identity, interests, and conflict:

IDENTITY, INTERESTS, AND CONFLICT:AN EVOLUTIONARY MODEL

August 25, 2001

A. Maurits van der VeenChristopher H. Browne Center for International Politics

University of [email protected]

Abstract

The end of the Cold War has brought a renewed interest in the sources of international conflict. Increasingly, national and ethnic identities appear to be important factors, although their causal contribution to the outbreak of conflicts remains unclear. At the same time, the growing salience of globalization, represented by ever denser international economic interactions, also has ill-understood implications for international co-existence and conflict. The fact that these two factors — identity and economic interactions — inevitably affect one another in various ways only complicates the analytical challenge. In this paper, I turn to agent-based modeling as a promising source of insights. Computer simulations allow us to vary in a systematic fashion the different variables associated with the identities and economic interests of actors. The findings presented here show that economic interests and identities interact in interesting and sometimes unexpected ways, even when identity has no direct impact on the success or failure of economic interactions. For example, allowing successful economic interactions to affect an agent’s choice of identities, however marginally, has a noticeable effect on the degree of local cultural assimilation we observe. In addition, the variables determining the choice of new economic contacts also have indirect implications for the degree to which those contacts will connect agents with similar identities.

IDENTITY, INTEREST, AND CONFLICT:AN EVOLUTIONARY MODEL

The fundamental source of conflict… will not be primarily ideological or primarily economic… the dominating source of conflict will be cultural. — Huntington (1993:22)

The end of the Cold War has brought a renewed interest in sources of conflict other

than diametrically opposed ideologies. Indeed, there seems to have been a resurgence in

international conflicts of various kinds during the 1990s. Increasingly, national and

ethnic identities appear to be important factors in these conflicts. However, the causal

contribution of identity-related considerations to the outbreak of conflicts remains rather

unclear. At the same time, the growing salience of globalization, as represented by ever

denser international economic interactions, also has implications for international co-

existence and conflict that remain ill-understood. The fact that these two factors —

identity and economic interactions — inevitably affect one another in various ways, only

complicates the analytical challenge of understanding the connections between identity,

economic interests, and conflict.

Since the real world does not allow us to perform systematic experiments, we must

look elsewhere for answers. In this paper, I turn to agent-based modeling as a promising

source of insights. Computer simulations allow us to vary in a systematic fashion the

different variables associated with the identities and economic interests of actors. The

findings presented here show that economic interests and identities interact in interesting

and sometimes unexpected ways, even when identity has no direct impact on the success

or failure of economic interactions. For example, allowing successful economic

interactions to affect an agent’s choice of identities, however marginally, has

1

Identity, interests, and conflict 2

implications for the degree of local as well as system-wide cultural assimilation we

observe. In addition, the variables determining the choice of new economic contacts also

have indirect implications for the degree to which those contacts will connect agents with

similar identities. An further contribution of agent-based modeling is that it allows us to

gain an understanding of the different assumptions necessary to make various arguments

in the literature about the sources of conflict plausible. The more heroic these

assumptions appear, the less credible we must judge the associated arguments.

The remainder of the paper is divided into three sections. First, I briefly discuss the

motivations for studying the relative implications for international conflict of identity-

related considerations and economic ties. In the second section, I present the simulation

model used to study these factors. The third, and longest, section presents the results of

several sets of simulation experiments.

Motivation

What are the implications of globalization for future international conflicts?

Globalization has become one of the most salient concepts in international relations over

the past decade (e.g. Held et al. 1999, Friedman 2000, Kaplan & Wright 2001), yet

observers disagree strongly about both its nature and its implications. One of the most

important disagreements relates to the comparative salience of cultural or identity issues

on the one hand, and economic considerations on the other hand: in the words of Barber

(1992): Jihad vs. McWorld.

Perhaps the most visible exponent of an emphasis on cultural issues as determinants

of international conflict is Huntington, whose article (and subsequent book) on the ‘clash


of civilizations’ (Huntington 1993) received much attention in the popular as well as the

academic press. Huntington argued that “the great divisions among humankind and the

dominating source of conflict will be cultural” (Ibid. 22). This will be the case because

differences between civilizations are “basic”, i.e. “the product of centuries” (Ibid. 25),

and because they are less mutable than political or economic differences (Ibid. 27).

Moreover, these differences are increasing in salience as the world shrinks, i.e. as

contacts between civilizations become more common (Ibid. 25-26).

One can criticize Huntington’s argument on many grounds. As Welch points out, it

“clearly fails to meet any reasonable test of logic or evidence” (1997:211). Yet it

resonates with many people, and variations of his thesis continue to figure prominently in

debates on globalization and its implications. One of the reasons for the visibility of the

tribalism-leads-to-war approach is the emphasis of those who concentrate on economic

considerations on the universalizing nature of global transformations: growing

international economic ties are argued to reduce economic differences (Barber 1992).

However, closer ties may produce growing friction just as easily as they may result in

international agreement and harmony. One need only look at the literature on

international competitiveness (e.g. Krugman 1996) or last year’s ‘banana war’ between

the U.S. and its Central American clients on the one hand, and the European Union on

the other hand to be convinced of this fact.

Of course, these forms of economic conflict fall well short of war, which

differentiates them from armed ethnic conflicts. However, I am interested here primarily

in potential sources of tension rather than the actual outbreak of armed conflicts.

Modeling the latter requires further extensions to our model, and it is worthwhile to gain

a better understanding of the interaction of different sources of tension first. I would


argue different assumptions regarding two important issues constitute the major sources

of disagreements about the implications of globalization for international conflict. The

first issue is that of the relative stickiness of economic and cultural ties. Huntington

emphasizes the fact that cultural features were less mutable than economic or political

ones. In this respect, he aligns himself with primordialists, who see culture and identity

as more or less constant (if not even fixed at birth). However, increasing empirical

evidence demonstrates that cultural and identity-related features do in fact change over

time, and can to some degree be chosen by individuals (*** get citations ***). If identity

is indeed relatively mutable, much of the force of Huntington’s argument disappears.

The second issue is that of the interaction between economic and cultural ties.

Huntington is rather vague on this issue. On the one hand, he argues that the world is

shrinking, because ties across cultural boundaries are becoming more common. On the

other hand, he emphasizes the growing degree of economic integration within cultural

boundaries. One would expect these two trends ought to have opposite implications in his

argument. If closer economic ties increase friction, then they should do so both within

and across cultural boundaries. If they help unify a region, on the other hand, this ought

to have implications across cultural boundaries too (see Barber 1992). In order to argue

that both trends increase the likelihood of conflict, Huntington has to assume that

economic interactions within a civilization give rise to conflict at a much lower rate than

do those across civilizations. This might be true if shared cultural markers significantly

increase the likelihood of mutual cooperation between economic actors, but whether or

not such is the case is open to question.

Both of these central issues require much additional empirical research to be

answered satisfactorily. In the meantime, however, we can learn a lot by examining the


relative impact on international tensions and conflict of different assumptions about the

nature of economic interactions and the evolution of identity. The best way to do this is

by means of computer simulation, in which we can systematically vary the constitutive

assumptions of the system (Axtell 2000). The model and the results introduced below

represent an initial investigation of the interaction of different variables associated with

economic and cultural contacts within and across borders. As such, they will not be able

to supply us with any definite answers about the relationship between either of these

factors and international conflict. However, they can shed some light on the ways in

which economic interests and cultural or ethnic identity interact. Although actual conflict

is not present in the model, changes in economic ties as well as in the identities of agents

tell us something about the most likely sources of tension and change in the system.

In this way, we can begin to develop a sense of the assumptions that would be

required to validate different takes on globalization. For example: what would we have to

assume about the relative stickiness of economic and civilizational connections for

Huntington’s argument to be at least vaguely plausible? And what happens to his

argument if we allow one to affect the other (e.g. civilizational connections shape

economic ties, or economic ties affect the identities of agents)? These are the kinds of

questions that will be addressed below.

Model

Our model is composed of a world with 400 inhabitants. These are located on a

square grid which wraps around in both directions (i.e. a torus). Each agent has a

repertoire of different identities, which together compose its overall identity. A total of


20 different identities are present in the world at the start, and each agent’s repertoire

contains eight of these. The contents of an agent’s identity repertoire are mutable, but

they have a certain stickiness, in line with empirical evidence about people’s identities.

The evolution of identity repertoires is modeled using the framework of Lustick’s Agent-

Based Identity Repertoire model (see Lustick 2000). This model derives its inspiration

from the constructivist literature, where new identities are added to the repertoire, and

old ones dropped, in response to the identities each agent sees around it in the population.

The model stipulates that only one identity in a person’s repertoire can be active (and

thus visible to others) at any one time. Identities that are not actively expressed by

anyone in an agent’s neighbourhood may well disappear from that agent’s repertoire in

the very long run. Conversely, identities that are popular are likely to be added and, if

active in enough neighbours, to be activated too. In addition, a ‘bias’ is associated with

every identity. This bias represents an environmental signal about the popularity of the

particular identity. It can be interpreted as indicating how favourable the structural

context is towards the active expression of different identities. Biases change over time,

randomly and independently from one another, but they change rather slowly — once

every 100 timesteps on average in our experiments here. The bias values vary from –2 to

+1 in most of our experiments reported here.1

The system is run for a number of rounds. Each round, every agent combines

information about its own identity repertoire with information about the active identities

in its Moore neighbourhood (the eight surrounding agents), and the biases provided by

the global environment. Every potential identity is assigned a weight. This weight will

vary from one agent to the next, as it depends on each agent’s immediate neighbourhood.

1 For more details on the identity repertoires and the process of updating them, see Lustick (2000) and van der Veen, Lustick, and Miodownik (2001).


All agents have three threshold levels that specify how to process this relative ranking of

identities. The first gives the threshold above which an identity that is already in an

agent’s subscription list will become the activated identity in the next round. Thus, if an

identity has a weight that exceeds that of the agent’s own active identity by more than the

threshold, the agent will activate on this new identity instead. The second, higher,

threshold indicates when an agent will add a new identity to its subscription list (often

preparing it for activation in a subsequent round). The third, and highest, value tells the

agent when to subscribe to a new identity and simultaneously activate it.2

Most agents’ identity repertoires are initialized randomly, with each identity equally

likely to form part of an agent’s repertoire or to become that agent’s activated identity.

However, we divide the world into four quadrants. In each of these, a national identity is

the initially active identity in nearly one quarter of the agents, whereas the other three

national identities are under-represented.3 This makes it more likely for these national

identities to come to dominate within their respective quadrants. However, other

outcomes are possible — for example, national identities may quickly lose out to another

identity that is present throughout the world, a ‘McWorld’ identity, for instance.

Agents take most of their identity cues from their immediate geographical

neighbours, the Moore neighbourhood. In addition, they also have economic contacts,

and these may be situated anywhere throughout the world, both within their own national

quadrant and ‘abroad.’ Each agent has the same number of economic contacts as cultural

2 Agents perform these calculations in asynchronous, random order. In other words, each round the agents update their identities sequentially in a random order that varies from round to round. This prevents any problems from occurring as a result of synchronous updating (see Huberman & Glance 1993).

3 To be precise, 20% of the population in a quadrant is activated on the national identity and has none of the other three national identities anywhere in its repertoire. The remaining 80% is initialized randomly. Given the fact that we have 20 possible identities, this means that an additional 4%, on average, will also be activated on the national identity.


neighbours, i.e. eight. Associated with each economic contact is the value of that contact

to the agent. When this value falls below a certain threshold, the agent is likely to sever

the economic contact and look around for a new agent to interact with economically. The

value of a contact is determined by the history of interactions with the other agent. Each

round, an agent decides whether to cooperate or defect in its economic interaction with a

contact. The probability of cooperating is proportional to the value of the contact. If the

contact cooperates, the contact’s value will rise; if it defects, the value will fall.

Clearly, an economic contact with a very high value is likely to continue over a long

period of time, even in the face of occasional defections by the other agent. Just how

‘sticky’ such an economic connection is (compared to an agent’s identity repertoire, for

example), is determined by the size of the adjustment to its value upon cooperation or

defection. The adjustment is expressed in terms of a fraction of the maximum possible

increment or decrement. For example, if a contact’s value is 0.5 and the adjustment

factor is 10%, then the other agent’s cooperation will result in an increase in value by

0.05 (10% of the maximum possible increase of 0.5). On the other hand, if the value is

already 0.9, then cooperation will result in an increase of just 0.01 (10% of the possible

0.1). By increasing or decreasing the adjustment factor, we can affect the relative

stickiness of economic ties.

In addition, it is possible for contacts that have a very high value to have an impact

on an agent’s identity. In effect, these contacts join the ‘peer group’ that an agent uses in

deciding whether to update its identity repertoire. In the model, we can vary the relative

importance of economic contacts compared to the agent’s Moore neighbourhood. This

variable allows us to examine the potential homogenizing effect of globalization: if an


agent develops strong contacts with a number of agents elsewhere in the world,

eventually it is likely to adopt some of their cultural markers.

A final feature of the model to discuss is the process of replacing economic contacts.

All contacts can be severed by either agent. Both agents are then placed on a list of

agents seeking new contacts. At the end of the round, agents on this list are paired up.

However, they are not always paired up randomly. Instead, we pick one agent at random

from the list, and then pair it with the agent remaining on the list that is most appealing

to it. The appeal of a potential contact derives from its proximity either to the agent itself

or to an existing contact. Empirically, we might expect agents to prefer proximity to

themselves because such contacts are likely to be culturally similar to them, and might

thus be more likely to cooperate in economic interactions.4 Proximity to an existing

contact may be preferred in the hope that the expected cultural similarity of such

candidates to the existing contact will make them similarly valuable as contacts.

Importantly, in the results presented here an agent’s identity repertoire does not actually

have any implications for its likelihood of cooperating. Nor does the degree of its cultural

similarity to that of its economic contacts. However, it will be interesting to see whether

the preferences in the selection of new contacts described here nevertheless lead to a

clustering of economic contacts or perhaps a gradual reorientation of contacts towards

interaction with one’s close neighbours — the opposite of globalization, in a sense.

Results

We have now seen enough of the model set-up to be able to discuss some results.

Table 1 shows the default values for a number of different variables in the model. Unless

4 As noted above, Huntington appears to make this assumption.


otherwise noted, these are the values used in the simulations presented below. As the

table shows, each simulation was run for a total of 500 rounds, and each specification

was run 10 times. The results presented below are averages over these 10 runs.

Variable Value Variable Value

Number of rounds 500 Number of replications 10

Initial value of contact 0.5 Adjustment factor 20%

Include in identity calculations at value

0.8 Weight in identity calculations 0.2

Consider severing contact at value 0.2

Over-weight proximity to self in contact replacement by factor

2

TABLE 1. Default parameter values.

Before we begin discussing the implications of varying the different variables listed

in table 1, it is worth briefly describing some general trends in each simulation run. As

one would expect, the four national identities each expand their sway over their

respective quadrant, occupying about 3/4 of the quadrant on average by the end of the

run. The other identities remain common within repertoires, but only are only

represented at the activated level in about 5 agents on average at the end of a run. About

one in three identities disappears from agent activations (but not from the subscriptions)

altogether in any given run of the system. Nonetheless, a few agents continue to change

their repertoires by adding new identities or activating on identities already in their

repertoire through the end of every run. In other words, the system does not reach a

steady state. Overall, the average identity-related ‘tension’ an agent is exposed to, as

measured by the number of agents whose activated identity differs from its own, also

declines steadily, from about 6 at the start to just over 2 at the end of the average run.


The average value of economic contacts gradually rises to 1. After 200 rounds it

crosses 0.95. This pattern is not surprising, as ‘bad’ contacts are eliminated whereas the

good ones are allowed to become stronger and stronger. In fact, after 300 rounds less

than one contact is severed on average every round.5 The average distance from an agent

to its economic contacts shrinks, but not as much as one might expect in light of the

preference for new contacts that are closer by. From an initial average value of 6.7, it

declines below 5.2 after about 50 rounds, and then climbs slightly to hold steady around

5.25.6 The preference for new contacts close to existing contacts has even less of a

noticeable impact, as the average distance among an agent’s contacts only falls from 6.7

to 6.4. The average similarity between an agent and its economic contacts does rise

noticeably, however. On average, an agent shares its activated identity with less than half

a contact at the start of a run. After 60 rounds, this figure has risen above 2, and after 500

rounds it has risen further to an average of 2.9.7

In the real world, it is not always clear whether another party has been cooperative in

an economic interaction. Other factors may intervene to make the other party’s behaviour

somewhat uncertain. In order to model this in our system, I introduced some noise. The

noise-factor was 5%, which implied that there was a 5% probability than an agent would

mis-interpret cooperation by its contact as being defection (and vice versa). Every

experiment discussed below was run both with and without this noise factor. When we

introduce noise, the basic patterns discussed above are radically altered. Not surprisingly,

5 Note also that if only a single contact is severed during a round, the agents have no choice but to re-start their contact, as no other agents have made themselves available. This has the effect, therefore, of re-initializing the contact’s value to 0.5.

6 Agents have more choice in terms of picking contacts early in the run, when many contacts are severed every round. The fewer contacts are severed in a round, the less choice, and therefore average distance to a contact may rise again.

7 Overall repertoires show rather little convergence. The fraction of identities that is shared rises from an average of 0.4 to 0.41 over the course of 500 rounds.


it becomes more common for agents to sever economic contacts, even during the later

rounds of a simulation. Indeed, the average number of severed contacts remains fairly

steady throughout, at about 16 connections per round. Moreover, the average value of

each contact is never able to climb much, rising from 0.5 to 0.51 quite rapidly but failing

to move beyond that.

One interesting side effect of introducing noise is that the continual severing and

renewing of contacts leads to a gradual relocation of contacts closer to each agent, given

the fact that local contacts are preferred. As a result, the average distance of each contact,

which was about 5.25 without noise, drops to 4.7 with noise. Along the same lines, the

number of contacts that share the agent’s active identity is 3.1 by the end, compared to

2.9 without noise. Overall it is quite striking how great an impact is associated with a

mere 1 in 20 chance of misinterpreting an agent’s behaviour in an economic interaction.

In the remainder of this section I present the results of three sets of experiments. In

the first, I examine the implications of increasing or decreasing the stickiness of

economic contacts. The second set of experiments considers the effect of adjusting the

degree to which agents prefer to establish new contacts close to themselves or to existing

contacts. The third, finally, studies the effect on the evolution of identity repertoires of

allowing contacts to be included in identity calculations and of making agents more

resistant to changing their identity repertoires.

The stickiness of economic contacts

With the default adjustment factor of 20%, we saw that the average value of an

agent’s contacts rises fairly rapidly, hitting 0.95 by 200 rounds. If we reduce the


adjustment factor, it ought to take longer for values to reach that level. On the other

hand, it will also make connections more stable, and thus should reduce the number of

severed contacts in each round. Conversely, if we boost the adjustment factor above

20%, we should see large numbers of severed contacts, even later in a run, as it takes

fewer defections by a contact for an agent to consider rejecting that contact altogether.

Round Variable 10% 20% 30% 40%

50 Avg. # replacements

Avg. value

8.5

0.56

17.6

0.73

17.4

0.86

13.2

0.95


Avg. value

6.4

0.63

8.9

0.85

4.5

0.96

1.3

0.99


Avg. value

3.7

0.77

1.4

0.98

0.1

1

0

1


Avg. value

1.7

0.90

0

1

0

1

0

1

TABLE 2. Changing the value adjustment factor.

In order to test the impact of the adjustment factor, I ran experiments where this

variable was assigned the values of 10%, 30%, and 40%, in addition to the default

experiment with a value of 20%. The results are shown in table 2. As one would predict,

reducing the adjustment factor reduces the number of severed contacts in the early

rounds, but it also means that contact values increase more slowly, increasing the number

of severed contacts later on. When one increases the adjustment factor, we see large

numbers of severed contacts in the very early rounds, but by round 50, the frequency of

broken contacts has fallen below that for the default adjustment factor of 20%. Moreover,

the large adjustments bring agents so close to full cooperation so rapidly that

replacements disappear entirely fairly early on.


The effects of introducing noise into the system are very similar to those discussed

above for the default 20% adjustment factor: average contact values never increase much,

and the number replaced connections remains pretty much constant throughout a run. It is

worth noting, however, that the number of replaced connections grows dramatically as

the adjustment factor rises. At 10%, the average number of replacements is just 1; at

20%, we saw that it hovered around 16. For an adjustment factor of 30%, however, the

average figure rises to about 52, and at 40% it reaches 110! The dramatic increase in the

number of replaced connections has only a very marginal effect on average contact value:

an adjustment factor of 10% means that average value remains stuck at 0.5; with 20% it

rises to 0.51, with 30% to 0.52, and with 40% just to 0.53.

Adjustment

factor

Avg. distance

to contacts

Avg. # contacts

w. same active ID

Avg. tension

level

10% 5.63 2.73 1.93

20% 5.26 2.85 2.05

30% 4.88 3.10 2.12

40% 4.57 3.32 2.03

TABLE 3. Variable preference for proximity of new contacts.

Of more immediate interest to questions of tensions and the evolution of identity are

data on trends in the number of economic contacts activated on the same identity as

oneself, and on the average tension levels throughout the system. Here we see a definite

impact for the different adjustment factors. The greater the adjustment factor, the smaller

the average distance to one’s contacts by the end of the run, as table 3 shows.8 Not

8 Note that since most contacts are replaced very early on as the adjustment factor rises, the average distance drops quite rapidly and then remains steady for the remainder of the run.


surprisingly, the average number of contacts with which an agent shares the active

identity increases accordingly as we move down the table.

The effects on overall tension levels are less pronounced. Tensions rise at first as the

adjustment factor increases, but drop slightly at the top level. This is likely to be a result

of the inclusion of economic contacts in decisions about updating one’s identity. As the

adjustment factor rises, fewer and fewer contacts are replaced later in the run, and as a

result they will have less impact on the continuing evolution of identities. On the other

hand, the large number of initially replaced contacts can be expected to have a noticeable

impact on the speed of the initial assimilation of identities. In fact, tension levels drop

more rapidly for the higher adjustment factors. Somewhere between 20% and 30%, the

first trend gains the upper hand over the second trend, as reflected in the table.

The implications of preferences in new contacts

In the first part of this section we saw that the preference for new contacts close to

oneself or close to an existing contact appeared to have relatively little impact on the

average distance between an agent and its contacts, or among those contacts. It could be,

of course, that this is a result of the choice of overweighting factor used. In order to test

that possibility, I re-ran the simulation with an overweighting of 0 (no locational

preferences at all), 1 (equal weighting for proximity to self and proximity to an existing

contact) and 3 (proximity to self weighs 3 times as heavily). Table 4, showing the results

of this experiment after 500 rounds, clearly indicates that preferences for contacts in

one’s own proximity can have a considerable impact on the localization of those

contacts, as well as on the degree to which they end up being culturally similar.


Overweight

factor

Avg. distance

to contacts

Avg. # contacts

w. same active ID

Avg. tension

level

0 6.70 1.63 2.24

1 6.05 2.25 2.04

2 (default) 5.26 2.85 2.05

3 4.59 3.64 1.79

TABLE 4. Variable preference for proximity of new contacts.

When agents have no preferences at all regarding the location of their contacts, the

average distance of those contacts remains steady at the initial value. As soon as a

preference is introduced, the average distance begins to fall. Moreover, a preference for

contacts located in an agent’s own proximity also has the attendant effect of increasing

the cultural similarity between an agent and its economic contacts, as the third column in

table 4 shows. Moreover, through the indirect effect of strong economic contacts entering

into identity calculations, the overall tension levels in the population also fall as over-

weighting increases.9 However, the decline in tension levels is much less salient than the

observed increase in the number of contacts with the same active identity.

When we introduce noise into the simulation, the increased number of severed

connections has predictable effects. For overweighting factors 0 and 1, the average

distance to an agent’s contacts remains nearly steady (6.70 and 6.65, respectively), and

the values for tension level and the number of economic contacts with the same active

identity are very similar to those shown in table 3. However, once we emphasize

proximity to the agent by raising the overweighting factor over 1, the changes become

very evident. At a value of 2, the average distance to one’s contacts falls to 4.72

9 Remember that tension levels indicate how many neighbouring agents are activated on a different identity than that of an agent itself. They can vary from 0 to 8.


(compared to 5.26 without noise) and at a value of 3, it drops further to 3.35 (compared

to 4.59). Not surprisingly, the degree of cultural similarity to one’s contacts is

accordingly stronger: the number of contacts with the same active identity rises to 3.15

(over-weight 2) and 4.12 (over-weight 3), whereas the average tension levels throughout

the system is the same as in the no-noise case (over-weight 2) or even rises slightly

compared to that situation (over-weight 3). This underscores the fact that introducing a

preference for local contacts has a much stronger effect on local cultural homogenization

than it does on the overall tension levels throughout the system.

The stickiness of identity repertoires

In the default simulations discussed above, strong economic contacts were allowed to

affect an agent’s identity calculations. The weight of an economic contact was set to

equal one fifth of that of an immediate neighbour. In the experiments reported here, I

changed that value to 0.1, 0.3, and 0.4, to see whether that would have an impact on the

degree of cultural convergence between agents and those with whom they interact

economically. As one might expect, larger contact weights result in a greater number of

contacts with which an agent shares the activated identity. Similarly, they generate an

overall context within which agents are surrounded by fewer neighbours activated on

other identities, on average. The exact values after 500 rounds are shown in the second

and third columns of table 5. However, the differences are quite small. It seems plausible

that this finding results from the particular value of the identity thresholds of the agents.

If a particular identity weight is right at a threshold, then even an additional weight of 0.1

will suffice to push it over that threshold. Adding 0.4 instead will make no difference. It


seems therefore, that a large amount of the impact of the economic contacts on identity

convergence takes place close to this threshold.

Contact

weight

Std. thresholds

Cont. w. same ID

Std. thresholds

Avg. tension

High thresholds

Cont. w. same ID

High thresholds

Avg. tension

0.2 2.85 2.05 2.27 2.93

0.3 2.97 1.91 2.77 2.63

0.4 3.05 1.86 3.41 2.11

TABLE 5. Variable contact weights, with two different sets of identity update thresholds.

In every simulation discussed until now, the three identity thresholds of the agents

were at 1, 4, and 6. In other words, an agent was willing to activate on an identity already

in its repertoire when the weight of that identity exceeded that of its currently active

identity by more than 1; it was willing to add a new identity to the repertoire if that

identity’s weight exceeded the weight of the agent’s active identity by more than 4, and it

would immediately activate that identity if its weight exceeded that of the active identity

by more than 6. We noted above that the overall repertoires of agents did not converge

much. Most of the changes take place at the activation level, i.e. the lowest threshold

value. Given the difference between it and the next threshold (from 1 to 4), it makes

good sense that increasing the weight of valuable contacts had relatively little impact. It

will be interesting to see, then, what happens when we shift the thresholds to make an

agent’s identity repertoire stickier. The last set of experiments examined the implications

of changing the thresholds to 2, 5, and 7, respectively, and once again varying the weight

of valuable economic contacts in the identity calculation. The fourth and fifth column in

table 5 display the results after 500 rounds.


As expected, the differences between the rows now are considerably larger.

Moreover, when we look at the number of contacts activated on the same identity as an

agent, we see that this number is smaller when contact weights are 0.2, and larger when

contact weights are 0.4 when we compare standard and high thresholds. This strongly

suggests that our interpretation of the relatively small effects under standard thresholds

was correct. It is also interesting to note that average tension levels throughout the

population remain higher when identity repertoires are stickier, even when the average

number of contacts with the same active identity increases. This indicates that we see

more of a clustering effect, with economic connections within culturally similar clusters,

but relatively little system-wide convergence in identities.

Finally, the introduction of noise into this set experiments suffices to eliminate any

noticeable effect from changing the contact weight. This is not surprising, since contacts

only have an impact when their value is over the threshold of 0.8 specified in table 1, and

as we have seen, introducing noise tends to result in severed connections well before

contact values ever reach that level.

Conclusion

The results presented here provide a preliminary impression of the ways in which

economic connections and the evolution of identities in a system can interact. We have

seen how the stickiness of economic contacts can have a considerable impact on the

average distance between an agent and its economic partners, as well as on the degree to

which an agent shares its active identity with its contacts, whereas the impact of this

variable on overall tension levels is less straightforward. Placing a high emphasis on the


proximity of one’s contacts has similar implications for the number of contacts with

which the active identity is shared, but it also has a clearer impact on tension levels: a

greater emphasis on local contacts also contributes to lowering tension levels system-

wide. Finally, the weight of economic contacts in identity calculations has interesting

effects on the stickiness of identity repertoires. Generally speaking, increasing contact

weights will raise the number of contacts with which the active identity is shared at the

end of a run, as well as lower overall tension. However, the size of the effect depends

strikingly on the intrinsic stickiness of identities: increasing agents’ resistance to

updating their identity repertoires simultaneously made the effect of changing contact

weights much more pronounced.

I argued at the beginning of the paper that simulations could tell us something about

the assumptions required for different arguments to be plausible or valid. For example,

the simulation that most closely resembles the kind of world Huntington imagines in his

“clash of civilizations” would probably be one with very high intrinsic stickiness of

identities, as discussed immediately above. At the same time, his model assumes fairly

high tension levels and low levels of economic conflict. If we take the number of

replaced contacts as a measure of economic conflict, we see that in his world it would be

essential that all economic interactions were perfectly straightforward, since the

introduction of noise dramatically increases the number of contacts that are severed.

Moreover, it appears that Huntington implicitly assumes high preferences for local

contacts, because those increase the tendency to trade with similar agents. Of course, as

we have seen here, this implies noticeable reductions in the degree to which economic

contacts span the system, and is therefore difficult to reconcile with the realities of

globalization.


The experiments presented here are but an initial attempt at investigating the

interaction between economic ties and the evolution of identities. Numerous additional

experiments and extensions to the model readily suggest themselves if we wish to

improve our understanding of the interaction between the different variables. First, it

would be worthwhile to compare the effects of preferring local contacts to those of

preferring contacts with the same active identity. It might be, for example, that the latter

produces pockets of cultural similarity that are not geographically contiguous, generating

systems that resemble situations where trade happens between emigrant communities or

trading posts and a larger home bases.

Second, it will be useful to model globalization more directly, by introducing borders

of greater or lesser porosity into the system. The model presented here had no physical

borders, although implicit identity borders emerged as a result of the prevalence of

national identities in different quadrants. One important feature of globalization,

according to the literature, is that physical borders are eroding, facilitating contacts

between previously separate states. It would not be difficult to set up our model to run

for a number of rounds with the quadrants separated, and then gradually increase

economic contacts across borders, as well as phase out the borders themselves, allowing

identities to come into direct contact.

Next, it is important to operationalize the concept of conflict. In this paper, we were

most interested in tensions and the rate at which economic contacts were severed.

However, neither necessarily leads to direct (armed) conflict. How can we model

conflict, and how should economic and identity-related tensions differ in the likelihood

that they lead to conflict? Moreover, how should we operationalize the implications of

such conflicts for the evolution of identities as well as economic contacts? These are


crucial questions to address as we increase the sophistication of the model. However, we

cannot proceed in this direction until we have a full understanding of the interaction of

the different factors in the absence of outright conflict; adding too many new variables at

once would rapidly make it impossible to investigate the net impact of individual features

of the model.

Finally, it is worth noting that the basic set-up of our model is not relevant only to

conflicts. It is also promising as a tool for investigating more generally the evolution of

identities in the context of different types of economic contacts. For example, there is a

growing interest in the impact of European integration on the identities of European

populations. Our model would allow us to separate out the effects of increased economic

interactions within Europe and of more direct contacts as a result of the phasing out of

physical borders. Moreover, the wealth of public opinion data available in the European

context about the connection between international contacts and changing identities can

provide a valuable source of information about the appropriate values of the different

variables in the model, such as the weights of economic contacts in the evolution of

identities, and the tendency to prefer local over more distant contacts.


References

Axtell, Robert. 2000. “Why Agents? On the Varied Motivations for Agent Computing in

the Social Sciences.” Washington: Brookings Institution, Center on Social and

Economic Dynamics, Working Paper No. 17.

Barber, Benjamin R. 1992. “Jihad vs. McWorld.” The Atlantic Monthly (March), pp. 53-

63.

Friedman, Thomas L. 2000. The Lexus and the Olive Tree.

Held, David, Anthony McGrew, David Goldblatt, and Jonathan Perraton. 1999. Global

Transformations: Politics, Economics and Culture. Stanford: Stanford University

Press.

Huberman, Bernardo A., and Natalie S. Glance. 1993. “Evolutionary Games and

Computer Simulations. Proceedings of the National Academy of Sciences of the

United States of America. 90(16)7716-7718.

Huntington, Samuel. 1993. “The Clash of Civilizations.” Foreign Affairs, 72(3):22-49.

Kaplan, Robert D., and Robert Wright. 2001. “Mr. Order Meets Mr. Chaos.” Foreign

Policy, May/June.

Krugman, Paul. 1996. Pop Internationalism. Cambridge: MIT Press.

Lustick, I. S. (2000). “Agent-based modeling of collective identity: Testing constructivist

theory,” Journal of Artificial Societies and Social Simulations, 3(1). Available at

http://jasss.soc.surrey.ac.uk/3/1/1.html

van der Veen, A. Maurits, Ian S. Lustick, and Dan Miodownik. 2001. “Studying

Performance and Learning with ABIR: The Effects of Knowledge, Mobilizing

Agents, and Predictability.” Social Science Computer Review, 19(3).

Welch, David A. 1997. “The ‘Clash of Civilizations’ Thesis as an Argument and as a

Phenomenon.” Security Studies, 6(4):197-216.

http://jasss.soc.surrey.ac.uk/3/1/1.html


Model notes

The simulation was programmed in Macintosh Common Lisp 4.3, on a Macintosh

Powerbook G3 running System 9.1. Code is available from the author upon request.

identity, interests, and conflict:

Documents