between typical individuals,
larger societies had difficulty limiting the incidence of violence. In most modern societies,
ceteris paribus, increasing scale increases the incidence of violence. Building on the ideas of
North, Wallis, and Weingast (2009, hereafter NWW), we propose a conceptual model of how
societies could have moved from hunter-gatherer scales to urban scales by creating arrangements
that constrained violence and, simultaneously, created social structures capable of coordinating
larger groups and organizations. A virtue of the model is that its implications can be tested by
utilizing new anthropometric information on the incidence of human induced trauma and recent
anthropological studies of the transition between bands, tribes, chiefdoms, and states.
Another important implication of the model enables us to address one of the paradoxes of
the urban revolution: why people tolerated the health costs of moving into towns and cities.
Evidence from skeletons shows that Neolithic cities and towns were unhealthy: their residents
were smaller in stature than hunter-gatherers, and their bones had relatively more lesions
4
indicating dental decay, infections and other signs of physiological stress. Since early city
dwellers had the option of living as healthier hunter-gatherers, why did they choose to live in
cities? What benefit of larger social organizations offset this cost? Modern evidence suggests
that violent crime increases with city size.1 If this was the true for the first cities, the transition is
even more perplexing. People formed and then moved into cities with higher rates of violence
and lower material standards of living! As we show, however, larger places were associated
with lower, rather than higher, levels of violence.
People may have gathered in cities as defense against violent attack. The weight of sheer
numbers would convey a substantial military advantage and social organizations that could put
larger numbers of people in the field should have enjoyed a significant advantage. The military
benefits of numbers, however, existed before there was agriculture. The theoretical question
again boils down to scale: understanding the new ability of some human societies to sustain
larger population groups. Neolithic institutions must have created new ways of structuring
human interaction.
We present a theory of social change consistent with the conundrums in the archeological
record and capable of explaining the emergence of large social units several millennia ago. The
heart of the theory is what NWW call the “natural state,” in which the political system
manipulates economic privilege to create rents that can be used to support social order and
reduce violence. In NWW elite individuals appear as an outcome of the social logic of the
natural state. In the model developed here, more micro-founded forces generate “elite”
individuals from the interaction of a population of equally endowed individuals who have the
option to form, or not form, alliances and groups. The model begins with individuals in a violent
environment. One stable form of group is a small egalitarian one. We show the conditions
1
Glaeser and Sacerdote, 2000.
5
under which larger groups can form that limit the level of violence in the society as a whole.
Individuals who form the connections between groups are exposed to higher levels of violence,
but they lower the probability of violence for others. In a world where transfers between group
members occur, the individuals exposed to a higher probability of violence enjoy a higher
material standard of living, and in that sense they are “elites,” but in utility terms they are not
necessarily better off since they also experience a significantly higher level of violence. Nonelites may voluntarily transfer resources to the elites in return for lower probabilities of violence.
The model provides a reasoned explanation for the urbanization paradox, as well as the empirical
information on the incidence of violence by size of place. The micro-founded model produces a
social logic that closely parallels the logic of the natural state.
The model is intended to be a parsimonious way of getting at the central logic of how
larger societies limit violence. Any method of lowering violence must induce powerful
individuals to stop fighting. The anthropological evidence strongly suggests that most huntergatherer societies are aggressively egalitarian, so that potentially powerful individuals are
constrained from coercing members of their immediate group (Boehm, 1999 and Kelly, 1995).
That success comes at a cost, however. The group must remain small, and both intra- and intergroup violence is relatively high.
We address the problem of forming credible coalitions of actors using simple
assumptions about violence and credible agreements. We begin with a world of small, isolated
individuals. Individuals face an exogenous probability of conflict. They can ally with others,
which simultaneously lowers the probability of fighting with allies but increases the probability
of having to fight alongside your allies if they are attacked. We provide a logic for why small
egalitarian groups form, and we show why larger egalitarian groups are problematic. Formation
6
of larger groups require connections between smaller groups, and the individual providing those
connections must, following the logic, expose themselves to more violence
The individual based logic is then extended to larger organizations and social networks,
following the logic of the natural state developed by NWW and Wallis (2011). The formation of
a coalition of powerful actors not only creates a social hierarchy, it fundamentally changes the
dynamics of small group interaction. Leaders of smaller groups are now backed by the
combined weight of the coalition of leaders of other groups. The aggressive egalitarianism of
small group hunter-gatherer societies no longer applies as a social logic. One of the important
implications of the natural state theory is that creating a viable way for powerful individuals to
cooperate makes the distribution of violence capability much more unequal among individuals,
while at the same time lowering the overall level of violence in society. Powerful individuals
may enjoy more resources, but they also incur higher levels of violence. Our two empirical tests
rest on these two predictions: first, when larger groups form that the overall level of violence in
society will fall and, second, larger groups will develop social hierarchies that use economic
privilege as a way to cement political relationships between powerful individuals.
The archeological record should show that residents of early population centers suffered
less from violence. The third section of the paper describes how deliberate trauma can be
inferred from skeletal remains, and then we apply the methodology to a sample of several
thousand skeletons from the Western Hemisphere. The fourth section presents results clearly
showing that early urban dwellers experienced significantly lower levels of deliberate trauma
over their lifetimes in comparison to hunter gatherers. Our argument does not hinge on whether
the level of violence in hunter-gatherer societies was high or low in an absolute sense, it hinges
on whether at comparable periods of time the level of human induced violence declined with
7
population size in the era of the Neolithic revolution.
There is a rich literature on the emergence of larger societies from hunter-gatherer
societies. This preliminary version of our paper sets aside that literature to focus on the
theoretical model and the empirical evidence.
II. Violence, hierarchy, and the natural state
A Formal Model
This section introduces a formal model of small-group interaction providing a simple
framework for discussing the interplay between social structure and violence. There are k groups
of n agents (or individuals) in the population. Agents can form alliances with one another.
Alliances form between individuals, and those individuals can be in the same group or in
different groups. Alliances are promises not to fight each other, and to come to each other’s aid
in case either member of the alliance is involved in a conflict. A group is “isolated” if the
members of the group have no alliances with anyone outside the group. Whether a group’s
members have internal alliances with each other comes endogenously out of the logic of the
model.
Groups that are not isolated, that is, groups with at least one member that has an alliance
with a member of another group, tend to be hierarchically organized rather than egalitarian. This
is because the individual with the outside alliance is potentially exposed to more violence. One
of the model’s main results is that egalitarian group structures tend to be isolated and experience
more violence than their hierarchical connector counterparts. Another important feature is that
egalitarian groups and hierarchically connected groups can coexist in the same equilibrium. The
outcome results solely from choices made by the agents about who they should ally themselves
with in the presence of potential violent conflict. It is not a result of intrinsic differences
8
between the agents’ productive or coercive abilities. We begin with the explicit assumption that
all agents are symmetric in the setup of the model.
For the
groups of
agents each, suppose that
2 and
2. The different groups are
denoted using uppercase letters ( , , …), while group members are denoted using the
corresponding lowercase letters ( ,
, …). Slightly abusing the notation, we sometimes refer to
arbitrary agents as , , and with the understanding that they may or may not be in the same
group. Each agent is endowed with one unit of a divisible good that can be thought of as the
result of his group’s production activities.
The agents are all interested in maximizing their expected consumption of this good (they
are risk-neutral). The complication they each face is that they are inevitably involved in conflicts
that sometimes become violent. There is one period in which a pair of agents is randomly
selected to be in conflict with each other. Conflicts potentially occur between any two agents. In
general, a conflict arises between agents
and
with probability
,
.2 We study the case
where the probability of conflict between two agents depends only on whether the two agents are
in the same group or not. The probability of a conflict between two individuals is higher when
they are members in the same group. Conflict arises through interacting with other agents, so that
conflict is more likely between agents who interact more frequently.3 Formally,
and
are in the same group and
,
otherwise with 0
,
1. Of course,
if
and
cannot be too large.4
Conflicts are symmetric so there is no distinction between
, and
, .
This seems paradoxical on its face, since members of the same group know each other better and can secure
agreements through the value of their ongoing relationship. The logic here, however, is the same logic that Daly and
Wilson use to understand why most murders occur among family members and close friends than strangers. It is not
that you are more likely to be killed in a given interaction with a family member than a stranger, it is that you are
involved in many more interactions with family members. As we develop the model, it will become clear why,
within group alliances can form that limit the intra-group conflicts that lead to violence.
4
There are
pairs of distinct agents in the model,
of which are pairs of agents in the
2
3
9
Conflicts do not have to become violent. Whether or not a conflict becomes violent
depends on the alliances agents have with one another and the agents’ incentives to uphold those
alliances. We are primarily interested in determining which sets of alliances can be considered
stable with respect to the incentives of agents to form and destroy alliances. To answer this
question we employ tools from the literature on the economics of networks, thinking of the
bilateral alliances as links in a network with agents as the nodes.
Formally, an alliance is an agreement between two agents such that each is involved in
the other’s violent conflicts. The agents agree to not fight each other when they are allied and
agree to come to each other’s aid if a conflict arises with agents outside the alliance. An alliance
is valuable to the agents because it eliminates violence between the two allied agents, but the
alliance comes with a cost by requiring that each agent participate in the other’s conflicts. Some
other potential consequences of alliances, such as the potential benefit of having additional
agents fighting on one’s behalf, are ignored to keep the model simple.
Given a network of alliances, the likelihood of conflict arising between any two agents,
and the consequences of conflicts becoming violent, we calculate each agent’s expected payoff
from the network when alliances are upheld. We are interested in the set of alliances that have
the property that given the existing alliances no agent can benefit from breaking an alliance and
no two agents can benefit from adding an alliance between them. In the economics of networks
this concept is referred to as pairwise stability. We do not assume that alliances will always be
upheld, but we confine our focus to situations where agents can expect that alliances will be
upheld because it is in the interest of all parties to do so. Since we will also be interested in any
transfers that might occur between agents involved in an alliance, we use a modified version
same group. Therefore, we must have that
1
2
1
2
1.
10
introduced by Bloch and Jackson (2006) called pairwise stability with transfers.
To provide a formal definition, we use notation from network theory (see Jackson
(2008)). The notation
refers to an alliance (or link) between
and
, … refers to the set of alliances or a network. The expression
network resulting from adding the
represent
alliance (
,
,
refers to the
is used analogously). Let
’s expected payoff (excluding transfers) in network . Then
is pairwise stable with
transfers (PST) if
i. for all
∈ ,
ii. for all
∉ , if
; and
;
The two conditions can be understood as requiring that adding or removing a link must weakly
reduce the joint surplus available to the two agents. The unspecified transfers may redistribute
the surplus of a link between the two agents, but they also may be used, for example, to allow
two agents to add a link that would harm one of the agents without the ability to transfer payoffs.
Since we only need to consider changes in joint surplus to apply this concept, we leave a
discussion of the transfers to a later section.
To calculate these expected payoffs, it is critical to determine when a conflict becomes
violent. We only want to label a network of alliances PST (pairwise stable) if the alliances are
never broken and those involved in the alliances benefit from them. We have already suggested
that if an alliance is to be upheld the agents participating in the alliance cannot fight, but
maintaining an alliance also places requirements on agents who don’t directly participate in the
alliance (specifically, those who are allied with the agents in question). Consider Figure 1, which
depicts a simple network of alliances between six agents. It seems reasonable to require that for
the
alliance to be upheld
cannot fight
or
. If
fought
, then there is no way to
11
resolve the intermediate alliances in a way that does not break one. For example, if
and
with
then
ends up fighting with
. We do, however, allow
because in that case no direct alliance is broken if
fights with
and
fights with
to fight with
with
,
.5
Figure 1: Example Network
When there are no consequences for the network, conflicts always become violent, and everyone
involved in the conflict loses their endowment with probability . Equivalently,
could
represent the conditional probability that one loses their endowment when involved in a conflict
that has no consequences for the network. With this assumption we ignore several potential
effects that a richer model might incorporate, such as the possibility that the amount of remaining
endowment depends on the relative numbers of agents protecting one agent or the other. This
might suggest an advantage in fights for agents who are protected by more agents. One way to
think about our assumption that
is fixed is that these effects are small compared the incentives
discussed in the model.
The rules governing conflict are formalized using concepts from network theory. A path
between
and
distinct. Let
distance between
and
,
5
is a series of links,
,
,
,…,
∈ , such that each agent in the path is
represent the length of the shortest path connecting
and
in
(let
,
0 and
in ). Assuming all alliances are upheld,
and
,
and
in
or the
∞ if there is no path connecting
don’t fight when in conflict if
3. This is the logic underlying Figure 1. Agents don’t fight if there is a path of length
A result of this logic is that the critical “distance” (which we define in a later paragraph) between agents within
alliances is 3. Agents linked in set of alliances will not fight with other agents who are less than 3 links away from
themselves, but will fight with agents who are further than 3 links away.
12
3 or less connecting them. Otherwise they fight when a conflict occurs.
|
Then
,
3 is the set of agents with whom
those agents is in conflict. Let
#
) and
represent the complement of
and one of
(i.e., the agents not in
represent the number of agents in that set (we use # to represent
∩
the number of agents in ). It is convenient to also have
#
fights if
to represent the set of agents in
that are also in
and
and the size of that set,
respectively.
A few more pieces of notation from network theory will be useful in the presentation of
|
the results. Define ’s neighborhood as
#
allies, and ’s degree as
,
1 , which includes
and his
1. Note that this is the usual definition of a node’s
degree but not the usual definition of a neighborhood, as it includes . A network is called rmax
regular if every agent’s degree is . The diameter of a network,
,
,
, is
distance between the two agents that are furthest apart. It will also be useful to refer to the
diameter of the group, or the longest distance between two agents in the same group, as
max
,
, ∈
. The diameter is useful because the condition that
3 is equivalent to saying that no agents within
fight with each other (
3 would
imply that no agents fight with each other).
Since maintaining the alliance between
do, it is interesting to look at how
and
were harmed by the
and
and
places restrictions on what
and
value that alliance. One could imagine that if either
alliance they could decide to break it purposefully by
fighting when they were in conflict. To capture this concern, consider the additional requirement
that
iii. for all
∈
and
∈
∪
,
.
13
An alliance will always convey positive externalities to the neighbors of those who ally, because
it can only decrease the number of fights in which each neighbor is involved (the cost of an
alliance is borne entirely by the allied). Therefore, it will never be in an agent’s interest to break
an alliance between a neighbor and another agent.
With this notation, the expected payoff of an agent
in network
before transfers,
assuming that alliances are upheld is given by
1
.
1
∈
The expression inside the sum is the probability that some agent
or outside of his group. The sum is taken over all of
network
fights with agents either inside
’s allies and includes
. Manipulating the
is therefore consequential to some agent as long as it changes the number of agents he
fights against either directly or through an ally.
Isolated Groups: Groups without outside alliances
To analyze the model, we first focus on the interactions within a single group,
example, when the group is isolated (i.e.,
is an isolated group if no agent in
agent outside of ). In an isolated group, the alliances between members of
for
is allied with an
in this case only
affect who fights whom within , and since the probability of a conflict between two members of
is constant, whether an agent’s payoff increases or decreases with an additional alliance only
depends on whether the number of fights the agent is involved in rises or falls.
Since we are only concerned here with just one group, which is a subset of the possible
alliances, we cannot yet identify PST networks. We are asking which sets of alliances within an
isolated group are themselves pairwise stable and so could be part of a PST network. To simplify
the exposition, we call the set of alliances within an isolated group isolated group stable (IGS).
14
Individuals in the group still have conflicts with people outside the group with probability q, but
we assume that there are no alliances between group members and outside individuals. [I think
this is right]. This amounts to holding fixed the set of alliances between other groups the
alliances within the group are PST. Calculating the change in
deletion of an alliance with another agent in
forms a new alliance with
Suppose that
the alliance. This new alliance benefits
’s payoff due to the addition or
is straightforward after a few observations.
, who was someone with whom
(and his neighbors) by reducing the number of fights
is involved with directly and through his neighbors, while the cost to
he now has to protect
if
fought without
of this alliance is that
is involved in a fight after the alliance is formed. The net benefit to
before transfers is
1
1
.
∈
The first term in the brackets is the benefit to
of forming the alliance and counts the reduction
in the number of fights (including conflicts involving his allies) that
forming the new alliance with
. The second and third terms in the brackets represent the cost to
in terms of the additional fights that
third term is the probability that
is involved with after
is now involved in after the alliance is formed. The
fights with a member of an outside group. This term is as
large as possible in magnitude, because an isolated group fights with all other groups.
Although any alliance cannot increase the number of conflicts that become violent, an
alliance always increases the amount of violence experienced by individuals. This is most clearly
illustrated by observing that the network with no alliances between members of
when
is isolated. When there are no alliances in , adding the
probability that
fights by , because he no longer fights
is always IGS
alliance reduces the
, but this reduction costs him an
15
increase in the probability of fighting of
1
1
. The net benefit to
is strictly
negative as long as there are more than two agents in a group, which we have assumed (clearly
the same is true of
’s net benefit). If an agent forms an alliance with someone inside his group,
he necessarily increases the number of potential conflicts with individuals outside his group.
Remember that an isolated group has no alliances with outside individuals, but it is still subject
to conflict with outside agents with probability q.
As a result, in order for any non-empty set of within group alliances to be IGS, intragroup conflict must be more important to individual agents than inter-group conflict. Intra-group
conflict becomes more important as the probability of intra-group conflict, , rises relative to the
probability of inter-group conflict,
or as the number of outside groups, , falls relative to the
size of the groups, . Proposition 1 gives a necessary condition for , ,
and
for any set of
alliances to be IGS.
It may help in developing your intuition about the model to note that all of the conditions
that matter involve , ,
and . Individuals within a group only form alliances with other
group members if the probability of within group, intra-group, violence is high enough. Since
group members interact with each other more often than with non-group members, the potential
for violent conflict within the group is higher, and group members are willing to form alliances
within the group. The benefits of forming additional alliances depend on the probabilities of
conflict inside and outside the group, as well as the size and number of groups. Together, , ,
and
provide the conditions that constrain the behavior of individuals.
At another extreme, if
is much larger than , then sets of intra-group alliances that lead
to no violence between group members are IGS as long as the alliances satisfy one condition.
This is the second part of Proposition 1, and we call it Condition 1. This condition simply states
16
that the probability of being in conflict with one member of one’s own group is at least as likely
as being in conflict with some outside group member given , ,
Condition 1:
and .
1
It is not true that every network of alliances that has a diameter within the group of less
than three is IGS. Networks can be over-connected in the sense that they involve alliances that
do not reduce the amount of intra-group violence. We call an alliance
this alliance would lead to at least one agent in
is, after removing the alliance
and
critical if removing
fighting at least one agent in
. That
are involved in at least one additional fight with each
other. For an example of an alliance that is not critical, consider a group of three agents ( ,
and
) where each is allied with the other two. The diameter of the group in this case is one, so
there is no intra-group violence, but none of the alliances are critical, as after removing any of
them the diameter of the group is still less than three. Alliances that are not critical cannot be part
of an IGS network when the group is isolated, because they carry with them the increased burden
of fighting outside groups more often.
The following proposition lays out the combinations of p, q, k, and n that can support
various configurations of IGS in isolated groups.
Proposition 1: If
4
1 , then the only IGS network is the empty network (i.e., the one
involving with no alliances). On the other hand, if
1
(Condition 1 holds) then both
the empty network is IGS and any network involving only critical alliances that leads to no
violence between members of the group (
3) is IGS.
Proof: To show that the first statement is true it is sufficient to show that under the condition the
most valuable possible alliance cannot form. An agent allied with
be involved in at most 1
1
other agents in a group can
/4 fights with members of the same group
(i.e., the number of fights that would occur if the agent and every one of his allies fought with all
17
of the remaining agents). The upper bound is only achieved if
Suppose that
and
are each involved in
each have
∗
is even when
∗
/2
alliances, and that no other alliances in . Then
1.
and
/4 fights and an alliance between them would eliminate all of these
/4
fights. Therefore, for either agent the net benefit to this alliance is
4
which can be positive if and only if
1
1
,
1 .
For the second statement, the fact that the empty network is group stable follows from
discussion preceding the proof (an alliance in this case reduces the probability of fighting by ,
1
but costs an increase in probability of
1
), as does the fact that non-critical
alliances cannot be part of a group stable network. What remains is to verify that no agent wants
3 and there every alliance is critical. The condition
to dissolve an alliance when
1
guarantees that if
and
each avoid at least one intra-group fight from this
alliance it cannot hurt them to form it. But they must avoid a fight, because after removing the
alliance they must fight directly. QED
Using Proposition 1 it is straightforward to construct examples of IGS groups. Figure 2
shows several examples of group networks to illustrate the restrictions that the Proposition puts
on intra-group networks. When Condition 1 holds, the networks in (a) and (b) are IGS since all
agents are no further than three alliances from each other and all alliances are critical. The
network in (c) is not IGS, because the only critical alliances are
network in (d) is not IGS, because
benefit both of them.
and
,
,
, and
fight in this network and the alliance
. The
would
18
(a)
(b)
(c)
(d)
Figure 2: Isolated Group Stability
We discuss the IGS group in more detail later, but it is important to point out how the
logic of the group works. Although any two individuals have alliances only with the individuals
next to them, because of the distance rule of 3, every individual is committed to coming to the
aid of every other individual. This group looks from the outside as if all the members are all
committed to each other. This is not a logic of one for all and all for one, however. Adding
another member breaks the group down, because now everyone in the group will fight one other
member of the group should a conflict between them arise.
What happens to the set of IGS networks as Condition 1 is relaxed and we move into
other areas falling under Proposition 1? Consider the case where
1
1
/2
. In this case, if two or more intra-group fights per allied agent (or four total fights for the
two agents in the alliance) can be avoided the alliance is mutually beneficial (but not otherwise).
Under this condition, (a) is still IGS because deleting any alliance causes the agents involved in
that alliance to each experience three additional fights. Deleting alliances in (b) is even more
costly for both agents, so it continues to be IGS as well. The conclusion changes about (d). The
alliance only eliminates one intra-group fight each for
longer beneficial (as it was under Condition 1). The
group fights by one for
and three for
and
, so this alliance is no
alliance decreases the number of intra-
, so it is mutually beneficial.
19
IGS groups that prevent intra-group violence still experience significant levels of
violence through fighting with other groups, and how the violence, and hence the expected
surplus, is distributed throughout the group depends critically on the structure of the alliances.
Two extreme examples are shown in Figures 2(a) and 2(b). Figure 2(a) is an example of
what we call an egalitarian network, in which each agent in the group experiences the same
amount of violence. To be precise, we might say that the group is “egalitarian in violence”,
because it is possible that transfers are specified to make payoffs unequal (we explicitly consider
transfers below). More generally an IGS network is egalitarian if every agent in the network has
the same number of allies (i.e., it is regular in the network theory sense). We include groups with
no alliances in this definition as well.
Figure 2(b) is called a star (a star is a group where one agent is linked to every other
agent and there are no other links). This structure is the most inequitable of the IGS networks,
because a single agent allies with every other agent and so experiences the highest amount of
violence. The remaining agents are each only responsible for one alliance. The importance of the
star stems from the fact that the agent at the middle of the star has the strongest possible
incentives to form alliances with other groups. This is a result of
benefitting from eliminating
fights with any of the members of .
This is in contrast to egalitarian networks which have weak incentives to connect with
each other, because they tend to minimize the maximum degree of the agents, among the
networks that are IGS. For any network with a diameter of at most three and
agents, a simple
argument establishes that
1
where
,
is the maximum degree of any agent in the network. This is a special case of the Moore
20
bound on the size of a network with given upper bounds on diameter and maximum degree (see
Bollobas 1982, p.171). It follows that the network in 2(a) for example achieves the lowest
possible
given six agents in a network of diameter less than four. In other words, we can rule
out the possibility that there is another network with diameter less than four and six agents in
which all agents have a lower degree (one in this case). Although this is obvious in 2(a), we can
find more complicated networks, such as the ones in 4(a) and 4(b) below, that are egalitarian and
achieve the lowest possible
when
12. For 5
7, arrange the agents in a circle and
connect each to the neighbors on either side. We have also found examples of egalitarian IGS
networks for
10,14 with
3. If
3, it is not possible for an egalitarian network to
have an odd number of agents, since the sum of the degrees in a network must equal twice the
number of edges (this is the so called Handshaking Lemma of Euler).
The Moore bound is provably unreachable in all but a limited number of cases (Bollobas
1982), and characterizing all of the networks that are close to achieving the Moore bound is an
open problem. However, regular networks (i.e., egalitarian networks) have been shown to come
close to achieving the bound,6 which supports the idea the intuition that egalitarian networks tend
to minimize the maximum degree of the agents.
Networks of Many Groups
We are now ready to evaluate which networks with many groups can be considered PST.
In particular we are interested in the conditions under which inter-group alliances form, the
relationship between egalitarian groups, and the incentives of any individual to form alliances
across groups.
Although there are now many more possible candidate networks and there are potentially
6
Bollobás and de la Vega (1982) who show that a regular networks chosen with uniform probability come close to
achieving the bound as is allowed to grow.
21
complicated interactions between the values of inter-group and intra-group links, under
Condition 1 we can focus attention on those within group networks identified in Proposition 1.
Consider an IGS network involving no intra-group group violence, but now allow some of the
agents to be allied with other groups (assume these alliances are critical). When the group is
isolated, the cost of adding an intra-group alliance includes the probability of having to protect
that additional agent from all outside group members. When there are inter-group alliances, this
cost falls, because agents do not fight with every outside group anymore. So if a pair of agents
could not benefit from breaking an intra-group alliance when the group was isolated, the pair will
not benefit from breaking the alliance when the group is not isolated. It is possible that with
inter-group alliances new intra-group alliances may be beneficial, but this possibility will not
arise in the cases covered here.
This suggests that PST networks can be constructed by taking multiple IGS groups and
checking whether any inter-group alliances will form. Inter-group conflict between any pair of
agents occurs with probability , and when no intra-group alliances are affected by the addition
of an inter-group alliance (and the shares are fixed as we continue to assume), the only
consideration for the agents in forming inter-group alliances is whether the number of intergroup fights they are involved in increases or decreases in response to the new alliance.
Our first observation is that, as long as there are enough groups, a network involving only
isolated egalitarian groups can be pairwise stable. For example, consider the case where there are
groups organized as they are in Figure 2(a) with six agents each. When all of the groups are
isolated, the pairwise stability with transfers of the intra-group alliances follows from the fact
that they are IGS. The network is therefore PST if no inter-group alliances are mutually
beneficial. The benefit of adding an alliance between any two of the agents across two groups,
22
say
, is that each avoid 11 fights involving members of the two groups.7 The cost is that
and
, for example, must now fight with
2 6
against groups , , … and well as
1 agents total. As long as there are at least four groups total (
would harm
and
, of which there are
4) this alliance
. The next proposition generalizes this logic to egalitarian groups of
arbitrary size.
Although a network involving four isolated egalitarian groups is PST, the network with
four isolated stars (Figure 2(b)) is not. The reason is that two agents that are the centers of each
of their respective stars have stronger incentives to ally with each other than any of the agents in
the egalitarian groups. Using the configuration in Figure 2(b), an alliance between the centers of
two of these groups carries a benefit of avoiding all of the fights between the groups, of which
there are 36 (
) in this case. This outweighs the additional 12 fights that one of the agents is
involved in after forming the alliance. Using similar logic, one can show that continuing to add
alliances between the centers of the stars is beneficial to both agents in the alliance, so we are left
with a network with each star center being connected to every other star center.
The intuition in this result underlies an important element in the empirical implications of
the model. In groups formed in stars, where one individual is connected to everyone, that
individual experiences a higher level of violence. When we look at “societies” with many
groups, , alliances across groups are much more likely if the component groups are organized
as stars than as egalitarian groups. The individuals forming the inter-group alliances will be the
individuals at the nodes in the stars.
A network of isolated stars is less likely than a network of isolated egalitarian groups, in
the sense that stronger restrictions on parameters are required. The next proposition generalizes
7
After adding the alliance, avoids directly fighting five of the
and (see Figure 2(a)) is three.
each of
agents. The number of fights encountered by
23
these examples.
Proposition 2: Suppose that there are egalitarian IGS groups of members each, in which
each member is allied with
2 fellow group members with
1
. The network involving
2
1 . If instead there are k IGS
no inter-group alliances is pairwise stable if 3
stars, the network with no inter-group alliances is pairwise stable if
2. Given an and
an satisfying
1
, if is large enough for the network of isolated stars to be stable, it is
large enough for the network of egalitarian groups to be stable.
Proof: Consider an alliance between any two members of distinct egalitarian groups, between
and
for example. By forming this alliance,
(or
) avoids 1
1
fights, since this is the number of agents that are no further than two links from
on
guarantees that there are no fewer agents in the group).
1
fewer times, and the new alliance requires that
2
1
away from
protect
cannot benefit from the alliance if 3
net benefit of connecting two stars is
direct
(the condition
protects his existing allies
times (the second term reflects the agents in
. Therefore,
1
in an additional
(if any) that are three links
2
1
2 to each agent in the center of the star, which
gives the inequality for IGS stars to be stable. For the last statement, assume that
1
2. Together these imply that
1
that the network of isolated stars is stable,
3
2
3
. The
and
2, so that the condition for the egalitarian network also holds. QED
For networks with no inter-group alliances to be PST, the conditions in the proposition
require that there be enough outside groups (i.e., a large enough ). The last statement of the
proposition shows that there needs to be more outside groups for isolated stars to be stable than
there does for isolated egalitarian groups to be stable.
By playing with different combinations of these group structures one can generate
heterogeneous networks with some inter-group fighting. Figure 3 is one example, where there
are two stars,
and , connected with six outside egalitarian groups. In this network,
and
24
are indifferent between maintaining and not maintaining their alliance. The most valuable
alliances not present are those between either
or
and any one of the 36 egalitarian group
members.
Figure 3: Pairwise Stable Network with an Inter-group Alliance
It is the presence of the different types of groups that allows intermediate outcomes like
this. As suggested above, if there are only stars and one alliance between groups is beneficial
then any alliance between any two disconnected groups is beneficial, so it is not possible to have
disconnected stars if the inequality in Proposition 2 does not hold. What keeps the network in
Figure 3 pairwise stable is the fact that an alliance between stars being beneficial does not imply
that an alliance between egalitarian groups is beneficial.
Wealth Distribution
The stability concept we have employed puts conditions on the changes in the joint
surplus associated with the formation and deletion of alliances. These conditions essentially
require that there be some set of pairwise transfers that are feasible and make both agents
involved in an alliance better off. By pairwise transfers, we mean direct transfers between allied
agents (i.e., ones that redistribute the surplus between the players forming an alliance). There is,
of course, no requirement that these transfers be unique, and there will generally be a continuum
of transfers that incentivize the agents to maintain the alliances that have been deemed stable. In
25
this section, we propose a specific method for determining transfers that we believe has intuitive
appeal for the current model.
,
To discuss transfers, we introduce a couple of additional pieces of notation. Let
denote the transfer from
indicating that
to
associated with the alliance
transfers the amount to
transfers in network
is
∑
∈
(let
,
\
,
in network
,
with
,
0
). Agent ’s payoff with
where the second term adds up the
transfers attributable to each of his neighbors. With transfers associated with each of his
alliances, the net change in ’s payoff when he breaks an alliance with
,
. If this expression is negative, then
is
does not have the unilateral incentive,
given the specified transfers, to break the alliance. The stability condition we have used
guarantees that there is some set of transfers that make this expression negative for every
every alliance that
,
is involved in, since
,
Let
alliance
and
0.
, ,
,
represent ’s marginal value of the
in network , or equivalently the loss experienced by
in the event that
is
severed. One might think of agents that are “less likely” to sever the alliances that carry higher
marginal values. There is thus a sense in which an alliance that has higher marginal values
associated with it is “more secure”. By adjusting the transfers between agents, obviously one
agent’s marginal value must increase while the other’s must fall. If the security of an alliance is
determined by the lowest marginal value for that alliance, then the security is maximized when
both agents received the same marginal value for the alliance.
The pairwise stability with transfers conditions ensure that it is possible to specify
transfers such that every alliance provides the same (non-negative) marginal values to both
26
involved agents. For such transfers,
, ,
/2, which is always non-negative in a pairwise stable network with transfers, so
that these transfers give each agent the incentive to maintain their alliances.
Since in egalitarian networks the amount of violence experienced by each agent is the
same, it is not surprising that the marginal values of the alliances can be equalized in such
networks, which would lead there to be no transfers occurring between the agents. This is the
case in the six person egalitarian groups used as examples so far. With more agents, there are
examples of egalitarian networks where the marginal values are not equalized, such as the
example in 4(a). Removing the
alliance only results in
transfers the alliance does not benefit
end up transferring wealth to
(note that
, leading
fighting
in this case, so before
at all. In the transfer scheme outlined here,
would
to capture a larger share of the groups resources
in this example). In the example in 4(b), the marginal values of each
alliance are equalized and there would be zero transfers under the scheme above, and the ex-ante
payoffs are equalized across members of the group.
27
(a)
(b)
Figure 4: Egalitarian IGS networks with non-zero transfers (a) and zero transfers (b)
In star networks, the “center” of the star plays a critical role in connecting groups to one
another, as is shown in Figure 3, and this leads the centers to capture more of the resources
available to the group. In Figure 3, there is a strong incentive for
, due to
’s alliance with
. This incentive is reflected in the transfer that
this case. Removing this alliance would lead
with all members of
is 5
6
36
at the cost of protecting
5
to maintain his alliance with
30 ). For
makes to
in
to fight all of his fellow group members along
(the net reduction in the probability of fighting
, the alliance is less valuable because for him it
eliminates five intra-group fights and one inter-group fight (the net reduction in the probability of
fighting is 5
36
transfer of 5 1
/2 to
receives 25 1
5
35 ). Equalizing the marginal values requires that
. Since the same is true of all of
’s intra-group alliances, he
/2 in total transfers (there is no transfer between
The source of the transfers that
make a
receives in this example is
and
).
’s relationship with
28
,…,
through his alliance with
non-central agents in
and
. In other words, it is the prevention of fighting between the
that drives the difference in marginal values of the intra-group
alliances, and hence the transfers. It is important that
have the alliance with
, too, as being
the center of the star does not ensure that transfers take place (under this transfer rule of course).
In the isolated star of 2(b), for example, each alliance provides the same marginal benefit to both
parties, so no transfers take place.
With transfers, the central agents who receive the transfers are wealthier than the other
agents in the group in the sense that they control more resources before fighting occurs, but they
are not necessarily better off than the other agents once one takes into account the amount of
violence that they experience. The stark difference in the number of alliances that they take on
means that the central agents are much more likely to encounter violence. In Figure 3, it is the
peripheral agents that benefit the most from the arrangement, even after the transfers take place.
The model presented in this section provides several predictions about the structure of groups
and the level of violence in human societies. First, in a society made up of small egalitarian groups
all individuals will (in conceptual terms) experience the same high level of violence. Second, larger
groups can emerge when some individuals make connections between groups. These connections
can reduce the overall level of violence in society by building networks of coalitions that will not
fight each other. The cost to individuals who provide the connection, however, is higher exposure to
violence. In the model with transfers, the individuals who are the recipients of net transfers are those
in the hierarchy of individuals providing connections between groups. Those individuals enjoy a
higher material standard of living, but whether they are better off in terms of utility is problematic.
Their higher material standard of living comes at the cost of higher probabilities of violence.
Symmetrically, most individuals in the world with transfers have lower material standards of
29
living, but they experience lower expected levels of violence. Later, in the empirical sections, we put
those predictions to the data.
A Less Formal Model
The individual based model yields predictions about violence, size of society, and hierarchy
in material living standards that we can take to the historical data. The hierarchical “connector”
model builds up from individuals to groups. There is another way to reach the same point by
working down from social orders towards the organization of groups and leaders within a
hierarchical society. This section develops the intuition of the model, connecting ideas about
political and economic organizations from North, Wallis, and Weingast (2009, NWW) particularly
as developed in Wallis (2011). The logic of the hierarchical connector groups on the left hand side
of Figure 3 does not per se connect directly with the modern world. Extending the model’s basic
intuitions about centers and alliances can, however, be put in framework that works quite well at
explaining many societies in the contemporary world. In those societies, powerful organizations
manipulate the economy to produce rents for the organization’s leaders, and then use the economic
rents to coordinate a political coalition. Coordination of the political coalition serves to limit
violence by the same logic as the individual model: by creating interlocking alliances between
centers. NWW call this the “logic of the natural state.” This section links the logic of the natural
state with the logic of the hierarchical connectors.
NWW considered the problem of how powerful individuals could reach credible
arrangements to limit violence. If two powerful individuals agree not to fight, and the first
individual puts down his rock, what is to prevent the second individual to bang him over the head
with his rock? Their solution was, in simple terms, to recognize the importance of organizations.
Starting from the same assumption that all individuals are endowed with basically the same amount
30
of violence potential, in order for violence to be effective it must be organized. The leaders of
organizations pose threats to one another, how can they reach credible agreements? They agree to
divide the land, labor, and capital in their world between themselves and agree to enforce each
other’s privileged access to their resources. In return for a promise to defend each other’s interest,
the connectors acquire exclusive (or limited) access to resources and functions. The connectors are
able to commit to one another, that is, they can each credibly believe that the other will not fight, if
the income of the specialists is sufficiently lower under conditions of violence than conditions of
peace. The rents they earn from peace enable the specialists to coordinate and cooperate, even
though the specialists remain armed, dangerous, and a serious threat to each other. Through their
combined coercion, each specialist can credibly threaten the people around them to ensure each
other’s rights.
The arrangement is represented graphically in Figure 5, where A and B are the two
connectors, the horizontal ellipse represents the arrangement between the specialists that create
their agreement/organization. The vertical ellipses represent the arrangements the specialists
have with the labor, land, capital, and resources they control: their “clients,” the a’s and b’s. It is
very important to understand that the horizontal arrangement between the specialists is made
credible by the vertical arrangements. The rents the connectors receive from controlling their
client organizations enable them to credibly commit to one another, since the rents are reduced if
cooperation fails and the connectors fight. There is a reciprocal effect. The existence of the
agreement between the connectors enables each of them to better structure their client
organizations, because they can call on each other for external support. The leader’s
organization is what NWW call the “dominant coalition,” but we will call the dominant network.
Figure 5 is simply the alliance between the two star groups depicted on the left side of
31
Figure 3. As the hierarchical connector model shows, it is possible for such alliances to be
maintained in the presence of egalitarian groups. By design, the connectors in Figure 3 are
individuals who happen to have patterns of network connections.
To move from the logic of hierarchical connectors to the logic of the natural state
requires us to be explicit about organizations. Organizations are groups of people with shared
interests and goals. An adherent organization is one where all of the members have an interest
in cooperating with each other (on the relevant dimensions of the organized activity) at all points
in time. In an adherent organization interests are structured in such a way that all individuals
have an interest in belonging to the organization, even if their interests are the result of being
coerced. In the connector model, where the only dimension that choices can affect is violence,
the IGS groups are adherent organizations.
In contrast, a contractual organization is one where relationships between the group
members are not inherently self-sustaining, and the group maintains itself only through the
presence (or potential presence) of an external third party. The third party may enforce
relationships within the organization or between the organization and other external parties.
There are no contractual organizations in the hierarchical connector model by design.
In Figure 5, the horizontal relationships between the connectors create an adherent
organization. The vertical relationships between the connectors and their clients are contractual
organizations because they rely on the external presence of the other connectors. The vertical
organizations might be organized as kin groups, ethnic groups, or patron-client networks. The
combination of multiple organizations, the “organization of organizations,” mitigates the
problem of violence between the really dangerous people, the connectors, creates credible
commitments between the connectors by structuring their interests, and creates a modicum of
32
belief that the connectors and their clients share a common interest because the specialists have a
claim on the output of their clients.
The key to the whole arrangement is that the rents A and B derive from their
organizations enable them to credibly commit to one another. The interests created by these
organizations must interlock, that is, the ability of A and B to form organizations depends on
their coordination and cooperation, since the vertical/contractual organizations are structured by
the third-party enforcement of the dominant network. Figure 5 suggests two important
implications of establishing even a nascent natural state. One involves the dynamics of group
interaction, the other about the new dynamics of social stability.
Hunter-gatherer groups are aggressively egalitarian. This feature is based on inherent,
genetic capabilities in all humans that make them capable of dealing with complicated social
situations. As long as individuals have experience with one another and continued social
interaction, groups are able to maintain cohesiveness at small sizes and they are naturally fair.
Egalitarian outcomes are also based, in part, on the inevitable fact that connectors must sleep and
when an oppressive leader sleeps he is easily killed. In a small group of people it is difficult for
one individual to suppress or expropriate the collective group.8 The logic of hunter-gatherer
bands is always the logic of an adherent organization, all members of the band must find it in
their interests to remain in the band or it breaks up. It is the logic of the egalitarian groups in
Figures 2 and 3.
The relationship between A and B in Figure 5 may change the internal dynamics of the
group. A can now call on the services of B to enforce his rights and privileges. The first “rule”
that may develop is that B comes to the aid of A whenever B calls, and vice versa. In simple
bands, connectors are unable to use coercion to coordinate behavior of group members. In
8
For detailed consideration of this point see Boehm, pp. 72-89, on antiauthoritarian sanctions.
33
contrast in natural states, connectors can credibly generate coercion by calling on people outside
of their immediate group. The emergence of a dominant network transforms the nature of the
vertical, client organizations in Figure 2 from adherent to contractual. The organization of the
client organizations is therefore tied directly to the organization of the dominant network, the
horizontal adherent organization. The rents A receives from his organization depend on the
contractual rules that he can enforce because of his relationship with B. A also receives rents
because of his unique identity within the coalition as the leader of organization A, and the fact
that he is the only person who can draw on the third-party services of B. Rules that strengthen
A’s organization, but weaken A’s mutual dependence on B, and thus weaken the ties within the
dominant network, are less likely to survive periods of instability. Organizational rules serve
two purposes, organizing the client organization and providing rents in the dominant coalition,
and the two purposes may be conflict.
The society depicted in Figure 5 is too simple to function, but it serves to illustrate how
we can move from the individual oriented world of the hierarchical connected model, to the
organization oriented world of the natural state model. In Figure 5, the connectors are “elites.”
They are elites because of their organizational arrangements with each other, not because of
some comparative physical advantage in violence. Their comparative advantage in organized
violence is the result of their alliance (in the connector model), or of the organization (in the
natural state mode). There seems to be little or no doubt in the historical record that when larger
societies form, trade is an integral part of the social structure, that the economy, particularly
trade, is controlled by the dominant network. Some individuals within the social hierarchy enjoy
privileged positions with respect to trade and markets.2 As Earle puts it when discussing the
2
For studies of “pristine” civilizations, that is the first large civilizations that arouse without a
geographic predecessor in Mesopotamia, Egypt, India, China, Meso America, and South
34
emergence of chiefdoms (groups of over 200 or so):
In chiefdoms, control over production and exchange of subsistence and wealth creates the
basis for political power... Economic power is based on the ability to restrict access to
key productive resources or consumptive goods... Control over exchange permits the
extension of economic control over broader regions,... The real significance of economic
power may be that the material flows through the political economy can be used by the
chief to nurture and sustain the alternative power sources...” (1997: 7)
The point is not to quibble about which came first, larger societies, social hierarchies, or trade
and markets, but to acknowledge that all three elements of societies appear together in the
historical record. Over historical time they have been intrinsically linked. Their simultaneous
relationship, their endogeniety, is not in question, but instead forms the basis for thinking about
the dynamics of their relationship.
The entire complex of organizations creates a set of incentives and interests for powerful
individuals leading to cooperative outcomes. Organizations occupy the central place in this
process and limiting access to organizations shapes interests. Organizations are a primary driver
of both the shape of institutions and their change over time. Cooperation cannot be sustained
unless powerful individuals believe that cooperation is in the interest of other powerful
individuals. Organizations structure interests to facilitate cooperation.
When we extend the logic of the hierarchical connectors to the logic of the natural state, we
maintain two testable implications. First, that as the size of social groups expands, the level of
violence should fall because powerful individuals have the military power to suppress conflict by
their clients and to reach stable agreements between themselves. Second, that as the size of societies
grows, from hunter-gatherer bands to cities and states, we expect to see the material standard of
living of the city dwellers fall. Ala the connector model, a falling material standard of living does
American, see Service (1975) and Trigger (2003). For the anthropological record on the
emergence of larger societies see Earle (1997 and 2002) and Johnson and Earle (2000).
35
not necessarily imply lower utility. The city dwellers are buying peace with a lower material
standard of living. The implications of the logic of the natural state that go untested in this paper,
are that not only will hierarchical discrimination between individuals arise, but the hierarchy will
take a particular form. A dominant network of interlocking political, economic, and social players
coordinate their activities through the creation of rents. These rents constrain the interests of
powerful individuals, enabling them to make credible commitments to one another. The rents are
created, in part, by providing coalition members with the means of creating more productive and
complicated organizations.
36
Figure 5
III. Measuring Violence from Skeletons
The model suggests that larger societies should, ceteris paribus, exhibit measurably lower
levels of humanly induced violence than their hunter-gatherer contemporaries. Evidence to test this
hypothesis is available in the archeological record, but interpretation requires understanding of ways
that deliberate trauma affects the skeleton.
Types of skeletal injuries. Skeletal trauma may be classified into three types: antemortem,
perimortem, and postmortem (Walker 2001). Antemortem fractures are easily visible because
healing produces a well-defined callus of new bone that usually persists throughout life. Wellaligned fractures in children may remodel and obscure an injury but in the absence of modern
medical technology such as pins, screws and casts that hold bones in place, the chances of good
alignment were poor. Thus in archaeological populations, most significant skeletal injuries prior to
37
death left permanent, visible evidence of repair.
Damage without visible signs of healing must have occurred near the time of death or was
the result of postmortem destruction in situ or during excavation. Using bone color and the pattern
of breakage, a trained physical anthropologist can distinguish between these causes. Living bone
breaks like glass or sheets of hard plastic, leaving oblique angles and sharp edges, but soon after
death bones typically lose collagen and become brittle. Dry bones break like a piece of chalk, at
right angles. Marked color differences between the surface and the interior of the bone suggest
breakage after death.
For obvious reasons, healed trauma is treated as evidence of violence and dry bone breaks are
not. One could argue either way on perimortem breaks, but inclusion is sensible because they could
have caused death; the downside is that some breaks might result from rough handling (or worse)
after death and before burial. At the risk of some overstatement of violence among the living,
perimortem trauma is treated as evidence of violence.
It is likely, however, that skeletal injuries understate violence on balance. Broken bones are
just the tip of the iceberg because many wounds affect only soft tissues. Only one assault injury in
seven can be classified as skeletal in the United States (Rand 1997). Philip Walker estimates that the
frontal view of the skeleton represent about 60 per cent of the target area for a projectile (Walker
2001). Of course, the area affected depends upon the size and number of the projectile(s), and
weapons technology in general. Hand-to-hand combat with clubs creates many skeletal injuries,
especially to the head.
The first step in interpreting skeletal injuries is to determine the proximate cause, which can
be learned by combining mechanical properties of bone with known characteristics of injuries
caused by blunt instruments, projectiles, and the like (Spitz 1980). Narrowing the range of plausible
38
mechanical causes prepares the next, more difficult step of determining social, cultural and
economic factors behind injuries, which is the objective of this paper.
Accidents versus violence. How can one tell whether a damaged bone is merely the result of
an accident as opposed to intentional violence? Clues are available from several sources, including
the location of the bone, the type of damage, presence or absence of multiple trauma and weapons
technology in use. A projectile point embedded in the back, as found in 9,000 year old Kennewick
man (Washington) and 5,300 year old Otzi the ice man (northern Italy), leaves little doubt about the
cause. Although there could be explanations other than violence, such as funerary rituals, cut marks
on bones usually signify deliberate trauma. In a famous example of misinterpretation, however,
Raymond Dart (Dart 1953) constructed a dismal picture of rampant violence in the human past by
mistaking carnivore tooth marks for blade cuts (Cartmill 1993). Modern osteologists know that
careful study in well preserved remains can distinguish the two causes.
People instinctively protect their head during a fall. A common site of accidental injury is
the lower arm, wrist and hand, which is extended to lessen the blow when hitting the ground.
Uneven terrain, characteristic of high elevations, complicates mobility and increases these types of
injuries (Larsen 1997). Bones of the ankles and legs also suffer many accidents and are more often
broken at high elevations. Not all lower arm fractures result from accidents, however. A type of
break called a “parry fracture” occurs to the middle of the ulna when an assault victim raises his arm
to deflect a blow to the head, but this type of break can also happen if the arm is twisted during a
fall. This qualification aside, arm, wrist, hand and leg fractures are a reasonably good indicator of
accidental trauma.
While deliberate aggression can affect any part of the body, the head is a common location of
interpersonal violence (Larsen 1997). Holes left by bullets and by weapons such as pikes, swords
39
and knives are reasonably clear evidence of intent, as are depression fractures to the cranium, left by
clubs or other blunt instruments. Broken noses and jaws also reflect interpersonal conflict. One
could imagine a few head injuries occurring by accident, but it is safe to argue that the vast majority
result from violent intent.
Modern Evidence Provides Clues of Ancient Patterns. Modern studies have identified
numerous factors associated with violence, including demographic characteristics of individuals,
geographic location, ethnic status, type and availability of weapons, poverty, and diet or substances
consumed. Although the literature considers many other covariates, the discussion here is tailored to
those available for analysis of the skeletal data. Some of the factors we discuss here cannot be
controlled for in the historical data, but we want to consider the possibility that the missing data may
explain the pattern of results reported here.
Age and sex are clearly associated with modern patterns of trauma. Men are considerably
more likely than women to experience injuries of all types (Baker 1992). Rand indicates that males
are responsible for 84 per cent of assaults treated in emergency rooms (Rand 1997). The median
age of perpetrators in American homicides was 20 years (87 per cent were males) and that of victims
was 25 years, of whom 78 per cent were male (Fox 1987). A relatively large share of women (57
per cent) compared with men (17 per cent) are victims of violence by family members or intimate
partners (Rand 1997). Fox reports that homicides have approximately the same pattern (Fox 1987).
Sex differences also apply to the bones affected. In England, 83 per cent of assault victims
had facial injury (Shepherd, et al. 1990), and a significantly higher proportion of women (56%)
compared with men (26%) had these fractures (Shepherd, et al. 1988). Whether people are
genetically programmed to assault the face is an open question, but contrary evidence comes from a
positive correlation between the rise of modern boxing and an increase in the proportion of
40
homicides caused by hitting and kicking (Walker 1997).
The NRC study of the United States (Reiss and Roth 1993) confirms the age and sex pattern
of violence found in England. Nearly 90 per cent of those arrested for violent crimes in the United
States are men (p. 72), and the lifetime risk of homicide is about three to four times greater for men
than women (p. 69). The risk of victimization by violent crime peaks at ages 16 to 19 for both men
and women and declines substantially with age. Somewhat higher rates of victimization apply to
minorities, with rates per 1,000 being 39.7 for blacks, 37.3 for Hispanics and 28.2 for whites (p. 69).
In 1990 one half of all homicide victims were black and their rate of victimization was six times
higher than for whites (p. 70). The connection with ethnicity may operate through other variables
such as income, education and place of residence. Violent crime rates increase monotonically by
community size, with the cities over 1,000,000 population having rates roughly five times higher
than communities of 10,000 or less in size, a differences that has widened since the early 1970s.
The classic study by Shaw and McKay (Shaw 1942) on the ecology of crime reported that
rates increased with poverty, ethnic heterogeneity and residential mobility. Two of these factors,
poverty and high mobility, were present when the twentieth century peak in homicide rates was
reached in the early 1930s (Reiss and Roth 1993). But is it absolute or relative poverty that matters?
Several studies point to the importance of community inequality (Reiss and Roth 1994). Other
studies have focused on the role of alcohol (Miczek, et al. 1993, Reiss and Roth 1993), weapons
(Ord and Benian 1995, Reiss and Roth 1993), and sugar consumption (Kanarek 1993). We cannot
control for any of these factors directly in our empirical tests, but we can consider whether they
might influence our results.
In recent years numerous studies have appeared on battered children (Donnelly and Oates
2000, Helfer, et al. 1997). It is difficult to determine whether the concern stems from a genuine
41
increase in the incidence of the phenomena or simply greater social awareness. By the midnineteenth century social commentators of the city, such as Charles Dickens, had the topic of
violence and neglect of women and children on their radar screens (Dickens 1966, Dickens 1965).
During the Victorian era children’s rights advanced with the passage of laws mandating education
and limiting work. In 1875 the New York Society for the Prevention of Cruelty to Children was
incorporated. Some studies, however, suggest that child abuse has existed throughout recorded
history (Bakan 1971). Study of skeletons may help to inform the debate.
Patterns of violence in ancient societies, distilled from several anthropological studies,
indicate some parallels with modern evidence. Angel found a higher rate of traumatic injuries
(especially to the head and neck) among males relative to females in the eastern Mediterranean
(Angel 1974). Robb reports that with the rise of agriculture cranial trauma increased for males
relative to females and by the Iron Age all types of trauma were higher among men (Robb 1997).
He suggests that gender roles evolved to expect violent behavior for men, but it is difficult to test
this hypothesis.
In sum, if modern patterns of violence extended to the distant past, what relationships might
be found in the skeletal data with respect to covariates that are available here for study? Certainly
there would be more trauma among men than women, and the pattern is so widespread in the recent
past that its absence would cast doubt on the veracity of the skeletal data. Estimated rates that are
roughly equal for men and women might then indicate deaths of men in battles at unexcavated
locations, which would tell in the sex ratio of deaths in the database. In the modern world, violence
is heavily concentrated against older teenagers and individuals in their twenties, and data since the
mid-nineteenth century points to some abuse, and therefore trauma, among children. Today violence
is highly correlated (positively) with community size, may rise with the power of weapons and
42
probably increases with consumption of alcohol, although social responses vary widely. To the
extent that modern patterns existed in the past, we expect to see rural hunter-gatherers exhibiting
lower rates of violence that urban village and city dwellers.
Poverty is a risk factor for violence in the modern world but it seems implausible to argue for
a biological imperative, or a level of material poverty below which people automatically descend
into violence. If absolute lack of material goods caused aggression, however, the distant past should
have been quite violent relative to the present given the enormous increase in living standards over
time. On the other hand, income or material goods might have been potent only in a relative sense,
in which case inequality was the driving force. Perhaps envy was the primary motive for crime and
aggression, and those locations known for inequality in material goods or power (e.g. cities), should
have been sites of violence. Plausibly a sudden decline in living standards, or thwarting of
expectations, caused by phenomena such as natural disasters or sudden climate change, could trigger
violence. Again, since village and city dwellers exhibit clear evidence of lower physical standards of
living, we might expect to see higher rates of violence in the urban populations.
IV. New Skeletal Evidence and Empirical Results
About 15 years ago Richard Steckel and Jerome Rose invited a large group of physical
anthropologists, economic historians, demographers and medical historians to document and analyze
the history of health in the Western Hemisphere using data from archaeological skeletons (Steckel
and Rose 2002). Anthropologists who contributed data on several skeletal indicators of health for
individuals who had lived at sites scattered from South America to southern Canada. The combined
dataset includes 12,520 skeletons from 65 localities representing populations who lived from 4,500
B.C. to the early 20th century. As explained below, some sites were deleted from the statistical
43
analysis and some skeletons lacked estimates of age or did not have the requisite bones for study of
trauma.
Table 1 lists information on the geographic, ethnic, and temporal distribution of the
skeletons. Nearly 80 per cent were Native Americans, with the remainder almost evenly split
between Euro-Americans and Afro-Americans. About two-thirds of the natives resided in North
America as opposed to Middle America (11.9 per cent) or South America (22.2 per cent). Slightly
more than one-half (52.6 per cent) lived in the Western Hemisphere prior to the arrival of Columbus
and nearly 14 per cent lived more than 2,000 years ago.
Health. Table 2 presents several skeletal markers of health for hunter-gatherers and for town
or city dwellers who lived in pre-Columbian America.9 All point to an important decline in health
following the transition to sedentary living. With the possible exception of dental abscesses, height
represented by femur length is the most familiar to economists. On average femur length is about
one quarter of adult height, so that the difference in the femur length of male hunter-gathers and
town dwellers (445.8 – 430.5 = 15.3 mm) translates into 6.13 centimeters or 2.41 inches of height.
This is a substantial difference, which approximately equals the gain in average height in men the
U.S. from the mid-1700s to the present. The implied difference in height for women was even
larger, amounting to 8.36 centimeters or 3.29 inches.
Two of the additional markers, signs of anemia (technically, porotic hyperostosis and cribra
orbitalia) and linear enamel defects (hypoplasias) in the permanent canine teeth are formed in early
childhood but the lesions are generally preserved in adult skeletons. Anemia could have been caused
by a heavy parasite load or an iron-deficient diet, whereas the enamel dental defects are often
associated with general dietary stress and/or prolonged bouts of diarrhea during weaning. A third
9
See Clark Larsen’s book on Biolarchaeology for a general discussion of these markers (Larsen
1997).
44
marker, tibia infections, typically follow from abrasions to the shin, which allow Staphylococcus or
Streptococcus organisms to invade the periosteum, a sheath of tissue that nourishes the bone, and
eventually leave scars on the surface of the bone. These infections may persist for years and they
signal a weakened immune system.
Sample Issues. Although archaeologists have purposely excavated some famous sites (e.g.
Herculaneum and Pompeii) most skeletons available for study originate with building projects.
Occasionally entire cemeteries are excavated, but in general one can seldom argue that skeletons
proportionally represent an entire society. Many collections in Europe, for example, are
disproportionately from modern cities and towns, where more construction has occurred relative to
rural areas. Conditions that make for large urban areas today, such as low-cost transport, may have
been relevant in the past. Thus, in assessing health from skeletons it is important to consider
settlement size.
More to the point of this paper, one may ask whether modern excavations might be
connected with violence in the past. Speculation is difficult for the pre-Columbian period because
little is known about the geographic sites of violence, but sex ratios of the dead provide important
clues. If current building projects target growing urban areas and if geographic conditions persisted
in giving rise to large settlements, then a geographic bias could exist if past violence was greater in
urban as opposed to rural areas. Thus, the statistical analysis controls for rural-urban location.
Potentially worrisome is that some violence occurred on battlefields, which may be remote
from urban areas and less likely to have been excavated. Most famous battlefield sites such as
Gettysburg are specifically protected by law. Others, such as the Taunton site from the War of the
Roses, were deliberately excavated and reveal a gruesome pattern of frequent, multiple trauma that is
characteristic of many battlefield burials (Fiorato, et al. 2000). A great deal is already known about
45
battlefield violence since the Medieval era, but without excavation of these sites the evidence will be
found in the skeletal record only to the extent that the wounded survived to be buried in other
locations.
There are good reasons to separate the study of military from non-military violence. Not
only are the people who perished in military battles usually buried in separate locations but the
forces giving rise to conflict are often quite different. Men volunteered or were conscripted into
armies that fought on behalf of governments or political movements. In contrast, interpersonal
violence usually occurs on a much smaller scale in the area where the people live, and the victims
frequently know one another. Their disputes are tempered by social norms, sometimes stimulated by
alcohol and often accompanied other crime, anger, personal hatred or a desire to get even. The vast
majority of the sites in the Western Hemisphere database are non-military and the few sites
associated with battles are removed from the study.10 Even then it is impossible to distinguish
purely military from other wounds, but the remaining evidence most likely reflects domestic and
interpersonal violence.
Empirical Results on Violence: We measure violence as a dummy variable indicating
whether a skeleton had trauma to the head or a weapon wound, such as an arrow point or cut mark.
The explanatory variables are age at death, sex, elevation, settlement size, time period and
ethnicity.11 To provide a backdrop for study of violence, accidental trauma is examined separately,
as are children under age 15, who usually lack dimorphic traits for reliable sex determination.
Tables 3 – 5 present logit regression estimates for Native Americans. The first of these
tables shows that accidents were higher among males, whose occupations contributed to the
10
11
Also deleted are individuals of unknown sex, ambiguous ancestry, or undetermined age.
Dental development, the pattern of growth-plate fusion on long bones, dental wear and other
systematic skeletal changes are used to estimate age at death. Sexually dimorphic traits of the
pelvis and skull appear in adolescence.
46
problem, and at high elevations, where terrain was a factor. There was also an age-dependent
relationship that was the net result of two processes. Since the signs of skeletal trauma persisted
throughout life, the wounds accumulated with age. On the other hand, death eliminated a few
individuals prone to accidents early in their careers, leaving the more agile (and less injury ridden) to
survive to older ages. The selective editing of people by age was probably more potent with regard
to violence.
Individuals who died at ages 45+ were about 7 percentage points more likely to have an
accidental injury as someone who died at age 15-24. The dy/dx values should not be interpreted as
prevalence rates because the denominator is the number who died, not the number who were alive at
a particular age. The difference in the numbers between adjacent ages is suggestive and might be
usefully compared with the number at risk (which declines with age as a result of deaths) implied by
a plausible model life table, but any such calculations await careful evaluation of model
specification. Size of settlement and time period were not systematically related to accidental
trauma.
The results for violence contrast sharply with those for accidental trauma, containing both
expected outcomes and surprises. In the first category are higher rates for men relative to women
and the large jump in trauma in the post-Columbian period. 12 Although we cannot identify the
ethnic source of the injuries, European-Americans clearly initiated some acts of violence. But
contact was very destabilizing and plausibly set in motion a chain of events whereby native-onnative violence increased from new weapons, disease, new labor regimes, alcohol and competition
for resources.
12
In a separate examination of types of violent trauma, there was no systematic sex difference in
injuries to the face. Consistent with modern evidence, women were more likely to suffer facial
trauma relative to other types of intentional violence. Otherwise the results for types of violent
trauma mirrored those for the summary measure.
47
There was no statistically significant difference in coefficients between ages 15-19 and 20-24
(based on a regression, not shown). If we accept the results at face value, as opposed to imagining
they follow from some unknown process of selection, one must ask why. An additional regression
(not shown) for the age group 0 to 29, which includes five-year age categories as regressors,
indicates a very low coefficient for those under age 10 (only 0.9% of deaths at this age showed
violence); a jump to 4.7% in the ratio of violence among those who died at age 10-14; essentially no
change to age 20-24; and then a doubling of the ratio to a plateau at higher ages. These ratios
suggest high incidence rates at ages 10-14 and 25-29. Other regressions (not shown) indicate this
double-plateau pattern is confined to the pre-Columbian period; after Europeans arrived, the
probability that a person’s skeleton had evidence of violent injury was low and flat throughout
childhood years and did not increase dramatically until the early twenties. One might ponder
whether changing tactics of conflict or a change in weapons technology influenced the changing age
structure of violence.
The fundamental empirical result is clear: a relative lack of violence in villages and cities
compared to hunter-gatherer bands. The evidence suggests there was a remarkable degree of social
or cultural control over violence in villages and cities that somehow eluded small bands of huntergatherers. Sex ratios of deaths suggests the pattern was not the bi-product of missing violent deaths
of men who were buried at unexcavated locations; the proportion male among deaths was very
nearly the same across all settlement categories (about 47.5 per cent). Moreover, Table 5 shows that
violence among children, whose sex ratio was probably unaffected by selective migration because
they were unlikely to have left home before age 15, was also low in urban areas. The absence of
sugar and alcohol in the pre-Columbian world cannot explain the pattern because hunter-gatherers
also lacked these products. Possibly, inhabitants of urban areas were so poorly nourished that they
48
lacked the energy for violence, and ritualized violence may have diffused interpersonal aggression.
We emphasize the role of the natural state in controlling violence, but we recognize that worse
nutrition in the cities could have made the task easier.
The probability of a violent skeletal injury was nearly 6 percentage points higher among
those who lived at high elevations, a result that could have occurred for several reasons. In the
Western Hemisphere, upland areas tended to have lower food productivity and often fewer food
types. In contrast, coastal areas often had not only abundant sources of marine foods, which
provided a critical food (protein), but also many foods found in upland areas. The struggle for food
in upland areas may also have led to violence. The food problem was likely compounded by
isolation in valley settlements, a living arrangement that probably enhanced suspicion of outsiders.
The common presumption that an encounter with a stranger would be aggressive may have led to
pre-emptive violence as a form of self-defense.
European and African-American patterns of injury provide a backdrop against which to
evaluate results for Native Americans. The pattern of accidental trauma, shown in Table 6, is similar
to that for European and African-Americans, given in Table 3. For all ethnic groups, accidents were
higher among males and strongly related to age at death. In contrast with natives, however,
elevation was less of a factor in accidents; the coefficient on high elevation is positive but not
statistically significant. There was no systematic difference between blacks and whites in accidental
trauma.
Two similarities across ethnic groups in violent trauma, apparent by comparing results in
Table 7 with those in Table 4, are higher rates for men and for localities at high elevations.
European and African Americans also had an age dependent process, but the probability
unexpectedly declined somewhat at the oldest age group. Plausibly the available weapons were so
49
powerful (i.e. guns) that individuals prone to violence were killed at younger ages rather than
surviving to die at older ages with lesser violent injuries.
All blacks and whites in the non-native sample lived in towns or cities, limiting the
geographic separation available in the data. Even though the cities of the nineteenth century were
small by standards of the late twentieth century, the absence of a size-of-place gradient is surprising
relative to modern evidence. As discussed below, the explanation cannot be found in a relative
absence of men among the dead in cities. Consistent with modern data, however, blacks had
significantly higher rates of violent skeletal injuries than whites. Given the social and legal setting
in the South of the late nineteenth century, it is likely whites were the source of some, if not much
violence against blacks.
Table 8 distills information on relative levels of violence. The table presents probabilities for
men aged 25-34 calculated from parameter estimates in Tables 4 and 6. The range of calculated
probabilities of the skeleton having a violent injury at the time of death is considerable.
Surprisingly, both the lowest and highest rates occurred in urban areas, natives who lived in preColumbian cities versus blacks who lived in nineteenth century cities.13 If pre-Columbians were
violent, it was concentrated among the hunter-gatherers, whose level was nearly twice that of
European Americans. And the rate for the latter may be biased upward by an excess of men who
lived in the cities. As expected, trauma was reasonably high among village tribes in the early postColumbian period, and the somewhat low proportion of males among the dead suggests the level
may be underestimated by burial of men in other locations. The skeletal evidence is clear: the shift
from hunting-gathering societies to sedentary urban societies was accompanied by a marked
13
The calculations allocated 50 per cent of individuals to each of the early and late pre-
Columbian periods.
50
reduction in the level of human induced violence. Despite the poor health and lower physical
standard of living of most residents of villages and cities, the urban areas were significantly safer
places to live.
VII. Conclusions
The Neolithic revolution involved three changes in human society: an increase in sedentary
lifestyles; the domestication of plants and animals; and an increase in the size of social groups,
eventually leading to villages and cities. Economic explanations of the revolution typically focus on
the relative returns to hunting-gathering versus agriculture, making the perfectly reasonable
argument that an increase in the relative return to agriculture drove the shift to farming. But such
explanations leave questions about the growing scale of societies completely unanswered. Huntergatherers should have benefitted from increasing social scale as well, if through nothing more than
specialization and division of labor.14 Why didn’t larger groups form?
We build on the new empirical archeological and anthropological evidence by focusing on
the changing social structures necessary to support larger organizations of people. Huntinggathering societies were made up of many small bands of individuals, loosely grouped in tribes, with
violence both within and between bands and tribes. Violence posed two problems in the Neolithic
world. Sedentary farmers would be natural targets for nomadic hunters. Implementation of new
farming technology required better provision of defense, and better defense required greater numbers
and a larger social organization. Violence also limited the size of sustainable social organizations.
If bands of 25 persistently fought one another, how could a well-organized social unit of 100 or 1000
people be established unless violence was reduced and controlled?
14
Marceau and Myers, 2006, explicitly argue that a “grand band,” a coalition of small bands
could coordinate resource use and improve productivity. There explanation for the rise of
agriculture is the technological change that leads to the dissolution of the grand band, declining
returns to hunting and gathering, and a shift into agriculture.
51
Our explanation builds on a comparative static result. We do not specify the pathway by
which hunter-gatherer societies were transformed into larger sedentary urban societies. Instead we
develop a micro based individual model in which alliances between individual can form within and
between groups. The hierarchical connector model shows that building larger groups is unlikely in a
world of egalitarian groups. Larger alliances of groups are more likely to emerge around a smaller
group of “connectors,” the central individuals in non-egalitarian groups who expose themselves to
more violence. In the model with transfers, the connectors enjoy a higher material standard of
living, but experience more violence, while the other individuals have a lower material standard of
living, but higher utility because of the reduction in violence.
The logic of the connector model can be extended by to the idea of the natural state. In
natural states, by which military, economic, and political-religious elites create and enforce
exclusive privileges to valuable resources and economic functions. Limiting access to these rights
creates rents and the rents serve to bind the coalition together. Because the rights elites hold in land,
labor, capital, and economic activities are less valuable if violence breaks out and because the elite
coalition includes military elites who are the most dangerous members of the society, elite leaders
are able to make credible commitments not to fight each other. The commitments are fragile, since
they are only credible if a leader loses more from lost economic rents if he is violent than he might
gain from acquiring more resources and activities, so we expect that violence is reduced, but not
eliminated.
The idea that the Neolithic revolution resulted from the new social institutions created by the
natural state has two testable implications. First, the level of violence should be lower in villages
and cities than in nomadic bands. We describe the methodology developed to ascertain whether
skeletal evidence of trauma is caused by human violence and then apply that methodology to a
52
sample of 12,000 New World skeletons. The empirical results show that violence is significantly
higher in hunting-gathering populations than in villages and cities. The second is that the standard
of living of individuals living in large populations should be lower. Both implications find support
in the data.
Another set of implications concerns the structure of social organization in larger social units.
Although we do not test these implications directly, anthropologists have gathered evidence on the
changing social structure of societies as they increase in size, as well as on archaic states that first
arose in the past. As societies grow, they develop social structures that enable elites to generate
rents within the dominant coalition. Social hierarchies and economic inequality developed in
sedentary societies involving a social hierarchy where the leaders formed an interlocking set of
military, economic, political-religious elites. The specific institutional arrangements that structure
elite interaction varied from society to society, but they show the same overall pattern of
organization predicted by the natural state theory.
The emphasis on institutional development does not eclipse the importance of technological
development in agriculture, building construction, irrigation, or other innovations during the
Neolithic revolution. Technological development was fundamental to the Neolithic period and the
millennia that followed. We do not speculate on whether institutional change or technological
change came first. Unless better archeological data or techniques unexpectedly allow us to ask and
answer the causation question more directly, answers to the question of which came first will remain
speculations. Suffice it to say that both occurred simultaneously. But technological progress as a
driving explanation is incapable of dealing with the basic conundrum of the Neolithic period: how
were healthy hunter-gatherers convinced to live in villages and cities where their health was
measurably lower? The answer is that they were safer. Town dwellers suffered a third to a quarter
53
of the human induced violent trauma experienced by hunter-gathers.
The Neolithic revolution included a revolution in social organizations that produced status
driven hierarchies and a much more unequal distribution of economic resources and wealth. The
emergence of privileged status was a way to produce economic rents that could be used to solve the
problem of limiting violence. Safer environments and protection from human predation induced
larger populations to live in closer proximity, organized by economic and political-religious elites.
The first large human societies emerged in the form of the natural state.
54
Table 1: Distribution of Skeletons in the Database
Native American
Period
North America Middle Americ
South America Euro-American Afro-American
1750+
627
0
0
1,201
1,380
1500 – 1749
2,580
0
39
113
0
1000 – 1499
888
236
1,095
0
0
1 AD – 999
1,642
594
382
0
0
1000 BC – 0 AD
250
0
247
0
0
Before 1000 BC
485
343
418
0
0
6,472
1,173
2,181
1,314
1,380
Total
Source: Western Hemisphere database. Grand total = 12,520.
55
Table 2: Markers of Health by Type of Social Organization
Hunter-gatherers
Variable
Town or city dwellers
Mean
S.E
N
Mean
S.E.
N
Femur length, M (mm)
445.8
2.88
98
430.5
1.48
298
Femur length, F (mm)
422.0
2.39
141
401.1
1.23
351
Anemia (proportion)
0.190
0.0057
737
0.250
0.0062
925
Avg. # abscesses
0.545
0.039
1408
0.823
0.044
1448
Avg. # hypoplasias
0.238
0.022
433
0.544
0.028
688
Tibia infection severity
0.408
0.018
1446
0.510
0.025
1021
Source: Calculated for adults from the Western Hemisphere database. Signs of anemia, femur length
and dental hypoplasisas reflect health conditions in childhood. Tibia infections can occur at any age after
early childhood and abscesses were relatively more frequent among adults.
56
Table 3: Logit Regression Explaining Leg, Arm or Hand Trauma
in Adult Native Americans
Variable
Coeff.
z
P>|z|
dy/dx
male
0.3588
2.38
0.017
0.0222
age25-34
0.5865
2.36
0.018
0.0410
age35-44
0.5121
2.11
0.035
0.0338
age45+
0.9002
3.61
0.000
0.0708
high elev
0.3867
2.04
0.042
0.0234
village
0.1459
0.54
0.590
0.0087
city
0.1728
0.51
0.609
0.0111
late-pre
0.1627
0.83
0.405
0.0101
early-post
0.1641
0.73
0.462
0.0103
late-post
-0.3858
-1.25
0.211
-0.0207
constant
-3.7111
-11.49
0.000
N = 2,745; LR chi2(10) = 34.23; Prob > chi2 = 0.0002; Pseudo R2 = 0. 0.024
Omitted categories: female; ages 15-24; elevation under 300 meters; mobile groups;
early pre-Columbian (lived prior to 0 A.D.).
Source: Western Hemisphere database
57
Table 4: Logit Regression Explaining Head or Weapon Trauma in Adult Native Americans
Variable
Coeff.
z
P>|z|
dy/dx
male
0.2932
2.34
0.019
0.0202
age25-34
0.7709
3.81
0.000
0.0628
age35-44
0.6073
3.05
0.002
0.0453
age45+
0.7158
3.32
0.001
0.0596
high elev
0.8584
5.12
0.000
0.0589
village
-1.3031
-7.46
0.000
-0.1043
city
-1.6339
-5.83
0.000
-0.0720
late-pre
0.1674
0.92
0.358
0.0117
early-post
0.9963
4.82
0.000
0.0842
late-post
0.4426
2.07
0.039
0.0345
constant
-3.0145
-12.43
0.000
N = 3,431; LR chi2(10) = 102.51; Prob > chi2 = 0.0000; Pseudo R2 = 0.051
Omitted categories: female; ages 15-24; elevation under 300 meters; mobile groups;
early pre-Columbian (lived prior to 0 A.D.).
Source: Western Hemisphere database.
58
Table 5: Logit Regression Explaining Head or Weapon Trauma in Native American Children
Variable
Coeff.
z
P>|z|
dy/dx
age 5-9
0.4626
0.74
0.457
0.0029
age 10-14
1.4470
2.67
0.008
0.0139
high elev
-0.5034
-0.83
0.405
-0.0029
village
-2.8578
-3.95
0.000
-0.0388
city
-2.1687
-2.42
0.015
-0.0068
late-pre
0.8591
1.05
0.292
0.0052
early-post
2.7622
2.64
0.008
0.0411
late-post
1.5280
1.51
0.131
0.0135
constant
-4.1718
-5.13
0.000
N = 1,490; LR chi2(8) = 41.35; Prob > chi2 = 0.0000; Pseudo R2 = 0.180
Omitted categories: age 0-4; elevation under 300 meters; mobile groups;
early pre-Columbian (lived prior to 0 A.D.). Note: sex is unknown for children.
Source: Western Hemisphere database
59
Table 6: Logit Regression Explaining Leg, Arm or Hand Trauma
in Adult European and African Americans
Variable
Coeff.
z
P>|z|
dy/dx
male
0.7936
3.60
0.000
0.0664
black
-0.1514
-0.70
0.485
-0.0129
age25-34
-0.0948
-0.19
0.852
-0.0079
age35-44
1.0452
2.48
0.013
0.1034
age45
1.4302
3.39
0.001
0.1579
high elev
0.5250
1.08
0.281
0.0542
city
-0.6368
-1.85
0.065
-0.0673
constant
-2.7944
-5.20
0.000
N = 1,042; LR chi2(7) = 59.30; Prob > chi2 = 0.0000; Pseudo R2 = 0.079
Omitted categories: female; ages 15-24; white; elevation under 300 meters; rural.
Source: Western Hemisphere database
60
Table 7: Logit Regression Explaining Head or Weapon Trauma
in Adult European and African Americans
Variable
Coeff.
z
P>|z|
dy/dx
male
1.2957
4.25
0.000
0.0612
black
0.7142
2.45
0.014
0.0336
age25-34
1.1903
2.11
0.035
0.0768
age35-44
1.1943
2.18
0.029
0.0695
age45+
0.9144
1.59
0.112
0.0525
high elev
1.3846
2.90
0.004
0.1171
city
-0.0701
-0.14
0.888
-0.0034
constant
-4.9643
-6.38
0.000
N = 1085; LR chi2(10) = 37.98; Prob > chi2 = 0.0000; Pseudo R2 = 0.073
Omitted categories: female; ages 15-24; white; elevation under 300 meters; rural or village.
Source: Western Hemisphere database
61
Table 8: Calculated Probabilities of Violent Trauma among Men Aged 25-34
Group
Expected probability (%)
Per cent male
Sample size
13.39
47.13
715
Pre-Columbian, city
2.70
48.63
183
Early Post-Columbian, village
9.48
44.43
673
European-American, city
7.25
59.88
496
18.53
49.90
511
Pre-Columbian, hunter-gatherer
African-American, city
Source: Western Hemisphere database
62
References
Angel, J. L. "Patterns of Fractures from Neolithic to Modern Times." Anthropologiai
kozlemenyek 18, no. 1 (1974): 9-18.
Bakan, David. Slaughter of the Innocents. San Francisco: Jossey-Bass, 1971.
Baker, Matthew J. "An Equilibrium Conflict Model of Land Tenure in Hunter-Gatherer
Societies." Journal of Political Economy 111, no. 1 (2003): 124.
Baker, Susan P. The Injury Fact Book. Edited by Susan P. Baker. 2nd ed ed. New York: Oxford
University Press, 1992.
Bandy, Matthew S. “Fissioning, Scalar Stress, and Social Evolution in Early Village Societies.”
American Anthropologist, 106(2), pp. 322-333, 2004.
Bender, Barbara. "Gatherer-Hunter to Farmer: A Social Perspective." World Archaeology 10, no.
2 (1978): 204.
Binford, Lewis Roberts. In Pursuit of the Past: Decoding the Archaeological Record. Edited by
John F. Cherry and Robin Torrence. New York, N.Y.: Thames and Hudson, 1983.
———. "Post-Pleistocene Adaptations." In New Perspectives in Archeology., edited by Sally R.
Binford and Lewis Roberts Binford. Chicago: Aldine Publishing Co., 1970.
Bloch, Francis and Jackson, Matthew O. (2006) Definitions of equilibrium in network formation
games, Int J of Game Theory 34:305-318.
Boehm, Christopher. Heirarchy in the Forest: The Evolution of Egalitarian Behavior.
Cambridge: Harvard University Press, 1999.
Bollobás, Belá. (1978) Extremal Graph Theory, London: Academic Press.
Bollobás, Belá and de la Vega, W. F.. (1982) The diameter of random graphs, Combinatorica
2:125-134.
Boserup, Ester. The Conditions of Agricultural Progress. Chicago: Aldine, 1965.
———. Population and Technological Change: A Study of Long-Term Trends. Chicago:
University of Chicago Press, 1981.
Braidwood, Robert J. Prehistoric Investigations in Iraqi Kurdistan. Edited by Bruce Howe.
Chicago: University of Chicago Press, 1960.
Brunton, Ron. "Why Do the Trobriands Have Chiefs?" Man 10, no. 4 (1975): 544.
Cartmill, Matt. A View to a Death in the Morning: Hunting and Nature through History.
Cambridge, Mass.: Harvard University Press, 1993.
Childe, V. Gordon. Man Makes Himself. [Rev ed. New York: New American Library, 1951.
———. What Happened in History. Baltimore: Penguin Books, 1954.
Claessen, H. J. M., and Peter Skalník. The Early State. The Hague: Mouton, 1978.
Cohen, Mark Nathan. The Food Crisis in Prehistory: Overpopulation and the Origins of
Agriculture. New Haven [Conn.]: Yale University Press, 1977.
Conrad, Geoffrey W., and Arthur A. Demarest. Religion and Empire: The Dynamics of Aztec
and Inca Expansionism. New York: Cambridge University Press, 1984.
Dart, Raymond Arthur. "The Predatory Transition from Ape to Man." International
Anthropological and Linguistic Review 1 (1953): 201-18.
Demsetz, Harold. "Toward a Theory of Property Rights." The American Economic Review 57,
no. 2 (1967): 347.
Dickens, Charles. The Adventures of Oliver Twist. New York: Oxford University Press, 1966.
63
———. Great Expectations. New York: Harper & Row, 1965.
Donnelly, Anne Cohn, and Kim Oates. Classic Papers in Child Abuse. Edited by Anne Cohn
Donnelly and Kim Oates. Thousand Oaks, Calif.: Sage Publications, 2000.
Dunbar, Robin. Grooming, Gossip, and the Evolution of Language. Cambridge: Harvard
University Press, 1996.
Earle, Timothy. How Chiefs Come to Power. Stanford: Stanford University Press, 1997.
-----. Bronze Age Economics: The Beginnings of Political Economies. Boulder, CO: Westview
Press, 2003.
Engels, Friedrich. The Origin of the Family, Private Property and the State. New York:
Pathfinder Press, 1972.
Fiorato, Veronica, Anthea Boylston, and Christopher Knüsel. Blood Red Roses: The
Archaeology of a Mass Grave from the Battle of Towton Ad 1461. Oxford: Oxbow, 2000.
Flannery, Kent V. "The Origins of Agriculture." Annual Review of Anthropology 2, no. 1 (1973):
271-310.
Fox, James Alan. Uniform Crime Reports (United States): Supplementary Homicide Reports,
1976-1983. Edited by Glenn L. Pierce and Research Inter-university Consortium for
Political and Social. 1st ICPSR ed ed. Ann Arbor, Mich.: Inter-university Consortium for
Political and Social Research, 1987.
Fried, Morton H. The Evolution of Political Society: An Essay in Politician Anthropology. New
York: Random House, 1967.
Gebauer, Anne Birgitte, and T. Douglas Price. Transitions to Agriculture in Prehistory. Madison,
Wis.: Prehistory Press, 1992.
Glaeser, Edward and Bruce Sacerdote. "Why is there more Crime in Cities?" Journal of Political
Economy, 107, 6, pp. S225-S258, 1999.
Hayden, Brian. "Models of Domestication." In Transitions to Agriculture in Prehistory, edited
by Anne Birgitte Gebauer and T. Douglas Price. Madison: Prehistory Press, 1992.
———. "Nimrods, Piscators, Pluckers, and Planters: The Emergence of Food Production."
Journal of Anthropological Archaeology 9 (1990): 31-69.
Helfer, Mary Edna, Ruth S. Kempe, and Richard D. Krugman. The Battered Child. Edited by
Mary Edna Helfer, Ruth S. Kempe and Richard D. Krugman. 5th ed. rev. and expanded
ed. Chicago: University of Chicago Press, 1997.
Isaac, Erich. Geography of Domestication. Englewood Cliffs, N.J.: Prentice-Hall, 1970.
Jackson, Matthew O. (2008) Social and Economic Networks, Princeton: Princeton University
Press.
Johnson, Allen W. and Timothy Earle. The Evolution of Human Societies, 2nd edition. Stanford:
Stanford University Press, 2000.
Johnson, Gregory A. “Organizational Structure and Scalar Stress.” In Colin Renfrew, Michael J.
Rowlands, and
Barbara Abbott Segraves, Theory and Explanation in Archeology. New York: Academic
Press, 1982.
Kanarek, Robin B. "Nutrition an Violent Behavior." In Understanding and Preventing Violence,
Vol. 2, Behavioral Influences, edited by Albert J. Reiss and Jeffrey A. Roth, 515-39.
Washington, D.C.: National Academy Press, 1993.
Keeley, Lawrence H. War before Civilization. New York: Oxford University Press, 1996.
Kelly, Robert L. The Foraging Spectrum: Diversity in Hunter-Gatherer Lifeways. Washington:
Smithsonian Institution Press, 1995.
64
Kremer, Michael. "Population Growth and Technological Change: One Million B.C. To 1990."
The Quarterly Journal of Economics 108, no. 3 (1993): 681.
Larsen, Clark Spencer. Bioarchaeology: Interpreting Behavior from the Human Skeleton. New
York: Cambridge University Press, 1997.
LeBlanc, Steven A. Constant Battles: The Myth of the Peaceful, Noble Savage. Edited by
Katherine E. Register. 1st ed ed. New York: St. Martin's Press, 2003.
Locay, Luis. "From Hunting and Gathering to Agriculture." Economic Development and
Cultural Change 37, no. 4 (1989): 737.
Marceau, Nicolas and Gordon Meyers. "On the early Holocene: Foraging to Early Agriculture."
Economic Journal, 116 (July), 2006, 751-772.
Miczek, Klaus A., Joseph F. DeBold, Margaret Haney, Jennifer Tidey, Jeffrey Vivian, and Elise
M. Weerts. Alcohol, Drugs of Abuse, Aggression, and Violence. Edited by Albert J.
Reiss, Jeffrey A. Roth and Klaus A. Miczek. Vol. 3 v., Understanding and Preventing
Violence. Vol. 3, Social Influences. Washington, D.C.: National Academy Press, 1993.
North, Douglass C., and Robert Paul Thomas. "The First Economic Revolution." The Economic
History Review 30, no. 2 (1977): 229.
Ord, R. A., and R. M. Benian. "Baseball Bat Injuries to the Maxillofacial Region Caused by
Assault." Journal of Oral and Maxillofacial Surgery: Official Journal of the American
Association of Oral and Maxillofacial Surgeons 53, no. 5 (1995): 514-7.
Polanyi, Karl. The Great Transformation. New York: Octagon Books, 1975.
Rand, Michael R. Violence-Related Injuries Treated in Hospital Emergency Departments / by
Michael R. Rand with Statistical Support from Kevin Strom. Edited by Kevin Strom and
Statistics United States. Bureau of Justice. Washington, D.C.: U.S. Dept. of Justice,
Office of Justice Programs, Bureau of Justice Statistics, 1997.
Reiss, Albert J., and Jeffrey A. Roth. Understanding and Preventing Violence. Vol. 3, Social
Influences. Edited by National Research Council. Panel on the Understanding and
Control of Violent Behavior. 5 vols. Vol. 3. Washington, D.C.: National Academy Press,
1994.
———, eds. Understanding and Preventing Violence. Vol. 1. Washington, D.C.: National
Academy Press, 1993.
Robb, John. "Violence and Gender in Early Italy." In Troubled Times: Violence and Warfare in
the Past, edited by Debra L. Martin and David W. Frayer, 111-44. Australia: Gordon and
Breach, 1997.
Rowthorn, Robert and Paul Seabright. "Warfare and the Multiple Adoption of Agriculture ater
the Last Ice Age." Working Paper, November 6, 2008.
Sauer, Carl Ortwin. Agricultural Origins and Dispersals. New York: American Geographical
Society, 1952.
Service, Elman Rogers. Origins of the State and Civilization: The Process of Cultural Evolution.
1st ed ed. New York: Norton, 1975.
Shaw, Clifford Robe. Juvenile Delinquency and Urban Areas, a Study of Rates of Delinquents in
Relation to Differential Characteristics of Local Communities in American Cities. Edited
by Henry Donald McKay. Chicago: University of Chicago press, 1942.
Shepherd, J. P., J. J. Gayford, I. J. Leslie, and C. Scully. "Female Victims of Assault. A Study of
Hospital Attenders." Journal of Cranio-maxillo-facial Surgery: Official Publication of
the European Association for Cranio-Maxillo-Facial Surgery 16, no. 5 (1988): 233-7.
Shepherd, J. P., M. Shapland, N. X. Pearce, and C. Scully. "Pattern, Severity and Aetiology of
65
Injuries in Victims of Assault." Journal of the Royal Society of Medicine 83, no. 2
(1990): 75-8.
Simon, Julian Lincoln. The Economics of Population Growth. Princeton, N.J.: Princeton
University Press, 1977.
———. The Ultimate Resource. Princeton, N.J.: Princeton University Press, 1981.
Smith, Vernon L. "The Primitive Hunter Culture, Pleistocene Extinction, and the Rise of
Agriculture." The Journal of Political Economy 83, no. 4 (1975): 727.
Spitz, Werner U. Medicolegal Investigation of Death: Guidelines for the Application of
Pathology to Crime Investigation. Edited by Russell S. Fisher and Werner U. Spitz. 2d
ed ed. Springfield, Ill.: C.C. Thomas, 1980.
Steckel, Richard H., and Jerome Carl Rose, eds. The Backbone of History: Health and Nutrition
in the Western Hemisphere. Cambridge; New York: Cambridge University Press, 2002.
Walker, Phillip L. "A Bioarchaeological Perspective on the History of Violence." Annual Review
of Anthropology 30 (2001): 573-96.
———. "Wife Beating, Boxing, and Broken Noses: Skeletal Evidence for the Cultural
Patterning of Violence." In Troubled Times: Violence and Warfare in the Past, edited by
Debra L. Martin and David W. Frayer, 145-80. Australia: Gordon and Breach, 1997.
Wallis, John. “Institutions, Organizations, Impersonality, and Interests: The Dynamics of
Institutional Change.” Journal of Economic Behavior and Organization. Vol 79 (1-2), pp.
35-48, June 2011.
Weisdorf, Jacob. "From Foraging to Farming: Explaining the Neolithic Revolution." Journal of
Economic Surveys, 19, 4, (2005) 561-586.
----------. "Stone Age Economics: The Origins of Agriculture and the Emergence of Non-Food
Specialists." University of Copenhagen Discussion Paper no. 03-34, 2003.
Wittfogel, Karl August. Oriental Despotism; a Comparative Study of Total Power. New Haven:
Yale University Press, 1957.
Yoffee, Norman. Myths of the Archaic State: Evolution of the Earliest Cities, States and
Civilizations. New York: Cambridge University Press, 2005.