A New Kind of Social Science: The Path Beyond Current (IR) Methodologies May Lie Beneath Them

A New Kind of Social Science:
The Path Beyond Current (IR) Methodologies May Lie Beneath Them
by
Valerie M. Hudson, Brigham Young University
Philip A. Schrodt, University of Kansas
Ray D. Whitmer, Jhax.Ltd.
Version 0.8B2
Last Update: 26 March 2004
Paper prepared for delivery at the Annual Meeting of the International Studies Association,
Montreal, Quebec, Canada, March 2004.
Authors can be contacted via email at the following addresses:
[email protected], [email protected], and [email protected].
This research was supported in part grants from the College of Family, Home, and Social
Sciences, Brigham Young University, the U.S. National Science Foundation grant SES-0096086,
and the General Research Fund of the University of Kansas.
The analytical web site for this project, which includes the graphic tools for analyzing event
patterns is http://ep.jhax.org. The data sets discussed in this paper, as well as a pdf
version of the paper with color graphics, can be downloaded from the KEDS project web site:
http://www.ku.edu/~keds/.
2004 by the authors
Abstract
Existing formal models of political behavior have followed the lead of the natural sciences
and generally focused on methods that use continuous-variable mathematics. Stephen Wolfram
has recently produced an extended critique of that approach in the natural sciences, and
suggested that a great deal of natural behavior can be accounted for using rules that involve
discrete patterns. Wolframs work generally does not consider models in the social sciences but
given the similarity between many of the techniques for modeling in the natural and social
sciences, his critique can readily be applied to models of social behavior as well. We argue
further that pattern-based models are particularly relevant to modeling human behavior because
human cognitive abilities are far more developed in the domain of pattern recognition than in the
domain of continuous-variable mathematics. We test the possibility of finding pattern-based
behavior in international behavior by looking at event data for the Israel-Palestine conflict for the
period 1979-2003. Using a new web-based tool explicitly designed for the analysis of event data
patterns, we experiment with three general patterns: the classic tit-for-tat, an olive branch
pattern designed to detect attempts at de-escalation, and four meta-rules that look at the
relationship between prior conflict and the propensity of the actors to engage in reciprocal
behavior. Our analysis shows that these patterns can be found repeatedly in the data, and their
frequency corresponds to changes in the qualitative characteristics of the conflict.
A New Kind of Social Science

Hudson, Schrodt and Whitmer
The world is a stupendous machine, composed of innumerable parts, each of which

being a free agent has a volition and action of its own, and on this ground arises the
difficulty of assuring success in any enterprise depending on the volition of numerous
agents. We may set the machine in motion, and dispose every wheel to one certain end,
but when it depends on the volition of any one wheel, and the corresponding action of
every wheel, the result is uncertain.
Niccol Machiavelli
For the past two centuries, the social sciences have aspired to produce law-like
generalizations about human behavior comparable to those found in the deterministic study of
mechanics in physics or the probabilistic models found in epidemiology. Unsurprisingly then,
social science has embraced the view that "the stature of a science is commonly measured by the
degree to which it uses mathematics" (Weinberg, 1975, 264, n.3/14) Considerable scientific
work has been done over the past sixty years in International Relations (IR) -- the social science
from whose vantage point we write this article, though its methods and conclusions can be
applied to social science more generally -- but this effort has produced virtually no law-like
generalizations, and what few might be said to exist give us almost no mileage in excess of what
common sense already provides (Walt, 1999; Green and Shapiro, 1995).
Indeed, a deep-seated discontent is growing in IR, in political science, in economics, and in
other social sciences: a deep methodological discontent. The most 'advanced' methods we can
use seem an ill fit with the types of questions we would like to pose and to answer in social
science. The more 'advanced' methodologically our fields become, the more removed from
reality and the more irrelevant to pressing human concerns the research seems to become. The
Post-Autistic Economics Movement in economics and the Perestroika Movement in political
science are but two recent manifestations of the yearning of social scientists, especially young
social scientists, to move beyond what is perceived to be the increasing barrenness of their
respective fields (see PS special issue July 2003; Fullbrook, 2001, Kasza, n.d.).
Yet the alternative methodological standpoints most often articulated propose to refocus
social science research on ancedotes, history and constructivism/discourse analysis, and thus
seem to have similar potential for controversy and paralysis. The issue here becomes
falsifiability and its relationship to causality (Yee, 1996). Unless some concept of a causal link,
however that be defined, is maintained in a methodology, it seems difficult to decide when a
particular historical or constructivist account is false, or at least less satisfactory than another
1

account. And since the findings of social science aim to inform social practice and policy, which
may have profound impact on the lives of individuals, some minimal falsifiability seems morally
imperative, if not theoretically imperative.
Some have viewed the social science enterprise as fated to be mired in stasis and
controversy, produced by the limitations of our available methodologies (Kuhn, 1962). Or is it?
Perhaps there is another way to look at this situation; a way which would produce more hopeful
possibilities. Rather than casting around the current landscape of methodologies, let us look
beneath it. Let us look at the common root of all these epistemological approaches, and work
upwards from there. Not only will we gain a new perspective on existing methodologies, but we
may be able to develop new methodologies, as well. These new lenses may be just the
prescription to move past stasis and controversy toward new social science capabilities.
In 2002, a methodological gauntlet was thrown done by Stephen Wolfram in his work, A
New Kind of Science. Though his book was not written from or for a social science perspective,
several of his assertions are pertinent to that endeavor. Wolfram asserts that most modern
scientific methods used in the physical and biological sciences are but idiosyncratic and limited
derivations from something much more basic, more fundamental, and more powerful. In place
of the continuous-variable mathematical structures that underlie classical mechanics and
statistics, Wolfram's approach focuses on the discrete transformation of patterns. Simple patternbased models can, through iteration, produce surprisingly complex behavior in physical and
biological systems. Biochemists, for example, search for patterns in amino acids as elements for
understanding the functions of a strand of DNA, and then the patterns of those strands combine
to produce the patterns formed by larger strands, then by chromosomes, then by the entire
genome. Though the patterns themselves are simple, they can ultimately produce highly
complex organisms, including human beings themselves.
Wonderfully for social scientists, humans do not only originate from patterns, but humans
seem hard-wired to perceive patterns and to find meaning in patterns. Indeed, it is not far off the
mark to suggest the ultimate basis of all human epistemology is discrete pattern identification.
As Wolfram puts it, "observers will tend to be computationally equivalent to the systems they
observe," an observation we will expound upon shortly (Wolfram, 2002, 737). Before we

discuss these issues in more detail, however, let us understand more fully the nature of our
methodological discontent in IR.
The Issue of Complexity as the Root of Discontent

When we measure too little, we need inference; when we measure too much, we need models
Willem Heiser1
All sciences are set up in such a way as to avoid, elude, and overlook complexity as much
as possible. Complexity is the bane of the would-be scientist, and yet complexity is a
characteristic of much, if not most, of the world around us.
Science is . . . unable to cope with [complexity], though its success with systems of its
own choosing has misled many scientists and politicians into thinking of science as a way
of effectively dealing with all systems. . . . The fruits of science are simple fruits, or more
precisely, the fruits of simplification. . . . We must begin to understand the limitations . . .
for its principal method is to squelch [complexity]. (Weinberg, 1975, 20-21).
For the social scientist, of course, the object of studyhuman beingsis irreducibly
complex:
Social scientists have had even less success, because their main interest--"humanness"-is a [complex] property that disappears when the system is taken part or abstracted and
averaged. . . . Perhaps we are reaching the useful limits of a science . . . whose
philosophical underpinnings are techniques restricted to systems of small and large
numbers. (Weinberg, 1975, 22)
We argue here that complexity is the root of our current methodological discontent in IR,
political science, and other social sciences. Looking at Figure 1, we see that the areas of study
justifiably approached through mathematical or statistical analysis and modeling are really quite
small. The vast remainder of reality--including social reality--cannot be effectively approached
in these ways.. For these approaches operate within very small confines, and if one's subject
matter cannot reasonably be placed inside those confines, use of the approaches will feel, and
indeed be, inadequate.
University of Leiden ISA-RC33 keynote address, Cologne, 5 October 2000. ISA-RC33 is the methodological
sub-section of the International Sociology Association. That ISA, not our ISA...

Mathematical
approaches
Statistical
approaches
Elements of understanding
Figure 1: Complexity and the Confines of Mathematical and Statistical Analysis
To date, the response to this situation has also been less than helpful. One response, very
common, is to simply apply mathematical or statistical methods beyond the confines where their
use is justified. This results in a strange methodological anomie, where one uses these methods
as if all is straightforward, while becoming ever more disengaged in one's questions and answers
from the reality wrestled with on a day-to-day basis by those who live within it. Statistically
speaking, we cannot justify assumptions of random or normal distribution in much of what we
are studying as social scientists. These assumptions, when used without justification outside of
areas noted in Figure 1 where aggregation and large-sample properties cause statistical
regularities, are actually an effort to avoid the issue of complexity. Mathematically speaking, we
have an enormous N-body problemin the sense of the problem in astrophysics, not statistics,
for which we have very little in the way of methodological capabilities. In a sense, we also have
a parallel problem in small N situations, where idiosyncratic factors such as the fact that the AlAqksa Mosque, the Western Wall and the Church of the Holy Sepulcher are within a 500-meter
radius of each other is intensely important to this conflict but it is decidedly not a large-sample
property subject to the normal or Poisson distribution. So we use simplifying assumptions that
evade the complexity with which we cannot cope.
4

But there is more. Most of these methods also derive from a strictly arithmetic view of
what can be the form of an interaction and usually involve a firm quantity-based definition of all
elements of understanding: models involving the analysis of interval-level variables are
substantially more developed than those involving nominal-level variables.2 But we as humans
know from our own lives that there are plenty of interactions in the world that have no
counterpart in continuous-variable operations3, nor can we define every concept in terms of
quantities. That is why we continue to have human diagnosticians, intelligence analysts, and
police detectives. As pattern recognition devices, our own brains are more powerfuland utilize
quite different mechanismsthan the most sophisticated mathematical and statistical methods,
and at a deep level, we realize that anew every time we read a piece of quantitative research in
the social sciences.
Mathematical and statistical approaches are a tiny and quite restricted subset of what the
human brain is able to bring to bear on a subject matter in pursuit of understanding. This is not
to say those approaches are not useful -- they are very useful, particularly in realms involving
large samples, high levels of noise, and variables that can be naturally operationalized using
continuous measures. But they are elementary methods compared to what we already know how
to do. As Wolfram puts it, "the field of mathematics as it exists today will come to be seen as a
small and surprisingly uncharacteristic sample of what is actually possible" (Wolfram, 2002,
821).
Humans were built to make sense of complexity. In a sense, the way to move past the
methodological discontent in our social science disciplines is to discover more about how our
minds in fact do this. "How we do this" is certainly the foundation of mathematical and
statistical approaches, but that foundation supports so much more in a methodological sense. If
we can explore that "more," we will give ourselves more powerful and less restricted
For example, most contemporary analyses of nominal variables use either dummy variable regression (nominal
independent variables) or variations on logit analysis (nominal dependent variables). Both techniques are
essentially mathematical tricks for treating the nominal variables as if they were interval, and their estimation is
entirely in the domain of continous variables.
As Wolfram puts it, "it is in many cases clear that the whole notion of continuity is just an idealization--although
one that happens to be almost required if one wants to make use of traditional mathematical methods." (Wolfram,
2003, 729).

methodologies specifically geared towards the understanding of complexity (see Political
Psychology, special issue 24:4, December 2003).
That, in a nutshell, is what we are about. In 1992, an issue of Science surveyed how
several new techniques in the physical and biological sciences had revolutionized not just the
methodologies, but also the theories, in their fields. The article observes:
Not everybody appreciates the importance of technique. Many scientists, in fact, are
"theory snobs" who dismiss technique as a kind of blue-collar suburb of science. ... [But
there is], clearly, enormous transforming power in techniques. In the absence of an
essential technique, a researcher or a field flounders, developing elegant theories that
cannot be decisively accepted or rejectedno matter how many intriguing circumstantial
observations are available. But with a key technique in hand, the individual and field
move ahead at almost terrifying speed, finding the right conditions to test one hypothesis
after another. Conversely, new techniques often uncover new phenomena that demand
new theories to explain them. (Hall 1992,345)
We believe similar theoretical progress can be made in IR through the exploration of new
methodologies.
Pattern Recognition, Human Understanding, and Human Action

It is becoming increasingly apparent that biological systems are much more complex
than the technological systems usually considered by the control engineer. A
technological system is usually designed on the basis of predesignated criterion of stability
and response which are expressed in some analytical form. Physiological systems, on
their other hand, have evolved slowly, continuously, adapting the performance of specific
tasks to a wide variety of conditions. Natural selection is not saddled with an
expediency demand. The evolution of physiological control systems might, therefore, be
expected to result in optimal systems chosen with the complexity of description not at all
entering as a limiting factor.
B. Pennock and E.O. Attinger.
Pattern recognition is the ability of an individual to consider a complex set of inputs, often
containing hundreds of features, and make a decision based on the comparison of some subset of
those features to a situation which the individual has previously encountered or learned.4 In
problem solving situations, recall can substitute for reasoning. For example, chess involves a
4The
literature on pattern recognition in human problem solving goes back to the 1960s, for example Newell and
Simon 1972, Simon 1982, and Margolis 1987.

well-defined, entirely deterministic system and should be solvable using purely logical
reasoning. Chess-playing computers use this approach, but Chase and Simon (1973) found that
human expert-level chess playing is done primarily by pattern recognition.
Before going further, we pause here to note that the chess analogy also helps us recognize
that pattern recognition by humans is also the basis for human action itself. Meaning for human
beings comes from the recognition of patterns (which imply that phenomena are rule-based), and
thus meaning in one's own behavior likewise comes from enactment of patterns, or, rather, the
rules that produce them. That is how chess is played, but that is also how virtually all social
behavior takes place, as well. To see the pattern-based nature of human understanding, then, is
to simultaneously see the pattern-based nature of human behavior. They are two sides of the
very same coin, and we ask the reader to keep this in mind for the discussion that follows.
Humans possess very large, albeit imperfect, long-term memories. Failures in the fidelity
of this memory are compensated by the fact that it is associative: we can recall a large amount of
information from a small number of features, even in the presence of noise. Information about
the fruit "apple" can be invoked by a smell, a taste, a variety of objects, a variety of words (e.g.
"apple", "Red Delicious", "Winesap", "cider" ) and a variety of social memories, as well as by
the noisy stimuli "Aqple" or the highly stylized trademark on an Apple computer.
Over the past two decades, international relations theory has begun to reemphasize the
importance of non-statistical patterns. The "international institutions" and "regimes" literature
(Krasner 1983) is quite explicit on this point. Robert Keohane, in his 1988 presidential address
to the International Studies Association, notes
..."institution" may refer to a general pattern or categorization of activity or to a
particular human-constructed arrangement, formally or informally organized. Examples
of institutions as general patterns of behavior include Bull's "institutions of international
society" as well as varied patterns of behavior as marriage and religion, sovereign
statehood, diplomacy and neutrality. ...What these general patterns of activity have in
common with specific institutions is that they both meet the criteria for a broad definition
of institutions: both involve persistent and connected sets of rules (formal or informal)
that prescribe behavioral roles, constrain activity, and shape expectations. (Keohane
1988,383)

While neither Keohane nor most of the international institutions literature have provided
formal definitions of these patterns, their emphasis on the importance of pattern in international
behavior is unmistakable.
Recall is preferred to reasoning because working memory5, which must be utilized in
deductive reasoning, is slow and constrained to handling only a few items of information at a
time. The long term memory used in pattern recognition, in contrast, is effectively unlimited in
capacity6 and works very quicklyon the order of secondseven when solving a complex
associative recall problems across thousands of potential matches. Purkitt notes:
Generally speaking, the power and complex of human cognition is tied to the almost
unlimited capacity of humans to store and combine vast amounts of information in longterm or associative memory. Research has also demonstrated that the active
processing of information is a serial process within the limited capacity of working
memory. In moving from the level of pieces of information to the level of factors or
indicators, it is now clear that individuals can only systematically process information on
a few (probably two or three) factors without explicit cognitive aids (e.g. algorithms).
(Purkitt 1991,40)
Associative memory is vast, effortless and quick; logical processing is limited, painful and
slow. Consider the following four questions:
Describe three uses of military force by the United States during the 1980s
Who was attorney general during the 1963 Cuban Missile Crisis?
What is 15% of $22.30?
Prove the Pythagorean Theorem.
The answer to the first question will come to the international relations specialist more or
less immediately "from memory". However, the method by which the answer was determined
cannot be describedfor example, were all instances of military force in memory searched, all
actions by the United States, or all international events in the 1980s? One cannot say; this is
instead done using "subcognitive processing", discussed below. The answer simply appeared,
Earlier known as "short term memory".
See Newell and Simon (1972), chapter 14. Newell and Simon argue that the capacity of associative memory is
effectively unlimited because the amount of time required to store items is sufficiently long that life-span, rather
than memory capacity, is the constraint.

without conscious operations, in a couple of seconds. Similarly, the second question can be
answered quickly despite being stated in a factually inaccurate manner.7
In contrast, most people can articulate the algorithm used to solve the third question. This
may be general-purpose school-book multiplication ("multiply 2 by 5, carry the one") or a
specialized algorithm developed because the problem is commonly encountered when
calculating restaurant tips ("divide the total by ten by shifting the decimal point to the left, then
divide that by two, then add these two amounts"). Failing these, one can solve the problem on a
calculator, and in any case the manipulation of the information can be verbalized without
difficulty. The final problem involves the logical manipulation of only a few axioms from plane
geometry, virtually every literate person has learned the proof in high school geometry, and yet
its solution is difficult for most people.
The latter two problems are far less information intensive than the first twothis is why
the third can be solved on a calculatorbut require deductive processing. In fact, the first two
problems require a very large amount of historical information and complex associative links.
The first problem could be solved using a large electronic data base such as NEXIS but
constructing a query that duplicates only the three examples usually retrieved by experts
(Grenada, Lebanon and Panama) is quite difficult. A slightly more difficult query, e.g. "Indicate
three changes in the NATO alliance between 1952 and 1972" becomes almost impossible even
for NEXIS. But such questions are nothing more than typical college examination questions and
barely worthy of consideration as expert political knowledge.
Origins of Pattern Recognition
In all likelihood, the human brain evolved with a strong bias towards pattern recognition
rather than deductive reasoning. This natural environment is comprised of two systems: the
physical and the biological. Many aspects of the physical world can be usefully described by
deductive axiomatic systems, and an information-processing system operating solely in a lawgoverned world would be able to survive with purely deductive reasoning; examples would
include computer viruses and programmed trading systems.
Robert Kennedy, and the crisis was in 1962, not 1963. The fact that the attorney general was the President's
brother and actively involved in the crisis aids in the recall; I suspect most people could not answer the same

The biological world, in contrast, is exceedingly complex and arbitrary. It is a world of
individuals constructed from complex feature vectors made of DNA, with billions of
components, and selected solely by the ability of their ancestors to reproduce, oftentimes in
unusual circumstances such as the aftermath of asteroid collisions. Such a world cannot be
described deductively in any practical sense, but because it is very repetitive, pattern recognition
is an effective information-processing strategy. If one Tyrannosaurus Rex tries to devour you,
the next one is likely to as well. Since critical decisions must be made in real time ("Is the object
approaching me sometime I can eat, something that will eat me, or something I can ignore?"),
evolution will select for high recall speeds under noisy environmental conditions. It does not
select for theorem proving or the minimization of quartic polynomials.
This neural bias would emerge early in the biological record, well before the development
of primates, or mammals, or even vertebrates. Homo sapiens sapiens is endowed with
sophisticated pattern recognition capabilities honed through eons of evolution, and it is
unsurprising that this capacity is put to use in social behavior. Deductive reasoning, in contrast,
is a comparatively recent development and is much more difficult. While we are very proud of
deductive reasoning, it is not necessarily more useful, particularly when dealing with social
behaviors which may also have some evolutionary roots.
Anderson and Rosenfeld trace the pedigree of this idea to William James:
As James points out [in Psychology (Briefer Course) (1890)] emphatically in several
places, the brain is not constructed to think abstractly it is constructed to ensure
survival in the world. [The design principles are:] do as good a job as you can,
cheaply, and with what you can obtain easily. If this means using ad hoc solutions, with
less generality than one might like, so be it. We are living in one particular world, and
we are built to work in it and with it. (Anderson and Rosenfeld 1988, 1)
Pattern recognition, unlike deduction, is easy.
An important consequence of the survival value of pattern recognition is a brain is biased
in favor of recognizing, rather than rejecting, patterns. As Hugh Kenner (Byte, November
1989,486) puts it, "The computer is a difference detector. The human mind is a similarity
detector." The survival costs of fleeing in terror from a dimly perceived and ultimately
nonexistent threat are substantially lower than the risks of not fleeing a genuine threat. The
question for the U-2 crisis.
10

ability of the brain to perceive patterns in random data is the bane of statisticians, but arises quite
naturally from the necessities of survival in a noisy environment.
Substantial parts of the brain are specialized for the social tasks of recognizing faces and
that most cognitively complex of all social interactions, language. It would not be surprising if
the brain had in addition some specialized hardware for handling at least some basic political
interactions, for example the social hierarchies present in many vertebrates.
Human associative
memory may be able to handle, subcognitively, complex episodic political information such as
precedent retrieval in part because the brain evolved in part to handle comparable problems.8
Subcognitive Processing: Seeing, Feeling, and Understanding
Studying associative recall is problematic because the process occurs in the non-verbal,
unconscious or subcognitive9 part of the brain: it is a form of information processing that we can
do but not articulate. In the foreign policy field such reasoning is typically called "intuition" or
"feel." But a better description might be "seeing" or "perceiving." A typical example of this
approach to foreign policy analysis is found in the following quote from Kissinger:
Statesmanship requires above all a sense of nuance and proportion, the ability to perceive
the essential among the mass of apparent facts, and an intuition as to which of many
equally plausible hypotheses about the future is likely to prove true.(Kissinger 1979,39)
Subcognitive information retrieval involves nothing mystical; the process can be, and has
quite extensively been, empirically studied (see Collins and Smith 1988; Reber 1993). One
can engage in very complex information processing without being aware of how one is doing
it, even using introspection. Lashley notes:
No activity of the mind is ever conscious. [Lashley's italics] This sounds like a paradox,
but it is nonetheless true. A couple of illustrations should suffice. Look at a complicated
scene. It consists of a number of objects standing out against an indistinct background:
Masters (1989) provides a thorough discussion of the possible connections between evolution and political
behavior; Axelrod (1984) and Simon (1990) note that evolution may have predisposed humans to altruistic
behaviors, a definite change for the bellicose "Social Darwinism" of a century ago.
This term is from Douglas Hofstadter (1985) via Holland et al (1986). The word is attractive since unlike
"unconscious" it implies active information processing; it avoids the Freudian overtones of "subconscious" and it
is more general than the term "nonverbal". Jackendoff (1987), while dealing with an entirely different set of
domains, provides an excellent discussion of subcognitive information processing and a guide to much of the
relevant psychological literature in the linguistic and visual perception domains; Springer and Deutsch (1981) give
a semi-popular review of the related literature on split-brain experiments.
11

desk, chairs, faces. Each consists of a number of lesser sections combined in the object,
but there is no experience of putting them together. The objects are immediately present.
When we think in words, the thoughts come in grammatical form with subject, verb,
object, and modifying clauses falling into place without our having the slightest
perception of how the sentence structure is produced. (Lashley 1956,4; quoted in
Jackendoff 1987,45).
As Lashley points out, the most obvious examples of subcognitive processing are sight and
language. For example, an illiterate speaker with no formal linguistic knowledge can continuously
and effortlessly assemble grammatically correct sentences consistent with a grammar of several
dozen, if not several hundred, rules. This skill is learned entirely by example and can be observed
in any average three-year-old child, in any culture. One can also construct grammatically correct
sentences through rulesas is done when one is learning a second language or in a machine
translation systembut this is slow, awkward and not characteristic of fluency.
Wittgenstein notwithstanding, a long series of empirical experiments demonstrate that one
can know or see things one cannot say, where "know" or "see" is defined as the successful
completion of complex information-processing tasks and "say" means verbalize. One of the
consistent lessons from the machine learning literature is that verbal problem solving protocols
provided by human experts are usually much more complicated than the set of rules logically
sufficient to solve the problem. An expert usually perceives or "sees" that much more
information is required to solve a problem than is logically needed, and when asked why will be
unable to describe what that additional information provides other than "feel".
A likely explanation for this is that the expert actually does most of the problem solving
using associative pattern recognition and therefore cannot articulate the process. The verbal
protocol represents what the expert thinks he or she is thinking, rather than providing an actual
description of the problem-solving process. Protocols are very useful in determining the
information that an expert uses, but they cannot provide an accurate step-by-step description of
the information processing.
This leads to a situation that is very interesting from the social scientist's point of view.
The more one's methods allow you to see, in a relatively unmediated way, complex social data,
the more likely it is the social scientist will be able to tap into their powerful subcognitive
processing abilities to uncover patterns, and by extension, rules that produced them. The more
our methods take us away from a relatively unmediated "seeing," the more difficult it will be for
12

social scientists to find the patterns that make possible human, and social science, understanding.
While standard techniques of mathematics and statistics can help us confirm we are not
imagining things, these techniques can only play a supportive role in social science, for social
science is based upon these deeply rooted and profoundly powerful types of human processing.10
Political Behavior as Rule-Based

McNamara asked [Chief of Naval Operations George] Anderson what he would do if
a Soviet ship's captain refused to answer questions about his cargo. At that point the Navy
man picked up the Manual of Naval Regulations and, waving it in McNamara's face,
shouted, "It's all in there." To which McNamara replied, "I don't give a damn what John
Paul Jones would have done. I want to know what you are going to do, now."
Allison 1971,131
One man alone can be pretty dumb sometimes, but for real bona fide stupidity there
ain't nothing can beat teamwork.
Edward Abbey
We have noted in a previous section that pattern recognition and rule-based human
behavior are but two sides of the same coin: the way that we impute meaning through pattern
recognition is the way that we create meaning through our behavior. While this is easy to see
for, say, individual chess players, most domestic and international political behavior is primarily
the product of bureaucracies rather than individuals. An individual may influence the direction
of a policy, but the implementation is still left to bureaucracies.11 In everyday language this
10
Indeed, it is interesting to speculate that so many of our students have trouble understanding social phenomena
through mathematics and statistics not because they are less intelligent, but because such techniques seem to the
normal mind to be less powerful and less informative than simple social intuition which, far from being
simple is in fact phenomenally complex. Perhaps the Post-Autistic Economics Movement may be seen as
having chosen its name with care, for perhaps those who find these methods easy to use and particularly insightful
may have brains that are somewhat atypical. This is not to necessarily suggest a pathological mental
conditionthe dramatic example of one of the giants of game theory, John Nash, notwithstandingbut rather that
the tools through politics can be most comfortably understood may be quite different for those attracted to the
contemplative abstractions of the academic life and those attracted to the rapid-paced, crisis-laden human
interactions of the policy world.
11
The focus here is on the Weberian "rational-legal" bureaucracy that dominates foreign policy decision-making
structures in the modern era. This applies even in "revolutionary" situations: For example, the taking of American
hostages in Iran during 1979-80 was outside the normal range of international behavior, but the attempts at
resolving that issue were very much normal, including the unsuccessful rescue attempt and the eventually
successful mediation by Algeria and other international agents. Similarly, the rhetoric surrounding the Iran-Iraq
13

organizational interaction is simplified"Hitler decided to attack Poland"but in virtually all
cases (and certainly in systems which have a strong democratic and/or bureaucratic component)
individuals are constrained to choose from a very small set of options that have been made
available through a bureaucracy. While the "Great Man" [sic] model attempts to allow the
cognitive processing of an individual replace bureaucratic decision-making in an organization,
the individual is still dependent on an organization to supply (and filter) information and
implement decisions. Behind every Great Man is a well-entrenched bureaucracy pleased to have
someone else taking responsibility.
The shift from individual to organizational information processing produces a paradox: An
organization, in order to obtain the capacity for large-scale information acquisition and policy
implementation, must make substantial compromises in its information processing capabilities.
Specifically, an organization must work within the bandwidth constraints of language, and it
cannot directly invoke associative recall. In contrast, the sequential processing of if...then rules
can be done within the constraints of working memory so this, rather than associative pattern
recognition, is the preferred mode of information processing in organizations. In much of their
behavior, the bureaucracies are not acting as if they followed rules; they are instead explicitly
following rules and are expected to do so, rule-following being a sine qua non of bureaucratic
behavior.12
Some of the early computational modeling projects in political science (for example
Carbonell 1978, Thorson and Sylvan 1982, Sylvan and Chan 1984, Andriole and Hoople 1988)
assumed that due to the rule-oriented nature of bureaucracies and the simplifications inherent in
popular ideologies, one would be able systematically to extract an organizations rules and
precedents from a sufficient quantity of debates, formal regulations and internal memoranda, and
from these rules one could simulate much of the decision-making process. Based on the
subsequent success of rule-based systems in replacing some routine managerial decision-making
in business, this was not an unreasonable proposition. In fact, had the computational modeling
war and the personal animosity between Saddam Hussein and Khomeini provided personalistic aspects to that
conflictas would the animosity between Saddam Hussein and various U.S. presidents named George Bush in
the period 1990-2003 and between but the military interactions were bureaucratically quite conventional.
12
Thorough treatments of the role of rules and heuristics can also be found in Majeski (1987), Mefford (1991), and
Sylvan, Goel and Chandrasekran (1990); the approach also permeates the Kahneman, Slovic and Tversky research.
14

projects focused on routine State Department activities, such as the processing of visa requests or
arranging golf games for visiting members of Congress, the approach might have worked.
Instead, because an extensive set of documentary evidence was required, the typical focus
of these projects was on critical decisions such as Vietnam. In crisis decisions explicit rules
proved insufficient. Sylvan and Majeski's study of Vietnam decision-making quickly
encountered the problem of "tacitness" and "tacit cultural rules" where many of the key elements
required to understand the bureaucratic debate were absent from the recorded discourse:
For example, the first few months of 1961 were marked by the eruption of a nonVietnamese war in Indochina. The country, of course, was Laos, a subject on which
Kennedy's advisers spent literally hundreds of hours. Yet the Vietnam documents of that
time barely mention Laos. It is only after the crisis fades that explicit references to the
Laos settlement are made, particularly in the form of 'lessons'. ... Our heuristics reflect
very little of situation-specific interpretations, and this means we miss many of the
allusions and other references in any given text. (Sylvan and Majeski 1986,11)
For example, the fundamental theme of anti-communism was never verbalized because it
was a shared assumption:
Most people employ a host of common sense rules for getting along in daily life; only a
small handful of these will ever be verbalized. In part, this is because of their
obviousness; in part, due to the embarrassment that would attend someone who reminded
others about them. If rules are shared in the [bureaucratic] culture as a whole (e.g.
communism is bad), they will never (or almost never) be made explicit. (Sylvan and
Majeski 1986,10)
Boynton noted a similar problem in trying to formalize the construction of narratives in
Congressional hearings on the 1983-84 Lebanon policy:
The apparent reason a narrative account was not constructed was that members of the
committee knew what had happened. They had been arguing about Lebanon policy for
well over a year, and they could take the knowledge of the events for granted. My
tentative conclusion is that narrative reconstruction is a fallback mode of reasoning.
When the subject is well known [it] is not necessary (Boynton n.d.,8)
Because understanding involves matching observed events to a pattern, the function of
political discourse is to provide sufficient informationin the forms of declarative knowledge,
event sequences and substitution principlesto cause the audience to understand (i.e. pattern
match) the situation in the same manner that the individual transmitting the information
understands it. Political information transfer attempts to stimulate pattern recognition in the
15

mind of the audience and thereby trigger a desired behavior. This process can occur between
competing organizations as well as within them: Signaling in a conflict situation involves
exchanging messages with an opponent in an attempt to get the opponent to take, or refrain from
taking, certain actions. Thus, even the existence of bureaucratic actors does not diminish the
human characteristics of pattern recognition as central to both understanding and action.
Agency, Computational Irreducibility, and the Question of Causality

If we as social scientists turn to this philosophical position--that pattern recognition is
central to human understanding and action--and we build methodologies to take advantage of
these powerful natural capabilities--we will have done more than simply proliferated
methodological alternatives. We will have made possible the social scientists' embrace of human
agency.
One of the most important critiques of quantitative social science methods is that they are
incapable of explicitly embracing human agency. Indeed, it seems that agency is a nuisance, a
perturbing variable and a source of noise, something complex that must be simplified out of
consideration, rather than a central focus of investigation. This sacrifice of agency has seemed a
justifiable trade-off for the ability to approach prediction, for prediction is seen as the side-door
to causality.
For example, in the statistical approach the focus is almost always on aggregate
behaviors. This is because such studies can take advantage of the Central Limit Theorem and
other regularities that occur when one aggregates large numbers of variables that behave as if
they were randomthat is, the values of these variables cannot be predicted by the independent
variables used in the study. This can yield some interesting insights, but they are insights of a
different kind than the natural meaning-from-pattern-recognition process that we have discussed
thus far. For example, one of the oldest and most solid statistical results in the study of war is
the Poisson distribution of wars over time, first established by Richardson (1960). In sufficiently
large samples, the distribution of wars occurs as if it was the result of a random selection of rare
events. This is an interesting finding, worthy of note in any study of war. Nevertheless, this
does not imply that the mechanisms or processes that led to any particular war are random: in
fact, in all cases one canand, indeed, the social scientist as human pattern recognizer may feel
16

positively impelled toassemble an extensive historical record indicating that the war resulted
from a massively complex series of agency-produced decisions that were made for reasons that
could be explicitly linked to various conditions in the world at the time. While those reasons
may have some elements that might be considered randommessages get lost or misunderstood;
individuals become incapacitated due to accident or disease; changes in the weather allow armies
to move or force them to remain in placein virtually all cases a narrative can be (and is)
constructed that explains the war based on a probing of human (individual and bureaucratic)
agency patterns linked to behavior, which provides meaning. Thus, although there are elements
of the aggregate behavior that follow, quite precisely, a purely random model identifiable by
statistical means, we cannot embrace human agency by using that pathway. We can only
provide an account of human agency in social science by the same pathway we provide an
account of it in daily life: seeing patterns and linking them to agency-based and meaningproviding rules.
It is fair to say, then, that this orientation suggests a whole new way of looking at causality
in social science, a way that allows for an explicit embrace of human agency and the change it
produces in human affairs. In a sense, this new approach redefines causality, falsifiability, and
other important concepts in science and the social sciences.
If pattern recognition is the basis of human understanding of human behavior, then while
one can specify rules that govern human behavior, it will be impossible to know for a surety in
advance all of the consequences produced thereby. Many consequences can be known, but never
all. As a result, there is, as Wolfram puts it, a "computational irreducibility" about rule-governed
human behavior. This means that there is a limit to prediction in any theoretical science of social
phenomena. In dealing with complex phenomena such as social behavior, a theorist will have to
readjust his sights: specification of the rules, and an understanding of less than all of the
consequences thereof, will now be his aim. Wolfram puts it this way:
When viewed in computational terms most of the great historical triumphs of theoretical
science turn out to be remarkably similar in their basic character. For at some level
almost all of them are based on finding ways to reduce the amount of computational work
that has to be done in order to predict how some particular system will behave. . . . If one
views the evolution of a system as a computation, then each step in this evolution can be
thought of as taking a certain amount of computational effort on the part of the system.
But what traditional theoretical science in a sense implicitly relies on is that much of this
17

effort is somehow unnecessary--and that in fact it should be possible to find the outcome
of the evolution with much less effort. . . . In traditional science it has usually been
assumed that if one can succeed in finding definite underlying rules for a system then this
means ultimately that there will always be a fairly easy way to predict how the system
will behave. . . . But now computational irreducibility leads to a much more fundamental
problem with prediction. For it implies that even if in principle one has all the
information one needs to work out how some particular system will behave, it can still
take an irreducible amount of computational work actually to do this. . . .And this, I
believe, is the fundamental reason that traditional theoretical science has never managed
to get far in studying most types of systems whose behavior is not ultimately quite
simple. [A]t an underlying level this kind of science has always tried to rely on
computational reducibility, [s]o when computational irreducibility is present it is
inevitable that the usual methods of traditional theoretical science will not work. And
indeed I suspect the only reason that their failure has not been more obvious in the past is
that theoretical science has typically tended to define its domain specifically in order to
avoid phenomena that do not happen to be simple enough to be computationally
reducible. (Wolfram, 2002, 737-742)
Our modified aims will then modify our definition of causality when attempting to
understand human behavior. The act of imputing causality is the act of identifying rule-based
patterns in the phenomena we study, with the caveat that the complete consequences of the rules
specified are probably not going to be knowable in advance. Wolfram states, "whenever
computational irreducibility exists in a system it means that in effect there can be no way to
predict how the system will behave except by going through almost as many steps of
computation as the evolution of the system itself" (Wolfram, 2002, 739). In general, then, there
are no valid shortcuts to take, for we are not operating in a context of computational reducibility
(generally speaking) in the social sciences.
Here is another way of approaching this redefinition of causality in the social sciences:
"[A]bstract descriptions will never ultimately distinguish us from all sorts of other
systems in nature and elsewhere. And what this means is that in a sense there can be no
abstract basic science of the human condition--only something that involves all sorts of
specific details of humans and their history. So while we might have imagined that
science would eventually show us how to rise above all our human details what we now
see is that in fact these details are in effect the only important thing about us. . . [The
computational irreducibility of social phenomena] is what allows history to be
significant--and what implies that something irreducible can be achieved only by the
evolution of a system. Looking at the progress of science over the course of history one
might assume that it would only be a matter of time before everything would somehow
be predicted by science. But . . . the phenomenon of computational irreducibility now
shows that this will never happen. There will always be details that can be reduced
18

further--and that will allow science to continue to show progress. But we now know that
there are some fundamental boundaries to science and knowledge. And indeed in the end
[computational irreducibility] encapsulates both the ultimate power and the ultimate
weakness of science. For it implies that all the wonders of our universe can in effect be
captured by simple rules, yet it shows that there can be no way to know all the
consequences of these rules, except in effect just to watch and see how they unfold."
(Wolfram, 2002, 846)
We will adopt this new view of causality in the empirical work that follows. To wit, we will
aim to see patterns in social behavior, and we will aim to specify rules that we have good reason to
believe the human agents involved are in fact using to produce those patterns. Unlike model
specification in traditional quantitative approaches, we cannot content ourselves with specifying
the model that produces the highest r-squared, to use an analogy. Our attempts at understanding
the behavior in question will only be successful if we embrace the human agency involved in
producing those patterns in the first place. We can only understand if we understand via agency.
Even with this new definition of causality, we previously spoke of the moral imperative in
social science to preserve a concept of falsifiability, for social science seeks to inform social
policy. What does falsifiability now mean in the context of this new view of causality?
Fortunately, we can still claim a form of falsifiability, while simultaneously asserting that
falsification cannot be completely determinable. Unlike the situation where one has asserted "all
swans are white," and then comes upon a black swan that falsifies the assertion in a universal,
time-independent sense, our new view of causality imposes a temporal boundedness to our
ability to falsify. Thus, while it is still possible to determine for a given period of time whether
one specific set of rules is more or less satisfactory than another specific set of rules in
accounting for the behavior during that period, we would no longer be able to say with any
degree of certainty that that judgment would hold outside of that time period. In a sense, then,
falsifiability is not a one-shot process. As the system evolves and unfolds, what was a less
satisfactory rule specification in one time frame may prove to be the most satisfactory
specification in another. As these judgments change, so we would expect policy to change, thus
meeting the moral imperative we have spoken of previously.
19

Seeing Human Behavior

As we noted previously, humans perceive patterns. There is a feeling or a seeing of patterns
that cannot be verbalized because of its strong dependence on associative reasoning. Thus, if we
wish to build new social science methodologies based on human pattern recognition capabilities,
we must invent a way of seeing human behavior. Wolfram encountered much the same challenge
in analyzing his cellular automata, which he created with the specification of arbitrary rules:
[O]ne of the major features of the new kind of science that I have developed is that it
does not have to [confine itself to computationally reducible systems]. . . .And that is
why--unlike most traditional works of theoretical science--this book has very few
mathematical formulas but a great many explicit pictures of the evolution of systems. . .
But if one has such a system, how does one decide what questions are interesting to ask
about it? Without the guidance of known theorems, the obvious thing to do is just to look
explicitly at how the system behaves--perhaps by making some kind of picture."
(Wolfram, 742, 793)
Wolfram himself has run hundreds of thousands of rules and sets of rules, noting the
behavioral outcome over a given time stream. This decades-long enterprise has yielded the
understanding that some rules and combinations thereof produce patterns that grow and
predominate, whereas others create patterns that dwindle and die out. Some patterns display a
complexity nearly completely understood from the rules themselves, while others are
unexpectedly complex and not completely comprehended by an examination of the rules
themselves. Figure 2 is an example shows an example of each general pattern: from left to right,
code 357 dies out; code 600 produces a complex but repeating pattern; code 1599 produces a
complex pattern that does not repeat for the duration of the run displayed, and code 2058
produces a pattern that does not repeat (it continues to expand beyond the limits of the display
shown here) but ceases to produce complexity.
20

Source: Wolfram, pg. 69
Figure 2: Examples of general categories of patterns

Patterns can also be analyzed at several levels when one is dealing with more than one
rule. Wolfram found that complexity reached its peak with three or four rules, and that
additional rules did not create additional levels of complexity, a finding whose significance we
will discuss in a moment. Yet in systems using, say, three rules, it is possible to look at the
patterns of each rule in isolation, and then in combinations of two, and then in the final
amalgamation of all three. Patterns at lower levels may produce unexpected patterns at higher
levels of observation, producing, if you will, meta-patterns from the contributions of the
individual rules. Figure 3 shows a wide variety of different patterns that can be produced from
relatively simple rules, along with Wolframs commentary on this.
21

In our quest to move towards this new methodological approach, we have had to confront
several challenges in devising a plausibility probe:
n do we have time stream data of human behavior in IR that we could use for a test?
n can we create a system of seeing that human behavior?
n can we specify a small set of rules understood to be used by the human agents involved?
n can patterns and meta-patterns be discerned in this way, thereby accounting for the
behavior in question?
It is to these questions that we now turn.
22

Source: Wolfram, pg. 107
Figure 3: Wolframs example of the diversity possible with simple rules
23

Why Events Data? Why the Israelis and Palestinians?

A mathematician may say anything he pleases, but a physicist must be partially sane.
Josiah Willard Gibbs
In this exercise, we attempt to show the possibilities and pitfalls of Wolfram's approach
to understanding human behavior in the arena of international politics. We selected events data,
not because we are not familiar with the weaknesses of events data, but because the noisy time
stream of events data closely resembles the types of inputs humans are constantly processing in
social pattern recognition. Data was coded into the WEIS scheme using TABARI, a computer
program from the Kansas Event Data System (KEDS) project that creates event data from
machine-readable text.13 The events were coded from Reuters News Service lead sentences
obtained from the NEXIS data service for the period April 1979 through May 1997, the Reuters
Business Briefing service for June 1997 through September 1999, and Agence France Presse
from October 1999 to December 2003. The coding software, coding dictionaries and data are
available at the KEDS web site, http://www.ku.edu/~keds.
For the event counts, we use the following categories based on the WEIS two-digit cue
categories:
verbal cooperation: WEIS categories 02, 03, 04, 05, 08, 09, 10
material cooperation: WEIS categories 01, 06, 07
verbal conflict: WEIS categories 11, 12, 13, 14, 15, 16, 17
material conflict: WEIS categories 18, 19, 20, 21, 22
This reduces the number of distinct event categories that can be used as independent variables to
a manageable amount and eliminates the problem of three-digit WEIS categories that have very
low frequencies. It is also likely to reduce the effects of coding error somewhat: Several of the
verbal conflict codes in WEIS are ambiguous even for human coders, and the automated
coding probably generates some misclassification within those categories.
For this first exercise, we also wished to choose a dyad whose behavior was highly
reactive one to another and has been the focus of sufficient media attention that we can be
13
Discussions of machine coding can be found in Gerner et al. (1994), Schrodt and Gerner (1994), Huxtable and
Pevehouse (1996), and Bond et al. (1997).
24

confident that the event data are a reasonably accurate description of the actual behavior in the
system. The dyadic relationship between the Israelis and Palestinians, while certainly affected
by the initiatives of third parties, is nevertheless quite internally reactive, as many scholars have
noted (see for example Bickerton and Klausner 1998, Gauss 1998, Gerner 1994, Goldstein et al
2001, Tessler 1994). Again, this particular dyad is a good place to start for this exercise.
While we will be using event data in this analysis, one could use this approach on almost
any information on human behavior over time. One could look at the time stream of memoranda
that comprises the Pentagon Papers, for example. One could look at market behavior over time.
One could look at negotiation moves between the parties involved in the North Korean talks.
And one need not only look at monads or dyads, but could examine N-ads, as well. Any human
behavior, or artifact thereof, that can be laid out in a time stream can be viewed from this
perspective. The data need not be at any level of measurement precision beyond categorical.
And one can combine types of data; one could look at patterns made by time streams composed
of behavior found in presidents' speeches and currency values, for example. Unlike
mathematics-based methods where data must be at the same or nearly the same level of
measurement precision to be combined, there is no such stricture in this method.
A Way of Seeing: Introducing the Website

We needed a mechanism by which we could "see" what the Israelis and Palestinians were
doing to each other. Accordingly, using the KEDS data, we created a web-based instrument of
sight, which is to be found at http://ep.jhax.org and which is entitled "Event Patterns."
Please use your browser to go there now.14
When you access the site, you will see its basic format is a variety of inputs to the system,
some set by default, some requiring input to proceed. At the end of this list of inputs, you
activate the "seeing" mechanism by hitting one of three buttons: "None," "Graphic," or "Text."
With "none" you will get no display. With "graphic" you will get a time stream graphic using
symbols to denote types of behavior by the Israelis (blue) and the Palestinians (green). With
14
There is a guidebook to using the website, which is provided in our appendix, but is also online at that same URL.
As of 15 March 2004, the default values in the web site are set to the rules analyzed in this paper, and therefore it
25

"text," you will receive a raw coding output of the KEDS data. For our purposes in this paper,
we will be choosing "graphic."
However, before you hit "graphic," there are a minimal set of inputs that you must provide.
At the very least, you must input a starting data and an ending date to bound the display. The
possible range is 19790415 (15 April 1979) to 20031231 (31 December 2003), the limits of the
data that we have available.15 Dates are entered full year first, then month, then day. If your
computer is not very powerful, we suggest you upload one year's worth of display first to test
your machine's capabilities. More capable machines should be able to handle the entire two
decades, if so specified.16 Try inputting a time frame and then hitting "graphic" now. Once you
hit that button, your computer may display generic "image icons" for a moment as it performs its
computations, but in a few moments those will resolve into graphic symbols.
The more interesting possible inputs are entitled "patterns" and "display." It is here that
you have the capability to perform discrete pattern transformations on the graphic output. This is
where you may "try out" putative rules that you believe the Israelis and Palestinians are using,
and then you may "see" whether those patterns account for any of the behavior in the set. We
will demonstrate those capabilities in a moment.
At the time of this writing, we have partially instantiated the capability to look at any of
the actors in any of the KEDS events data sets. When you click the button to the right of the
default "data source" input, you will see options of looking at events in the Balkans, Africa,
China, Iraq, and many arenas. What is not fully instantiated at the time of this writing is to
provide a list to you of alphabetic actor codes and subnational codes. That will come before we
submit this paper for publication; they can also be obtained by downloading the appropriate
coding dictionaries from the KEDS web site..
can be used to generate a complex display without entering any specific values. The discussion that follows
assumes that you want to look at something other than the default rules.
15
The Kansas Event Data System project has a Levant data set at http://www.ku.edu/~keds/data.html/levant.html
that is updated every three months, and at the present time (March 2004) is current to 31 December 2003.
However, due to a change in the availability of Reuters report, that data set switches to Agence France Presse as a
source after 30 September 1999, and the frequency of events reported by AFP is substantially greater than that
reported by Reuters. In addition, the final months of the Reuters data appear to be extremely sparseprobably due
to the gradual deterioration of Reuters services that eventually led us to switch to AFPand consequently the
apparent absence events in 1999 may be an artifact of the data.
16
Which is to say the display works just fine on Schrodts Macintosh G5 but crashed Hudsons Dell.
26

There are other possible inputs to play with. Most concern the size and characteristics of
the graphic display itself. But there are others that may have more substantive use when probing
for rules-based patterns in the data, such as "days per unit."
Rule Specification and Visual Inspection in this Probe

In this initial probe of the approach, we desired to specify some very simple rules and
then see how well they "accounted for" the behavior of the Israelis and Palestinians. Respecting
agency as the source of both Israeli and Palestinian behavior, we needed to specify rules that we
could justify as having meaning and making sense for the actors involved. Thus, as noted
previously, rather than selecting rules on the basis of how well they conformed to an event
stream, we selected rules on the basis of how well they corresponded to what experts in the field
assert are the rules these actors do use.
Wolfram himself provides encouragement that the rules need not be many, and neither do
they need be complex. For example, he states, "Simple and definite underlying rules can
produce behavior so complex that it seems free of obvious rules" (Wolfram, 2002, 752) and then
goes on to elaborate that in his years of experience analyzing complex systems,
But when in general does complexity occur? [I]f the rules for a particular system are
sufficiently simple, then the system will only ever exhibit purely repetitive behavior. If
the rules are slightly more complicated, then nesting will also often appear. But to get
complexity in the overall behavior of a system one needs to go beyond some threshold in
the complexity of its underlying rules. The remarkable discovery that we have made,
however, is that this threshold is typically extremely low. [I]t ultimately takes only very
simple rules to produce behavior of great complexity. . . . Instead, once the threshold for
complex behavior has been reached, what one usually finds is that adding complexity to
the underlying rules does not lead to any perceptible increase at all in the overall
complexity of the behavior that is produced. (Wolfram, 2002, 105-6)
Indeed, Wolfram found that the most complex behavior could be obtained with the use of
approximately three rules. We feel that there is reason to believe that the set of rules being
employed by the Israelis and Palestinians in enacting what they feel to be meaningful behavior
toward one another is also not very large, nor very complex. Signalling between organized
human collectives, especially those in conflict, almost mandates that only a small set of simple
rules be used in order to maximize the chances that the other group will understand the meaning
intended by the action.
27

Furthermore, because international politics is a complex problem solving environment,
heuristicssimple rules used to partially solve complex problemsare of particular importance.
Purkitt observes:
To cope with limited cognitive capabilities, individuals selectively process information
and use a limited number of heuristics or mental rules of thumb as cognitive aids in their
effort to manage information. This apparently universal reliance on intuitive heuristics to
solve all types of problems seems to be due to the need to compensate for the limitations
of short-term memory and information processing capabilities. By using intuitive mental
heuristics, people can develop a problem attack plan which permits them to develop a
subjectively acceptable problem solution. (Purkitt 1991,43)
In fact, rational choice and balance of power theories are both heuristics in the sense that
they are relatively simple; they come with a complex set of side-conditions; and they are
intended as general rules to guide decision-making, without providing a complete specification
of actions to be taken. To the extent that an heuristic is shared by the decision-makers in a
political systemfor example balance of power in 19th century European diplomacy or the
Chicken game in 20th century nuclear deterrenceit reduces uncertainty and becomes selfvalidating.
Therefore, we endeavored to come up with a small set of fairly simple rules that could be
justified on the basis of scholarship concerning Israeli and Palestinian actions.
The first rule we used was the classic "tit-for-tat" approach immortalized by Anatol
Rapoport and, more recently, Robert Axelrod (1984). Country experts have asserted that the
Israelis and Palestinians consciously use this rule. If the Israelis followed Palestinian aggression
(or cooperation) of a certain threshold and in a certain time period by aggression (or cooperation)
of their own (or vice versa), then we could account for such behavior using the tit-for-tat rule.
We could also identify instances where tit-for-tat was not operating, thus providing a way of
"seeing" which behaviors could be accounted for by the rule and which could not.
But we also wanted to compare tit-for-tat with another rule, for two reasons. First, we
wanted to see the comparative ability of two rules to account for Israeli and Palestinian behavior.
That is, we wanted to demonstrate that such a comparison, which is necessary for our modified
definition of falsifiability, is possible. Second, we wanted to see the operation of two actorbased rules operating simultaneously. We wanted to know if it was possible to "see" such a
simultaneous operation of more than one rule.
28

Therefore, the second rule we used was one that we have labeled the olive branch: one
side responds to a period of conflict with cooperation rather than reciprocating the conflict. The
olive-branch rule is the standard gambit for breaking out of the mutually-destructive
DD/DD/.../DD sequence in the classical prisoners dilemma game.
In addition to analysis of actor-based rule specification, we also believed it was possible
that the Israelis and Palestinians were creating a "meta-pattern." Now, to this point, we have
only been interested in using rules defined by actor agency to account for that actor's own
behavior. But what if we performed a discrete pattern transformation that allowed us to see what
behavior the Israelis and Palestinians were, to use an analogy, weaving together? The
combination of the idiosyncratic use of the actor-rules might be producing a pattern all its own--a
pattern that neither party intended. Viewing such a meta-pattern might allow us to reflect not on
the meaning each actor intended, but the meaning produced togetherwhich might have an
altogether different character.
Patterns and Meta-Patterns in the Israeli-Palestinian Dyad:

Accounting for Behavior with Rules
All of the analyses that follow are derived from a set of rules implemented at the default
web page at http://ep.jhax.org.(accessed 12 March 2004). As noted above, these
combine a tit-for-tat (TFT) analysis with several meta-rules based on whether or not the actors
are adhering to the TFT pattern of behavior. We define TFT with respect to behaviors aggregated
over a six-day period, and look only at material conflict and cooperation; these are compiled
separately. An actor is defined as showing the behavior only if the number of events exceeds a
set thresholdthis is 4 events in a six-day period for Israel and 2 event for Palestine (the
asymmetry is used because the media report Israeli action with a much greater frequency than
Palestinian actions). The various other indicators and meta-rules are built from these threshold
events. The indicators were downloaded from the web site using the Text option, and then
reading those results into MS-Excel.
Because there are 4485 points in the complete data set (15 April 1979 to 31 December
2003 in two-day intervals), it was necessary to construct some means of summarizing the results.
The figures below give 32-day moving totals of the number of times that a pattern was matched:
29

this measure has the value of 16 if the pattern matched in every one of the two-day periods in the
32-day interval. The values for Israel and Palestine are mirrored across the X-axis, with the
counts for Israel above the axis and Palestine below.
As we anticipated, there is a very large discontinuity in all of the graphs following 1999.
Unfortunately, this is likely due to two factors, one involving the situation and one involving the
data. The situational discontinuity was the outbreak of the second Palestinian intifada in
September 2000. The level of violence during this period was substantially greater than that
experienced during the first intifada in 1988-1991, and consequently the number of reports of
violence is objectively higher. In addition, the Palestine Authority had achieved some degree of
international legitimacy following the Oslo peace agreement, and the international response in to
the second intifada has been more sustained and open than that during the first intifada, where
the U.S. in particular engaged in negotiations with the Palestinians very reluctantly. Objectively,
one would expect the two periods to appear different.
However, we also have a change in data sources here, with Reuters prior to October 1999
and Agence France Presse (AFP) following that period. As discussed in Schrodt, Gerner and
Simpson 2001, AFP generally has a much higher density of coverage of the Israel-Palestine
conflict than Reuters has in the periods where we can examine the coverage of both sources.
Consequently some of the increased intensity of coverage is, in all likelihood, due to the change
in data sources.
Whatever the cause, the net effect of these two changes is that the data during the period
2000-2003 generally overwhelm our indicators, which either spike to their maximum values for
the entire period or go to zero. Because we frequently see this occurring for measures of both
conflict and cooperation, it seems more likely to be an artifact of the increased coverage of AFP.
The solution to this problem would be to use higher thresholds for the AFP data; we intend to
experiment with this adjustment at a later date.
Tit for Tat Analysis
The first analysis that we undertook was to look for the classical tit-for-tat behavior. Titfor-tit is one of the most widely studied patterns of political behavior, and in addition it has long
been known that reciprocity is one of the strongest patterns in event data (see for example Dixon
1986, Ward and Rajmaira 1992, Goldstein and Freeman 1992, Goldstein and Pevehouse 1996).
30

We were confident that we would find tit-for-tat patterns; the question was whether they would
occur in a meaningful fashion.
Figures 4 and 5 show the 32-day moving totals for incidences of conflictual and
cooperative tit-for-tat, which were compiled separately (i.e. conflictual TFT refers to a period of
material conflict by one side followed by a period of material conflict by the other). Several
characteristics are evident in these displays.
20
15
10
5
0
-5
-10
-15
-20
ISR
PAL
Figure 4. Tit-for-tat conflict
31

15
10
5
0
-5
-10
ISR
PAL
Figure 5. Tit-for-tat cooperation

First, the behaviors are generally, but not totally, symmetricgenerally when one side is
engaging in TFT, whether cooperative or conflictual, the other side is doing so as well. There is
no reason that this must be the case, but the fact that we observe it suggests that the two
antagonists are implementing a classical TFT solution to the prisoners dilemma game.
Unsurprisingly, give our qualitative understand of the conflict, they are far more likely to be
playing DD than CC.
Second, most of spikes in the conflictual TFT correspond to periods of substantial violence
such as the first and second intifadas and Israels 1982 invasion of Lebanon. Theoutbreak and
decline of the first intifada from December 1987 to August 1990 shows the same exponentialdecay shape that is seen in Goldstein-scaled data for the period (Schrodt and Gerner 1994).
Similarly, the negotiations following the Oslo agreement in September 1992 and prior to the
outbreak of the second intifada in September 2000 are evident.
The surprising aspect of these two graphs is the juxtaposition of TFT conflict and
cooperation during the post-Oslo period. This is not an error and is an illustration of the utility
of objective events patterns over vaguely remembered narratives: one tends to forget that while
the Oslo period saw nowhere near the levels of violence seen in the second intifada, there were
32

periods of substantial conflict, such as the four suicide bombings in Tel Aviv and Jerusalem and
subsequent Israeli reactions to these in the spring of 1996, shortly after Israels military
withdrawal from Palestinian urban areas. Conversely, negotiations have continued at both the
official and unofficial levels (e.g. the recently concluded Geneva Accords between Israeli and
Palestinian citizen elites) during the second intifada.
Olive Branch Rule
The second set of rules we looked at were the olive-branchinstances where one side
engaged in cooperation despite having experienced conflict from the other side. These instances
are shown in a 32-day moving total in Figure 6.
30
25
20
15
10
5
0
-5
-10
-15
-20
ISR
PAL
Figure 6. Olive branch pattern

The olive-branch pattern turns out to be far better than the cooperative TFT pattern at
delineating the Oslo period. Like cooperative TFT, one also sees olive-branches occurring
during the second intifada, but we believe that this is consistent with the narrative record. It is
also interesting to note that one sees a number of olive-branch instances following the outbreak
of the first intifada, and continuing in a sporadic pattern until the Oslo agreement. This would
33

appear to be consistent with changes in Israeli policy as they experimented with a variety
different levels of response to the intifada, and to international pressure for a resolution of the
dispute following the first Iraq war in 1991.
The olive-branch patterns show substantially less symmetry than was seen in the TFT
graphs, even in the post-Oslo period. Some of this may be a calibration issue: events with Israel
as an actor are consistently reported more frequently than events with Palestine as an actor.
However, this is unlikely to be the only explanation, given the symmetry for the TFT rules. A
more likely explanation is that the Palestinians, as the weaker party, are more likely to be the
follower than the leader in offering cooperation. This story is also consistent with the pattern
seen in the cooperative TFT graph prior to Oslo: there are a number of instance when the
Palestinians engaged in cooperative TFTthat is, cooperated following cooperative behavior by
Israelbut this was rarely reciprocated, as seen by the absence of spikes above the X-axis. This
might have led the Palestinians to get less inclined to initiate olive-branches, since their
experience is that unilateral cooperation would not be reciprocated. A final possible explanation
may simply be the fact that Israel has a major stronger and cohesive state structure and is
therefore better able to implement policy shifts.
An interesting example of the olive-branch pattern in the display can be seen in Figure A3
in the Appendix. We selected this because it was a frame that jumped out when the display as
a whole was being scannedsuddenly a long series of orange arrows (the olive-branch
indicator) appear on the Israeli side, with some limited reciprocity on the Palestinian side.
This appeared sufficiently unusual that we went back to the original source texts to see
what was going on. As the figure caption indicates, this was the period of U.S.-mediated
negotiations at the Wye Plantation conference center that attempted to get the Oslo process back
on track amid a deteriorating security situation. As the negotiation were presented in the press
reports, Israelunder U.S. pressurewas offering a number concession, while the Palestinian
negotiators were generally considering that the U.S. was taking a position that favored Israels
interests and did not consider these offers to be adequate implementation of Oslo. While the
display obviously does not provide all of that information, it does signal that something
interesting is going on, and in this instance, we were alerted to this by the display.
34

Meta-Rules
Our final set of experiments dealt with looking at the incidence of meta-rules. These
were implemented as four boxes of different colors at the right-hand edge of the display, and
were intended to detect both the level of escalation and de-escalation in the activies, as well as its
consistency. The four rules are
Red box
Black box
Purple box
Yellow box
Meta-rule 2: no cooperation in previous period; only one side using conflict

Meta-rule 1: no cooperation in previous period; both sides now using conflict
Meta-rule 3: no conflict in previous period; currently mixture of conflict and
cooperation
Meta-rule 4: no conflict in previous period; currently cooperation only
In essence, the red and black boxes represent escalation; the purple and yellow de-escalation. If
none of the meta-rules apply, the final column is left blank.
Making sense of the meta-rules has thus far proven more difficult than we anticipated, in
part because they fluctuate quickly. Graphs similar to those used for tit-for-tat and the olivebranch rules are not particularly informative, so instead we spent some time simply looking
through the entire display and trying to determine patterns using that highly sensitive pattern
recognition device, the human visual cortex. Using this method, two meta-patterns seem to be
evident.
First, to use Wolframs phrasing, purple grows. In the parts of display prior to the Oslo
Agreement in 1993, signals are usually umambiguous: there is either conflict or the absence of
conflict, and the meta-boxes are either black, red, or absent (cooperation is rare during this
period, as we saw in Figure 5). In the post-Oslo period, in contrast, purple becomes more and
more dominant, indicative of mixed conflict and cooperation.
This is, in all likelihood, a reflection of a critical change in the situation: the rise of Islamic
militant groups to challenge the legitimacy of the Palestinian Authority and the Palestine
Liberation Organization as the sole voices of the Palestinian opposition to Israel. A second
factor that may also be contributing to this is the increase in visible international mediation,
particularly by the United State and Europe, who know strongly encourage talks between Israel
and Palestinian representative even there is a high level of violence on the ground. Prior to Oslo,
when there was no mutual recognition between the two sides, talks (if present at all, and they
usually werent) were in secret or through intermediaries, whereas now both sides have to
respond to public calls for negotiations even in times of conflict.
35

The second meta-pattern that we have noted is that the majority of the meta-boxes are red:
following a period of no cooperation, conflict is initiated by one side or the other, rather than
simultaneously. Lest this seem obvious, one should note that this is quite a different pattern than
one sees in a conventional war, where the normative pattern would be that the two sides meet
on the field of battle and engage in conflict simultaneously. In the Israel-Palestine conflict, the
dominant is instead asymmetric conflict incidents, typically in the form of brief Israeli military
raids and even shorter small-scale Palestinian attacks such as violent demonstrations, attacks on
Israeli settlements, and (following 1996, when the tactic was introduced) suicide bombing.
While generally these occur in a tit-for-tat fashion, there is a significant time lag between the
stimulus and response and this therefore triggers a red meta-rule.
A couple of other patterns appear to suggest themselves, but we need to do further
systematic analysis to determine whether these are actually occurring at a level beyond that
expected by chance. First, it appears that in the pre-Oslo period, Israel tends to get the last
word in a period of extended conflictthat is, it is an Israeli action that triggers the last
occurrence of a red meta-box. Following Oslo, these shift over to the Palestinians. If this shift is
in fact real, it may be another manifestation of the decentralization of Palestinian militant activity
following Oslo.
The other general change between the pre- and post-Oslo patternand this involves an
overall assessment of the display, not just the meta-rulesis the increase of material cooperation
(typically, agreements) by the Palestinians. Prior to Oslo, cooperation by the Palestinians was
simply the absence of conflict; following Oslo we start to see cooperation events, even during
periods of conflict. Again, much of this can be explained by the fact of mutual recognition that
came with Osloprior to Oslo, the two sides had no public arena in which to cooperate. Israel,
as the occupying power, could engage in unilateral concessions to the Palestinians (for example
easing restrictions), but there could be no parallel official Palestinian response. This situation
changed following the Oslo agreement.
Conclusion
Event data have in employed in the analysis of international behavior for over four
decades, but arguably we are still trying to learn how to use them effectively. This type of
36

dataa nominal time seriesis rarely encountered in other fields such as economics and
medicine that make extensive use of statistical methods, and consequently there are relatively
few places from which we can borrow techniques. Instead we must invent our own.
This paper has been an initial foray into the realm of using discrete patterns similar to
those pioneered by Wolfram as a tool for event analysis. Our work here has been primarily a
descriptive validation of the potential of this approach: since no one had looked for patterns in
this fashion before, we first needed to demonstrate that we could find them, and that the patterns
had some plausible correspondence to our underlying qualitative understanding of the situation
we were analyzing.
One of our concerns when we embarked on the analysis was whether we would posit
plausible patterns and find nothing in the data. Our experience has, instead, been the opposite
the problem is not that we are finding too little, but we are still finding too much. When one
combines the remarkably rich set of patterns that can be constructed using the quite simple
methods aggregation methods available in the pattern-specification language with the ability to
rapidly construct colorful, web-based displays at a very fine time interval, it is difficult to figure
where to go next with the analysis. On the other hand, with a few exceptions, we are finding very
credible patterns in the patternsthese do not occur at random, but instead their rise and fall
generally tracks changes in the political situation which we know about from qualitative narratives.
In this conclusion, we will suggest three issues that we see most readily in need for further
research using this method. These are: calibration of the thresholds; the interaction of multiple
sets of rules; and the issue of developing abstract representations for rules.
Calibration
The time lags and event-count thresholds that we used in this analysis were arbitrary firstguesses based on our general familiarity with this data set. One calibration decision we
madethe use of different event count thresholds for Israel and Palestinewas clearly correct:
with the distinct thresholds, the patterns frequently track each other nicely, whereas if we had
used the same threshold for both actors we would have had severe problems with either overcounting or under-counting. However, as was evident in all of the figures, we have at least one
calibration problem: Reuters-based data versus AFP-based data. It is possible, however, that
there are others as wellfor example the period around 1982 and 1985 appear a bit too quiet.
37

Calibration will become an even greater issue if we start comparing rules across different
cases. While we have done all of this analysis on the Israel-Palestine case, there are about a
dozen different data sets available for analysis on the web site, and clearly one cannot use the
same thresholds for each of them due to the differences in the level of media coverage.
Ideally one would like to find some sort of scale-free patterns that would give similar
results for various different levels of event densities: in other words, patterns that would be
sensitive only to the relative frequency of events rather than to the absolute frequency. It may
also be possible to construct rules that can automatically compensate for lower levels of media
coverage by looking at longer periods of time. That is, if one has a region that receives, say,
only 10% of the media attention that the Israel-Palestine conflict receives, adjust for this by
aggregating over periods that are, in some sense, ten times longer.
The second issue in calibration is simply to see how the conflict appears at increasingly
higher thresholds, almost as though one was focusing a telescope or microscope to get an optimal
view of a specimen. When the threshold is set too low, a pattern will fit too easily, and the
resulting series will largely consist of noise. When the threshold is set too high, the pattern fits
too rarely and almost never occur. In between these two extremes the pattern fits just right and
provides the best analytical view of the situation.
Multiple rules and actors
The analysis we have done in this paper is at a very high level of aggregation. Israel
includes not just all parts of the Israeli governmentincluding the actions of opposition leaders
and partiesbut also non-governmental actors such as settler groups and citizen activists.
Palestine is even more diffuse, and encompasses over time the PLO, various militant groups
such as Hamas and Islamic Jihad, the quasi-governmental Palestinian Authority (after 1994), and
individual Palestinians.
Each of these groups may be operating according to rules, but they are not necessarily the
same rules. In some instances groups that we have included within a single actor are working
directly at cross-purposes. The on-going competition between the PLO/PA and Islamic militant
groups is probably the most important example in the data, but conflicts have also occurred
between the Israeli government and settler groups, and within Israeli governments. One can
38

extend this further to note that standard theories of bureaucratic behavior would suggest that the
operating of competing rule sets will be the norm rather than the exception in political behavior.
In the analysis we have presented here, we have not explicitly represented these alternative
rules, although we can see at least one cases where they were applied: the patterns of tit-for-tat
cooperation that occur simultaneously with the tit-for-tat conflict during the second intifada
period (Figure 5). Islamic military groups and the Israeli military are engaging in high levels of
conflict, while simultaneously one gets externally mediated negotiations between Israel and
individuals representing the Palestinian Authority and, at the very end of the series, the
independent Geneva Accord negotiations between non-governmental elites from both Israel
and Palestine.
A potential strength of the pattern-based approach would be the ability to explicitly model
these multiple agendas. That modeling could be further enhanced if we used the full capabilities
of automated coding to code sub-state actors.
A Language of Patterns
The final issue we wish to consider is the question of how patterns should be represented.
W have an abundance of choices, and only time will tell which is the most effective method.
Wolframs work is all done using a very simple graphical representation of rules: he
simply shows what pattern will replace each possible existing pattern. The rules aresomewhat
ironicallyimplemented in the mathematical programming system Mathematica, which was
originally designed to work with the continuous variable equations of the type that Wolfram now
seeks to replace as the foundation of analytical work in the natural sciences. This simple method
of representing rules was sufficient to keep Wolfram busy for ten years producing A New Kind of
Science; the fact that Wolfram developed Mathematica and owns the company of produces it
gave him the wherewithal to take ten years on the project.
When Hudson and Schrodt originally thought about this project, they envisioned
something similar to Wolframs rules. However, as they worked with Whitmer, who has done
extensive computer programming for Web-based applications, it was clear that because of the
stochastic aspects of event data and the lag times involved in political behavior, we would need a
more complicated representation. Whitmer, using the general-purpose programming language
Java, developed a pattern description language that provides only a small number of operators
39

for specifying patternsbasic arithmetic, time lags and intervals, and the minimum and
maximum functions. However, consistent with Wolframs core notion that complex systems can
be built from very simple components, these rules were quite adequate to develop a rich set of
patterns, rules, and meta-rules; in fact we have just begun to explore the possibilities with this.
But there are still [many] other alternatives. Early in the project, Schrodt decided to do a
quick run through the data looking for tit-for-tat patterns to make sure that these existed and had
a reasonable qualitative interpretation before we invested a lot of additional effort. Most of
Schrodts recent programming has used the text-processing language perl, which contains its
own very rich set of textual pattern recognition operators based on the regular expressions used
in Unix. To look for tit-for-tat using regular expressions, the event stream was first converted to
weekly strings of text where C and c represent cooperative events (WEIS <110) for Israel
and Palestine respective;D and d represent conflictual events (WEIS>=110). This reduces
the event stream to a series of lines that look like the following, where the date is in <>.
<871115>cC
<871122>DcDcCcD
<871129>DdCCCDDcd
<871206>DdDDDCDcCD
<871213>dDDDDDDDcdDD
<871220>DDDdDDDDcdDCD
<871227>cDCdddDD
<880103>DDdDDDDDCDDDDcDcDdDdDddDdDD
<880110>DCdCDddDDDDDDdcDDdDcdCcDdDDddCDCddDdDDDDDDDDCcdcd
<880117>cccCCDcCcDDDdddDDCddDdddDD
<880124>dCDDCDdDCdDdDDDDdCdDDD
<880131>dDDDDcDDDcdDDCCdDccdDcDcdDDDDDcCDD
<880207>ddDDDDDDDCDcCcCcddDDDDCCcDDcC
<880214>CccCDCDDDcCDCcDDDDDDDD
Using this as the input, tit-for-tat on a weekly basis can be found using a short perl
program where the core code is:
if (($last =~ m/(C[^C]*){$thres,}/g) && ($line =~ m/(c[^c]*){$thres,}/g))

{$ic = 1;} else {$ic = 0};
40

if (($last =~ m/(D[^D]*){$thres,}/g) && ($line =~ m/(d[^d]*){$thres,}/g))
{$id = 2;} else {$id = 0};
if (($last =~ m/(c[^c]*){$thres,}/g) && ($line =~ m/(C[^C]*){$thres,}/g))
{$iC = 3;} else {$iC = 0};
if (($last =~ m/(d[^d]*){$thres,}/g) && ($line =~ m/(D[^D]*){$thres,}/g))
{$iD = 4;} else {$iD = 0};
if ($iC || $iD) {$iISR = 2;} else {$iISR = 0};
if ($ic || $ic) {$iPAL = 1;} else {$iPAL = 0};
In this program fragment, $last is the string from the previous week; $line is the string from
the current week; and $thres is the minimum threshold for the number of events. The variables
$iISR and $iPAL
take the value zero if the actors are not using tit-for-tat and non-zero-values if
they are. When these variables are read into Excel, this produces a chart such as the one in
Figure 7.
ISR TFT
PAL TFT
Figure 7. Weekly tit-for-tat behavior computed using a perl regular expressions

So, one year into this project and we have at least three very different ways of representing
patterns: Wolframs graphs, Whitmers set of arithmetic operators, and Unix/perl regular
expressions. All three methods are very abstract; all three methods are capable of expressing a
wide variety of rules.
41

Which method is best?we dont know, and at this point we dont want to know, since
we are still in the early stages of investigating this mode of research. Our point is simply that we
need to develop a language (or languages) for expressing patterns, much as mathematicians in
the 16th and 17th centuries found ways to standardize the notation and operations of classical
continuous variable mathematics. The power of any notation comes from its ability to express
complex ideas in a compact form, and the ability of the user to quickly read idiomatic
expressions in that notation. We expect (sometimes in vain...) that our students can recognize
that the statement
4x = 2x - 2
can be solved to give a unique value for x. With a bit of coaching, we can get them familiar
with the fact that
y = ax + b
is a line, and then graduallyprovided the student has sufficient background in algebra and
calculusintroduce idioms such as the calculation of a variance from a sum and sum-of-squares,
interpretation of the logit equation, and eventually derivation of the ordinary least squares
estimator in matrix notation.
However, we are able to do this with suitably trained and motivated students because 99% (in
fact usually 100%) of the notation was standardized outside of the field of political science: a
student who learns matrix algebra in the mathematics department or economics departmentor
for that matter learns it in China before attending graduate school in the U.S.finds no changes
when applying that notation to the analysis of political behavior. This is not the situation with
abstract pattern recognition.
Computer languages such as perl may, in fact, be the first step in such a standardization.
One of Perl's key features as a language is regular expressions; in fact, Perl has probably
done more to evolve regular expressions than any other language. If you are not familiar
with regular expressions, think of them as the ultimate string manipulation tool for
serious string processing. Regular expressions are to strings what math is to numbers.
Andrew Clinick, Program Manager, Microsoft Corporation, January 22, 1999
(http://msdn.microsoft.com/workshop/languages/clinic/scripting012299.asp; accessed 18
December 2000)
42

However, the utility of perl depends on ones ability to read the language idiomatically,
and until one has had considerable experience with abstract syntax of regular expressions, perl is
anything but transparent. Whitmers arithmetic rules have the advantage of being built from
simple operations that any student who remembers middle-school algebra should be able to
understand. However, even though the operations are simple, figuring out their combined
effects, and learning the idiomatic tricks that one uses to get the arithmetic expressions to
effectively describe patterns that fundamentally are not arithmetic patterns is still a complex
process. We undoubtedly still have a ways to go on this issue.
43

Bibliography
Abelson, Robert P. 1973. "The Structure of Belief Systems." In Computer Models of Thought
and Language, edited by R.C. Schank and K.M. Colby. San Francisco: Freeman.
Allison, Graham T. 1971. The Essence of Decision. Boston: Little, Brown.
Anderson, John R. 1983. The Architecture of Cognition. Cambridge: Harvard University Press.
Anderson, James A. and Edward Rosenfeld. eds. 1988. Neurocomputing: Foundations of
Research. Cambridge: MIT Press.
Andriole, Stephen J. and Gerald W. Hopple. 1988. Defense Applications of Artificial
Intelligence. Lexington MA: Lexington.
Axelrod, Robert. 1984. The Evolution of Cooperation. New York: Basic Books
Bickerton, Ian J. and Carla L. Klausner. 1998. A Concise History of the Arab-Israeli Conflict.
(3rd edition) Upper Saddle River, NJ: Prentice-Hall.
Bond, Doug, J. Craig Jenkins, Charles L. Taylor, and Kurt Schock. 1997. Mapping Mass
Political Conflict and Civil Society: The Automated Development of Event Data. Journal of
Conflict Resolution 41: 553-579.
Boynton, G. R. 1991. "The Expertise of the Senate Foreign Relations Committee." In Artificial
Intelligence and International Politics, edited by Valerie Hudson. Boulder: Westview.
Boynton, G. R. n.d. "Interpretive Triples and Reasoning About Foreign Policy." Photocopied
manuscript.
Bull, Hedley. 1966. "International Theory: The Case for a Classical Approach." World Politics
18:361-377.
Carbonell, Jaime G. 1978. "POLITICS: Automated Ideological Reasoning." Cognitive Science
2:27-51.
Chase, W.G. and H.A. Simon. 1973. "Perception in Chess." Cognitive Psychology 4:55-81.
Collins, Allan and Edward E. Smith, eds. 1988. Readings in Cognitive Science: A Perspectice
from Psychology and Artificial Intelligence. San Mateo CA: Morgan Kaufmann.
Dixon, William J. 1986. "Reciprocity in United States-Soviet Relations: Multiple Symmetry or
Issue Linkage." American Journal of Political Science 30:421-45.
Fullbrook, Edward, "A Brief History of the Post-Autistic Economics Movement," November 21,
2001, http://www.paecon.net. Accessed 14 March 2004
Gause, F. Gregory. 1999. "Systemic Approaches to Middle East International Relations."
International Studies Review1,1:11-32.
Gerner, Deborah J. 1994. One Land, Two Peoples: The Conflict over Palestine. (2nd ed.)
Boulder, CO: Westview Press.
44

Gerner, Deborah J., Philip A. Schrodt, Ronald A. Francisco, and Judith L. Weddle. 1994. The
Machine Coding of Events from Regional and International Sources. International Studies
Quarterly 38: 91-119.
Gilovich, Thomas. 1981. "Seeing the Past in the Present: The Effect of Associations to Familiar
Events on Judgments and Decisions." Journal of Personality and Social Psychology 40:797808.
Goldstein, Joshua S., and John R. Freeman. 1990. Three-Way Street: Strategic Reciprocity in
World Politics. Chicago: University of Chicago Press.
Goldstein, Joshua S., and Jon C. Pevehouse. 1996. "Reciprocity, Bullying and International
Cooperation: Time-Series Analysis of the Bosnia Conflict." International Studies
Association, San Diego.
Goldstein, Joshua S., Jon C. Pevehouse, Deborah J. Gerner, and Shibley Telhami. 2001.
Dynamics of Middle East Conflict and US Influence. Journal of Conflict Resolution 45, 5:
594-620.
Green, Donald and Ian Shapiro, 1995. Pathologies of Rational Choice Theory: A Critique of
Applications in Political Science. New Haven, CT: Yale University Press, 1995
Hall, Stephen S. 1992. "How Technique is Changing Science." Science 257:344-349.
Hofstader, Douglas. 1985. Metamagical Themas. New York: Basic Books.
Hogarth, Robin M. and Melvin W. Reder, eds. 1987. Rational Choice: The Contrast between
Economics and Psychology. Chicago: University of Chicago Press.
Holland, John H. 1975. Adaptation in Natural and Artificial Systems. Ann Arbor: University of
Michigan Press.
Holland, John H. 1986. "Escaping Brittleness: The Possibilities of General Purpose Algorithms
Applied to Parallel Rule-Based Systems." In Machine Learning 2, edited by R.S. Michelski,
J.G. Carbonell and T.M. Mitchell. Los Altos,CA: Morgan Kaufman.
Holland, John H., Keith J. Holyoak, Richard E. Nisbett and Paul R. Thagard. 1986. Induction:
Processes of Inference, Learning and Discovery. Cambridge: MIT Press.
Huxtable, Phillip A., and Jon C. Pevehouse. 1996. Potential Validity Problems in Events Data
Collection. International Studies Notes 21, 2: 8-19.
Jackendoff, Ray. 1987. Consciousness and the Computational Mind. Cambridge: MIT Press.
Jervis, Robert. 1976. Perception and Misperception in International Politics. Princeton:
Princeton University Press.
Kasza, Gregory. n.d. Perestroika: For an Ecumenical Science of Politics. undated,
http://www.paecon.net. Accessed 14 March 2004
Keohane, Robert O. 1988. International Institutions: Two Approaches. International Studies
Quarterly 32:379-396.
Khong, Yuen Foong. 1992. Analogies at War. Princeton: Princeton University Press.
45

Kissinger, Henry. 1979. The White House Years . Boston: Little, Brown.
Kohonen, T. 1984. Self-Organization and Associative Memory. New York: Springer-Verlag.
Krasner, Stephen D., ed. 1983. International Regimes. Ithaca: Cornell University Press.
Kuhn, Thomas S. 1970. The Structure of Scientific Revolutions (2nd ed.) Chicago: University of
Chicago Press.
Lashley, Karl. 1956. "Cerebral organization and behavior." In The Brain and Human Behavior,
edited by H. Solomon, S. Cobb and W. Penfield. Baltimore: Williams and Wilkins.
Margolis, Howard. 1987. Patterns, Thinking and Cognition: A Theory of Judgement. Chicago:
University of Chicago Press.
Majeski, Stephen J. 1987. "A Recommendation Model of War Initiation: The Plausibility and
Generalizability of General Cultural Rules." In Artificial Intelligence and National Security,
edited by Stephen Cimbala. Lexington, MA: Lexington Books.
Masters, Roger D. 1989. The Nature of Politics. New Haven: Yale University Press.
Mefford, Dwain. 1991. "Steps Toward Artificial Intelligence: Rule-Based, Case-Based and
Explanation-Based Models of Politics." In Artificial Intelligence and International Politics,
edited by Valerie Hudson. Boulder: Westview.
Newell, Allan and Herbert Simon. 1972. Human Problem Solving. Englewood Cliffs, NJ,
Prentice-Hall.
Political Psychology. 2003. Special Issue: Neuroscientific Contributions to Political
Psychology. Political Psychology 24,4.
PS. 2003. Special issue: Methodological Pluralism. PS 36,2.
Purkitt, Helen E. 1991. Artificial Intelligence and Intuitive Foreign Policy Makers Viewed as
Limited Information Processors. In Artificial Intelligence and International Politics, edited
by Valerie Hudson. Boulder: Westview.
Reber, Arthur S. 1993. Implicit Learning and Tacit Knowledge. New York: Oxford University
Press.
Schrodt, Philip A., and Deborah J. Gerner. 1994. Validity assessment of a machine-coded event
data set for the Middle East, 1982-1992. American Journal of Political Science 38: 825-854.
Schrodt, Philip A., Deborah J. Gerner, and Erin M. Simpson. 2001. Monitoring conflict using
automated coding of newswire sources: a comparison of five geographical regions. Paper
presented at the PRIO/Uppsala University/DECRG High-Level Scientific Conference on
Identifying Wars: Systematic Conflict Research and Its Utility in Conflict Resolution and
Prevention, Uppsala. June 2001.
Sniderman, Paul M., Richard A. Brody and Philip E. Tetlock. 1991. Reasoning and Choice.
Cambridge: Cambridge University Press.
Sylvan, David J. and Stephen J. Majeski. 1985. "Tacit Cultural Rules in Foreign Policy." Paper
presented at the Merriam Research Conference, University of Illinois, October.
46

Sylvan, David and Stephen Majeski. 1986. "Problems of Contextual Specificity in
Computational Models of Foreign Policy." Paper presented at the International Studies
Association, Anaheim.
Sylvan, David, Stephen Majeski and Jennifer L. Milliken. 1991. "Theoretical Categories and
Data Construction in Computational Models of Foreign Policy." In Artificial Intelligence and
International Politics, edited by Valerie Hudson. Boulder: Westview.
Sylvan, Donald A., Ashok Goel and B. Chandrasekran. 1990. "Analyzing Political Decision
Making from an Information Processing Perspective: JESSE." American Journal of Political
Science 34:74-123.
Sylvan, Donald A. and Steve Chan, 1984. Foreign Policy Decision Making: Perception,
Cognition and Artificial Intelligence. New York: Praeger.
Simon, Herbert A. 1982. The Sciences of the Artificial. Cambridge: MIT Press
Tessler, Mark. 1994. A History of the Israeli-Palestinian Conflict. Bloomington, IN: Indiana
University Press.
Thorson, Stuart and Donald A. Sylvan. 1982. "Counterfactuals and the Cuban Missile Crisis."
International Studies Quarterly 26:537-71.
Vertzberger, Yaacov Y.I. 1990. The World in their Minds: Information Processing, Cognition
and Perception in Foreign Policy Decision Making. Stanford: Stanford University Press.
Ward, Michael and Sheen Rajmaira. 1992. "Reciprocity and Norms in U.S.-Soviet Foreign
Policy." Journal of Conflict Resolution 36,2: 342-368.
Walt, Stephen M. 1999. "Rigor or Rigor Mortis? Rational Choice and Security Studies,"
International Security 23, 4: 5-48
Weinberg, Gerald M. 1975. An Introduction to General Systems Thinking. New York: John
Wiley and Sons.
Wolfram, Stephen. 2002. A New Kind of Science., Champaign, IL: Wolfram Media
Yee, Albert S. 1996 "The Effects of Ideas on Policies" International Organizations 50, 1: 69-108
47

Appendix 1: Instructions for Using the JHAX Event Pattern Web Site
Created by Ray Whitmer
Last revision: 2003/12/18 19:48:36 GMT
1. Introduction
The JHAX Event Prediction Website http://ep.jhax.org.is a web-based program
that manipulates and displays data about world events. This page focuses on the technical
aspects of the program, ignoring many significant things such as qualities of the data and the
search for rules that might usefully predict events, etc.
2. Source Code
The Java Programs and Java Server Programs on these pages were written by Ray
Whitmer and have been released under the GPL, meaning roughly that anyone can use them
freely (both in terms of source rights and cost) as long as they are not redistributed as part of a
non-free program. The sources may be viewed or downloaded on the internet without
registration of any form.
When you follow the Java Programs link above, you see a list of several files that contain
some of the main parts of the code. If you click on the version number, you will that version of
the file. From either of these links, you can click on "download" to get an exact copy of the file.
Clicking on the file name takes you to a history of the file.
EventDisplay is the main object that constitutes the event display of the program. It knows
what all the program parameters are, how to get them and send them, and how to use them to
create a textual or graphical display. Event is a simple object representing a single event that
may be read from a file. EventTypePattern assigns a name to patterns of these event types, each
interfacing to the scheduling. Servlet causes a virtual graphic to be drawn from a special
graphics URL referenced by pages where a graphical display has been requested. Mostly this
just calls back into the EventDisplay class. Displayable controls drawing each symbol for the
graphical display.
The JHAX.org main source page has links to sources and documentation of various
systems.
48

3. The Page is the Program

The main page of the web site presents a set of field values that control the program, filled
in with working default values. This page describes how to modify the values to get exactly the
desired output. When the form values have been modified, pressing any button at the bottom of
the form causes the values to be incorporated into the URL shown at the top of the browser, so
the address of the page with the settings built-in can be easily bookmarked or emailed.
3.1 Data Source
The data source field allows selection of which file of data to filter and display. Currently
only one may be selected at a time. Each data source is displayed by the name of the file as
unarchived from http://www.ukans.edu/~keds/data.html. See that website information on the
data sets, and to see what each contains. When using a data set with the program, it may
important to know when the data set starts and ends, which actors are represented in the data, and
so on to produce a non-empty display.
3.2 Actors' Map
The actors' map field identifies all actors in the data that will be uniquely represented on
the display, as defined further on in the display description. This field value might typically map
source actors to A and B, for dyadic interactions between two parties. Event data often has far
more actors represented than can be well-represented seperately in the display. Mapping specific
actors to generic names (letters of the alphabet) also permits the same patterns and display
description to be more-easily reused with different actors.
The actors' map field may be changed to map different primary actors into the existing
scheme, to add more specific mappings into the dyadic scheme, or to go beyond the dyadic
scheme (for example using further letters of the alphabet for n-way event displays). All
otherwise-unmapped actors are mapped to UNK, which can also be referenced in the patterns or
display.
The actors' map value is a list of comma-seperated mappings. Each mapping has a source
followed by equals and a target name, as follows:
<source actor>[*] = <target actor>[, ...]
49

Like most parameters of this program, the source and target actors are case-sensitive, so
the display description must be coordinated using the proper case.
The source actor may end in an asterisk to map all actors that start with that source. Where
multiple mappings match the same source actor, the first matching declaration takes precedence.
For example, the in the mapping list FO*:A,BA*:B,BAR*:A, the mapping BAR*:A does not
map anything since BA*:B already included everything starting with BAR*, but in the example
FOO*:B,FO*:A,BAR*:A,BA*:B, the mappings FOO*:B and BAR*:A make exceptions to
FO*:A and BA*:B respectively since they occur first.
3.3 Event Codes
The event codes field maps the numeric codes of source events to a usually-smaller set of
event types to be presented in the display.The comma-seperated values in this list alternate target
verbs and the source integer codes in ascending order starting with the target for negative infinity
and concluding with the target for positive infinity, as follows:
<code>[, <verb>, <code>[...]]
For example the list previous,10,foo,100,bar,500,unknown maps all values below 10 to
previous, values from 10 to 99 inclusive to foo, values from 100 to 499 inclusive to bar, and
values greater than or equal to 500 to unknown. The values previous, foo, bar, and unknown will
be presentable in this case in the display description. To add a distinct verb for code 200, the list
becomes
previous,10,foo,100,bar,200,distinct,201,bar,500,unknown.
Note that the bar mapping has to be reinstated at 201 or the distinct mapping would have
been in effect until 500.
3.4 Days per Unit
The days per unit field contains an integer that controls, among other things, how many
days of events will be displayed on each line of the display. When the number of days per row is
changed, this also affects patterns that all reference time units explicitly or implicitly. With 2
days per unit a sum of 10 units duration will sum events from 20 days, with a week per unit this
becomes 10 weeks, and with a day per unit, this becomes 10 days, etc.
50

Although the default setting is 2, a better choice for the setting may be 1, because it does
not degrade the acuracy of the data source and interprets durations more-simply as days. This
would double the size of the output and represent fewer events per row, which may also be a
good thing.
3.5 Patterns
The patterns text box allows assignment of names to counter expressions, one per line.
Assigned pattern names may be used in other patterns or in the display. Each line has a pattern
identifier followed by the counter expression, as follows:
<pattern identifier> = <counter expression>
Counter expressions are described in section 4.
3.6 Display
The display text box specifies how specific events will be represented on the display. Each
line of text consists of four comma-seperated values, as follows:
<column number>, <symbol shape>, <symbol color>, <counter expression>
A description follows of these fields except for the counter expression, which is described
in section 4.
3.6.1 Column Number
The column number specifies the column of the graphical output to which the line outputs
symbols (the column's width, symbol size, offsets, etc. are governed by the previously-described
fields). If two lines in the display description output symbols to the same column, the symbols
come out in the order that the lines appeared in the display description. Referencing higher
column numbers causes the the graphical output to become wider to include the specified
column.
0 is interpreted as the number of an invisible column, which can be used to temporarily
remove something from the graphical output. Positive column numbers produce output of
symbols in the respective columns starting at the left side. Negative column numbers output to
the same columns as the corresponding positive numbers, but starting at the right side of the
51

column and proceeding to the left, possibly colliding with output to the positive column number,
but with reverse offset to avoid collisions as long as possible.
3.6.2 Symbol Shape
The symbol shape specifies the program-supported shape that is to be drawn to symbolize
the particular output. The program originally supports BoxFill, PointFill, BoxOutline, and
PointOutline, which correspond to boxes or arrow-points that be filled or drawn as an outline.
The arrow points always have their tip in the direction of progression, meaning that outputting to
a negative column number makes them point left and outputting to a positive column number
makes them point right. Support for additional symbol shapes will likely be added at some time
in the future.
3.6.3 Symbol Color
The symbol color specifies which color to use to paint the symbol using an 8-digit
hexadecimal (base 16 with digits 0123456789abcdef) number which specifies, two digits at a
time the value of the alpha, red, green, and blue components of the color (00,01,02,...fd,fe,ff).
The red, green, and blue components specify how much of the respective colors are mixed
to produce the corresponding color simulation in humans (dogs and partially-colorblind humans
detect two color components, shrimp detect twelve, and publishers may only detect one or less,
making the system biased towards normal humans).
The alpha component tells how opaque the color is.
For example ff000000 is opaque black (red green and blue all 0), 00000000 is transparent
black (totally invisible), 80000000 is half-transparent black (things show through 50%),
e00000ff is an eighth transparent blue (things show through 12.5% with non-zero blue),
e0808080 is an eighth transparent gray (things show through 75% with red green and blue
equal), etc.
3.6.4 Counter Expression
The counter expression describes the counter which specifies how many of the symbols to
draw on a particular row of the display. See the section on counter expressions.
52

3.7 Starting and Ending Date
Starting or ending date fields may be used to limit (or even artificially extend) the display.
This is especially useful to spare the browser (and the web server) the load of redisplaying the
entire data set graphically. A browser which is in the process of downloading too much data
might also be stopped by pressing the stop button or escape key (or closing and restarting the
browser). If left unspecified, the starting or ending date of the data source will be used.
A properly-specified date for this program consists of 8 digits which are run together with
no seperating spaces or punctuation. The first four digits are the year number, the next two the
month, and the last two are the day, as follows:
<year digit><year digit><year digit><year digit><month digit><month digit>
<day digit><day digit>
Anything else (except leaving the field blank to specify no date) should cause an error.
3.8 Symbol Size
The symbol size field tells the program how many dots high and wide in the resulting
display of a symbol should be in the graphical output. If specified as a single number, the same
value is used for width and height, but two comma-seperated values may be given to
independently specify width and height for less-symmetrical sizes as follows:
<symbol width>, <symbol height>
For example, 6,12 specifies symbols that are twice as high as they are wide.
Changing the symbol size is one easy way to change the width and length of the resulting
page since symbol size directly contributes to page size.
3.9 Pixel Offset
The pixel offset field tells the program how many dots to shift when offsetting symbols to
avoid total occlusion of different symbols in the graphical display. This effects the total width of
the display, because when space is reserved for symbols in the display, each possible offset
reserves this much space horizontally and vertically. If specified as a single number, the same
value is used horizontally and vertically, but two comma-seperated values may be given to
independently specify the horizontal and vertical pixel offsets, as follows:
53

<horizontal pixel offset>, <vertical pixel offset>
These values may also be specified as negative numbers to cause the offsetting to occur on
the opposite side of the symbol, i.e. the value 0,-3 eliminates horizontal offsetting and makes the
vertical offsetting occur by three pixels at each offset in the opposite direction from default
offsetting. The offsetting is also reversed when dealing with left-justified versus right-justified
columns in order to make total occlusion of symbols from opposite sides of the column less
likely.
3.10 Dimensions
The dimensions field contains three other comma-seperated values that are used in the
layout of the graphical output, as follows:
<column width>, <height>, <offsets per row>
Column width is the number of symbols that may be placed in each column. As with
symbol size and pixel offset, this directly effects the total width of the graphical output.
Height is the number of rows of symbols that may be placed in a single graphic before
exhausing the space. This effects the total height of each graphic and inversely effects the total
number of graphics that will be required to present the requested output.
Offsets per row is the number of offsets that can be distinguished in a particular row before
total occlusion occurs. Together with pixel offset, this controls how much extra space is
reserved for each symbol to allow offsetting, also effecting the width and height of the graphical
output.
3.11 Image Type
The image type field specifies which browser-supported image format to use, Jpeg or
PNG. The best choice is according to what your browser supports. While JPeg is slightly older
and may be supported in more browsers, PNG is more-suitable to the task because Jpeg is
designed to compresses pictures containing continuously-varying shades and loses things in the
process, which means that if you choose Jpeg, the graphics will take longer to download and the
symbols will be fuzzier in color and position.
54

GIF is an older format that deals better with non-continuously-shaded graphics such as the
output of this program, but after the format became popular, Unisys made it legally risky to use
the format in a program, which is why the World Wide Web produced PNG (Portable Network
Graphics) an unencumbered and more-advanced replacement that is supported by later versions
of standards-compliant browsers.
3.12 None, Graphic, and Text Buttons
Pressing one of these buttons makes a request of the server to incorporate the current form
fields into the URL (so they can be bookmarked), adding the specified display of events to the
page.
n If none is pressed, any graphic or text display of events is removed, which may be
useful for bookmarking the undisplayed form.
n If graphic is pressed, the graphical output of the program is displayed.
n If text is pressed, a textual version of the output of the program is displayed.
4. Counter Expressions
A counter expression computes some number usually based on event counts. This can be
an event type, a pattern name, a number, or an expression.
4.1 Event Type
When as a counter expression, the event type refers to the number of matching events in
the current unit of time.
An event type is a source actor, followed by dash, followed by a target actor, followed by
dash, followed by the verb, as follows:
<source actor>-<target actor>-<verb>
The event type is aways interpreted after mapping the actors and codes, so the source and
target actors must each be targets in the actor map field and the verb must be from the event
codes field. If, for example, the actor map produces source and target actors A and B and the
code maps to the target verb MaterialCooperation, then the corresponding event type to refer to
the count of such events is A-B-MaterialCooperation.
55

4.2 Pattern
A pattern is a name (with no dashes) that refers to the counter expression of a previouslydeclared pattern.
4.3 Number
A number is an integer constant that is returned instead of a counter. This is most-useful
as part of a more-complex counter expression, since it is boring to display the same number of
symbols on every row of the display. Asking for the sum of a constant over a time range will
return the constant once for every time unit, because it behaves like other expressions during
ranges, but as a constant value rather than as the result of counting something.
4.4 Expression
A counter expression describes how to combine event counts in the data set to determine
how many symbols to output for identifying interesting patterns. The expression is represented
as an operation type (sum, min, or max), followed by parentheses encloding one or more commaseperated constants, event types, or nested counter expressions that describe what is to be
summed, minimized, or maximized. Optional brackets may follow the parentheses-enclosed
values, allowing the operation to gather data from before the current time being presented.
Inside of the brackets is a non-negative integer specifying how many units before the current
time the operation begins. A colon and second number may follow the first number specifying
in how many time intervals the data should be processed. An optional slash and divide value
may follow the operation and optional bracketed numbers.
Note that in the random sampling of examples follow, with descriptions of the
interpretation by the program, previously-defined patterns can be used instead of event types.
sum(A-B-MaterialCooperation,A-B-VerbalCooperation)
The sum of material and verbal cooperations from A to B at the current time.
sum(A-B-MaterialCooperation,A-B-VerbalCooperation)[5]
The sum of material and verbal cooperations from A to B five units of time before the
current time.
56

sum(A-B-MaterialCooperation,A-B-VerbalCooperation)[0:10]
The sum of material and verbal cooperations from A to B during the ten units of time
ending at the current time.
sum(A-B-MaterialCooperation,A-B-VerbalCooperation)[5:10]/3
The sum of material and verbal cooperations from A to B during the ten units of time
ending five units of time before the current time, divided by 3.
min(A-B-MaterialCooperation,A-B-MaterialConflict)
The lessor of the material cooperation or the material conflict from A to B at the current
time.
min(A-B-MaterialCooperation,A-B-MaterialConflict)[0:50]
The least count material cooperation or material conflict from A to B in the 50 days ending
at the current time.
min(sum(A-B-MaterialCooperation,A-B-MaterialConflict))[0:50]
The least of the sums of material cooperation and material conflict in each day from A to B
in the 50 days ending at the current time.
min(sum(A-B-MaterialCooperation,A-B-MaterialConflict)[3:5])[0:50]
The least of the 3-day-old 5-day sums of material cooperation and material conflict from A
to B in the 50 days ending at the current time.
min(sum(A-B-MaterialCooperation,A-B-MaterialConflict)[0:5])[3:50]
The same interpretation as the previous example.
min(max(sum(A-B-MaterialCooperation)[0:1])[0:1])[0:1]
Since minimums, maximums, and sums of a single value are that value and an age of 0 and
duration of 1 is just default interpretation in the current time, this expression should be simplified
to the unadorned event type with no additional expression, i.e. A-B-MaterialCooperation.
min(A-B-MaterialCooperations)[5]/2
57

Which operator (min, max, or sum) was specified is irrelevant since it applies to a single
event type and time unit, but some operator was required so that dividing and aging could be
specified.
Note that the division makes little sense in this sort of case because the counter is at most 1.
min(A-B-MaterialCooperations)[0:5]/2
The operator is now relevant, because minimization is applied over a period of 5 units of
time, the result of which is divided by 2.
min(1,sum(A-B-MaterialCooperations,B-A-MaterialCooperations)[0:5]/10)
Count the number of material cooperations between A and B in either direction over the
latest 5 days, scaling by a tenth so that a symbol for 10 or more, but never display more than 1
symbol.
sum(-1,min(1,sum(A-B-MaterialConflict)[0:5]/10),min(1,sum(B-AMaterialConflict)[0:5]/4)[5])
Return 1 symbol whenever at least 4 material conflicts from B to A spread over at most 5
units of time is followed by at least 10 material conflicts from A to B spread over at most 5 units
of time.
Note that the -1 sum of this expression could produce a negative result, but the display
output treats negatives the same as zero, so it was not necessary to surround the expression with
max(0,<expression>), which could have prevented negative results.
58

Examples of the display:

The following figures give examples of the display produced by the web site. The symbols are:
Column 1:
Green box
Green arrow
Palestine cooperation (empty for verbal; solid for material)

Palestine conflict (empty for verbal; solid for material)
Column 2:
Blue box
Blue arrow
Israel cooperation (empty for verbal; solid for material)

Israel conflict (empty for verbal; solid for material)
Column 3:
Green arrow
Red solid box
Red empty box
Yellow solid box
Yellow empty box
Palestine conflict above threshold

Palestine tit-for-tat conflict
Palestine unilateral conflict
Palestine tit-for-tat cooperation
Palestine unilateral cooperation
Column 4:
Blue arrow
Red solid box
Red empty box
Yellow solid box
Yellow empty box
Israel conflict above threshold

Israel tit-for-tat conflict
Israel unilateral conflict
Israel tit-for-tat cooperation
Israel unilateral cooperation
Column 5:
Black box
Red box
Purple box
Yellow box
Meta-rule 1: no cooperation in previous period; both sides now using conflict

Meta-rule 2: no cooperation in previous period; only one side using conflict
Meta-rule 3: no conflict in previous period; currently mixture of conflict and
cooperation
Meta-rule 4: no conflict in previous period; currently cooperation only
59

Figure A1. 7-December-1987 to 8-March-1988: Outbreak of first intifada
60

Figure A2. 25-March-1994 to 25-June-1994: Period between signing of Oslo Agreement

and Rabin assassination
61

Figure A3. 6-October-1998 to 6-January-1999: U.S.-mediated Wye Plantation negotiations
62

Figure A4. 11-July-2000 to 11-October-2000: Outbreak of second intifada
63

A New Kind of Social Science: The Path Beyond Current (IR) Methodologies May Lie Beneath Them

Uploaded by

Copyright:

Available Formats

A New Kind of Social Science: The Path Beyond Current (IR) Methodologies May Lie Beneath Them

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A New Kind of Social Science: The Path Beyond Current (IR) Methodologies May Lie Beneath Them

Uploaded by

Copyright:

Available Formats

A New Kind of Social Science:

2004 by the authors

A New Kind of Social Science

The world is a stupendous machine, composed of innumerable parts, each of which

A New Kind of Social Science

A New Kind of Social Science

The Issue of Complexity as the Root of Discontent

A New Kind of Social Science

A New Kind of Social Science

A New Kind of Social Science

Pattern Recognition, Human Understanding, and Human Action

A New Kind of Social Science

A New Kind of Social Science

Earlier known as "short term memory".

A New Kind of Social Science

A New Kind of Social Science

question for the U-2 crisis.

A New Kind of Social Science

A New Kind of Social Science

A New Kind of Social Science

Political Behavior as Rule-Based

A New Kind of Social Science

A New Kind of Social Science

A New Kind of Social Science

Agency, Computational Irreducibility, and the Question of Causality

A New Kind of Social Science

A New Kind of Social Science

A New Kind of Social Science

A New Kind of Social Science

Seeing Human Behavior

A New Kind of Social Science

Source: Wolfram, pg. 69

Figure 2: Examples of general categories of patterns

A New Kind of Social Science

A New Kind of Social Science

Source: Wolfram, pg. 107

Figure 3: Wolframs example of the diversity possible with simple rules

A New Kind of Social Science

Why Events Data? Why the Israelis and Palestinians?

A New Kind of Social Science

A Way of Seeing: Introducing the Website

A New Kind of Social Science

A New Kind of Social Science

Rule Specification and Visual Inspection in this Probe

A New Kind of Social Science

A New Kind of Social Science

Patterns and Meta-Patterns in the Israeli-Palestinian Dyad:

A New Kind of Social Science

A New Kind of Social Science

Figure 4. Tit-for-tat conflict

A New Kind of Social Science

Figure 5. Tit-for-tat cooperation

A New Kind of Social Science

Figure 6. Olive branch pattern

A New Kind of Social Science

A New Kind of Social Science

Meta-rule 2: no cooperation in previous period; only one side using conflict

A New Kind of Social Science

A New Kind of Social Science

if (($last =~ m/(C[^C]){$thres,}/g) && ($line =~ m/(c[^c]){$thres,}/g))