“Two Paths to a Science of Politics” presents two
strongly argued points of view about where we should be going
in our discipline. The authors by and large agree on the nature
of science and the variety of ways to do it, but they disagree on the
most fruitful way to proceed at this moment. Granato and Scioli believe
that we need to do more to test theories with data and to formulate
theories that are based on data. We need, in their phrase, to consider
the empirical implications of theoretical models (EITM). Rogers Smith
thinks we need to take history, context, and meaning more seriously in
order to understand phenomena such as political identity.
To get some perspective on these two approaches to political science,
it is useful to think about the nature of scientific argument, and it
is very useful to move beyond using physics (especially physics as
reconstructed by the logical positivists) as the model for science.
Biology, I believe, provides a better model for social science, partly
for the very reasons that logical positivists and their allies had
difficulty accepting it as a science. Once we develop a more subtle
conception of science that avoids the snare of trying to imitate
physics, it turns out that linking theories to data almost inevitably
leads to more concern with history, context, and meaning. So, although
these authors take different routes, their two paths may lead in the
same direction, if not to the same place.
What is Science?
When Charles Darwin wrote The Origin of Species in 1859,
scientific explanations for the diversity of life competed, often
unsuccessfully, with creationist views about the immutability of
species and the presumed inability of science to explain organic
phenomena. In his book, Darwin made a powerful case for how descent
from common ancestors coupled with slow but continuous modifications
from natural selection could account for the vast array of species in
the fossil record and on the earth today. In this way, as well as
bringing science into the study of natural history, he brought history
back into science. Although he referred to evolution just once in
The Origin of Species, and only as the last word in his last
sentence where he concluded that “from so simple a beginning
endless forms most beautiful and most wonderful have been, and are
being, evolved,”1
Darwin's theory of
evolution replaced stasis and immutability with history, context, and
change. Darwin called his book “one long
argument,”
2 and
it is perhaps, along with Galileo's
Dialogue Concerning the Two
Chief World Systems, the best scientific argument ever put
forth—more so because it showed how science could attack problems
that were thought beyond its ken. Good social science, it seems to me,
should look more like the arguments in these books than the stilted and
overly mannered explication of science foisted upon us by the logical
positivists and by Milton Friedman in his essay on “The Methodology
of Positive Economics.”
3 Although the logical positivists usefully made us think more
carefully about the nature of theories, laws, explanations,
verification, and observation, they also developed a reconstruction of
science that was overly confining.4
The heart of their approach was a schema for the axiomatization of scientific
theories leading to universal covering laws that are true, have empirical
content, and are not accidental. Thus, a “law,” such as the
clearly flawed (but empirically true) assertion that “men who
take birth control pills will not get pregnant” must be rejected
because it is only accidentally true. One way to show that it is an
accidental generalization and not a law is to demonstrate that its
counterfactual—“if some men did not take birth control
pills they would get pregnant”—is false. Another tenet of
the logical positivists, argued eloquently and forcefully by Milton
Friedman, was that the test of a law is its ability to predict
events.
The emphasis upon empirical content, truth, and counterfactuals is
clearly useful. But thinking of theories as axiom systems leading to
universal covering laws and the emphasis on prediction as the test of a
theory was very limiting. In addition, as pointed out by Thomas Kuhn
and others, the logical positivists said very little about the
formation of concepts and the development of new scientific
theories.5
Perhaps the biggest failure of the
logical positivists was that their reconstruction did not seem to have
much to do with science as it was actually practiced.
What is Darwin's idea of science? Science requires a number of
interrelated steps. It requires clarifying concepts and offering causal
mechanisms that may be probabilistic as well as deterministic.6
To develop a coherent theory of the
origin of species, Darwin clarified concepts such as variation,
species, and variety; he also described a causal mechanism—the
struggle for existence leading to the natural selection of the more fit
varieties of a species—that could, over very long periods of
time, lead to the diversity and complexity found in nature. Unlike the
deterministic physical theories of Isaac Newton and Pierre Laplace,
Darwin proposed a probabilistic causal process in which the generation
of hereditary variants and the fate of any particular organism is a
matter of chance, although the ultimate fate of the variant is
determined by its overall fitness.
7 Science also requires considering alternative hypotheses such as
creationism. Time and again in The Origin of Species, Darwin
asked whether the evidence better accords “with the common view
of the immutability of species, or with that of their slow and gradual
modification, through descent and natural selection.”8
In addition, science demands testing theories by logical
argument
and empirical evidence wherever possible. Darwin
showed no sign of believing that one particular method of testing is
superior to another. To elaborate and test his theory, he used detailed
cases studies on the breeding of pigeons, on fossils found in
geological strata, and on the natural history and ecology of such
isolated areas as the Galapagos Islands. He also employed simple
mathematical calculations akin to formal models to demonstrate Thomas
Malthus's principle that the geometric increase of any population
would soon outstrip its food supply unless something kept the
population in check. He produced quantitative data on the relationship
between the geographic range of a species and the number of varieties,
finding, as he predicted, more varieties when a species covered a wider
geographic area. He described experimental data on breeding programs,
and he did simple experiments himself. Perhaps most importantly, Darwin
brought history into the study of the similarity and the classification
of organisms.
I have long thought that Darwin's scientific method is much
closer to what political scientists should embrace than the physical
science model that formed much of the basis for the logical
positivists' reconstruction of the scientific method, as
exemplified in the works of Carl Hempel, Ernest Nagel, and Abraham
Kaplan.9
Rather than look for
universal covering laws
10 Hempel and
Oppenheim 1948.
that are true in all
times and all places, political scientists should be cognizant of
history and context, and they should cultivate the systematic study of
particular places and times. They must also take chance and contingency
seriously. And they must employ whatever method they can muster to
check their theories against reality.
Evolutionary theory takes history, context, and contingency
seriously. It takes them so seriously that some philosophers have
worried that the theory's reliance upon probabilistic mechanisms
unduly limits its ability to make predictions—which are
considered by some to be the hallmark of real science. Even more
damningly, in this view, evolutionary theory's dependence upon
history and context means that it is not scientific by the standards of
the hypothetico-deductive (H-D) approach of Hempel and Nagel, which
requires universal covering laws. The probabilistic nature of
evolutionary theory requires, at the very least, that the Laplacean
ideal of exact theoretical predictions—which is embedded in the
“deductive-nomological” theory of Hempel and Peter
Oppenheim11
that was the
first complete version of the H-D approach—be abandoned in favor
of statistical tests of evolutionary theory as proposed in the revised
inductive-statistical (I-S) theory of Hempel.
12 Although moving from a deterministic to a statistical formulation of
theories is a big step, the importance of history and context in
evolutionary theory requires an even bigger step if evolution is to be
considered a scientific theory. Several decades ago, many philosophers
refused to take this step, and they argued that evolution is not a
scientific theory. For the philosopher J.J. Smart “there are no
laws in the strict sense”13
in biology because
“the theory of evolution and ecology are not, in the
logician's sense, typically ‘scientific’ in nature.
They are quite obviously ‘historical’ subjects. They are
concerned with a particular and very important strand of terrestrial
history.”
14
Karl Popper agreed with Smart.
15 Although other
philosophers have tried to rescue evolutionary theory by proposing that
it is an “immature” H-D approach, these attempts—such
as treating the Hardy-Weinberg equation for the frequencies of
genotypes after selection as “laws of nature”—seem
tortured, and the real problem seems to lie with the requirements of
the H-D approach itself. After all, it seems presumptuous and just
plain wrong for philosophers to claim that evolutionary theory is not
scientific.
Philosophers themselves have recognized the folly of this position,
and today none would claim that evolutionary theory is not a science.
As one alternative, philosophers such as Bas van Fraassen, Patrick
Suppes, and Frederick Suppe have gone beyond the logical
positivist's approach and put forth a “semantic”
conception of theories as ideal systems, models, or concepts which tell
us something about reality but which are more akin to tools for probing
reality than to laws such as F = ma or E =
mc2.16
These tools propose possible
structures for phenomena such as suggesting that gases are composed of
large number of swiftly moving molecules or that light propagates like
the waves on a pond. These structures make it possible to ask questions
about phenomena and to provide some explanation of them—but they
will inevitably have limits such as when molecules in gases are slowed
to the point where quantum-mechanical effects take over or when light
waves act like particles (quanta) to produce electricity in discrete
amounts in the photo-electric effect.
Elisabeth Lloyd and other philosophers who embrace a semantic
perspective happily concede that there are no universal, physically
necessary laws in modern evolutionary theory, but they argue that the
theory is scientific because there are concepts such as variation
within a species and models such as natural selection which guide
biologists' thinking about the development of species.17
Moreover, within genetics and ecology there is a profusion of concepts and
models which can be used to explore various aspects of evolution. These
concepts and models are devices for understanding reality. Sometimes they
are merely abstractions or simplifications or even “fictions”
used as conceptual devices for illuminating what might or might not be true.
Sometimes they are predictive models. But unlike the logical positivist
and related perspectives
18 that view prediction or
falsification as the crucial activity in science, the semantic
perspective allows for the possibility that even very approximate
theories may tell us something and that theories can be confirmed or
disconfirmed through strategies that go beyond their predictive power.
These confirmation strategies include the fit between the model and
data, support for various assumptions embedded in the model, or the
wide range of evidence that accords with the model. These are precisely
the strategies used by Darwin in
The Origin of Species. The
semantic approach suggests that science is about creating a workshop of
tools, all constantly being used and redesigned to fashion an
understanding of reality. There is no master tool, but there is
constant attention to improving the relationship between the tools and
the projects at hand. The result is a growing capacity to use the tools
to explain the world and to shape it to our purposes.
This perspective turns us away from seeking universal covering laws
and towards finding useful models for understanding the world. In
political science, Mancur Olson's theory of group membership,
which introduced the “free rider problem,” might be
considered a powerful fiction that raised questions about group
formation.19
James C. Scott's notions of the
differences between the public performances and the hidden transcripts
of those in subordinate positions raise questions about how power is
experienced and deflected and how visible evidence should be understood
and interpreted.
20 Gary Cox's extension of
Duverger's law on how the number of parties varies with the form
of the electoral system provides some mechanisms, namely strategic
voting and strategic entry by politicians, to explain the law's
assertion that there are typically no more than two parties in
single-member plurality systems.
21 These models are used in
different ways, but they do not claim to produce universal laws. Their
results can be subject to historical contingency, and they can even
contradict one another because they have different purposes—as in
models of the motion of fluids which assume a continuous medium versus
models of the chemical behavior of fluids which assume individual
molecules.
Should Political Science be Scientific?
The essays by James Granato and Frank Scioli and by Rogers Smith
suggest that Darwin, and the biological sciences in general, do,
indeed, provide an excellent template for political science. Both
essays agree that we should be committed to what Gary King, Robert
Keohane, and Sidney Verba have called a unified logic of scientific
inference.22
King, Keohane, and Verba
1994.
Smith says that:
If we are to judge, for example, to what conceptions of their
identities and interests particular political actors are giving
priority, we need to form some hypotheses based on what we think we
know about those actors. Then, we define the different implications of
alternative hypotheses. Finally, we look for observable data about
their lives that we can use to falsify some of the hypotheses. That
logic is constant, though the techniques of falsification will vary
with the types of problems particular data present and with the tools
currently at our disposal.
Granato and Scioli make a similar point:
In an ideal world, where there is unification in approach,
political science research should have the following components: (1)
theory (informed by case study, field work, or a “puzzle”);
(2) a model identifying causal linkages; (3) deductions and hypotheses;
(4) measurement and research design; and, (5) data collection and
analysis.
Granato and Scioli go on to argue that “the three most common
current research practices—formal modeling, case study analysis,
and applied statistical modeling—deviate from this ideal. They
therefore limit the possibilities for substantial enhancement of
knowledge.” They argue that formal models sometimes fail to
respect the facts; that case studies sometimes theorize by proverb; and
that statistical analysis sometimes amounts to data mining, garbage can
regressions, and statistical patches. They believe that the result is a
series of unreliable findings and the “proliferation of
non-cumulative studies” that do not provide useful knowledge,
especially knowledge that can be used by policymakers. Their solution
is to connect formal and empirical analyses so that concepts are
clarified and causal linkages specified. That is the goal of a new
National Science Foundation program called Empirical Implications of
Theoretical Models which will “encourage political scientists to
build formal models that are connected to an empirical test” and
“lay the groundwork for (social) scientific cumulation.” In
short, Granato and Scioli want more research to accomplish what Darwin
did in The Origin of Species.
What Special Problems Does Political
Science Face?
In principle, I suspect that Rogers Smith does not disagree with what
Granato and Scioli want to do, but in practice, he probably worries
that it will lead to an undue emphasis upon universal covering laws.
“We cannot,” he says, “fully understand the sources
and character of changes in political identities by looking only at
timeless behavioral constants.” Much of his article, in fact, is
devoted to two examples, both involving the study of political
identities, where he feels that formal theory and a-historical
behavioralism have grasped for and failed to produce “timeless
behavioral constants.” For Smith, “[I]f political
identities are socially, indeed politically, constructed, we need to
attend to the historical and contextual processes through which
identities are constructed differently among different groups in
different times and places.”
To get at these processes, he recommends “richly interpretive
methods that involve discursively grasping the consciousnesses and
senses of value and meaning that identities give to people.” For
Smith:
We are likely to gain more insight on these topics through
interpretive textual analyses; ethnographic fieldwork; biographical
studies; in-depth interviews; individual and comparative case studies,
both historical and contemporary; participant observation research;
narrative historical institutional analyses; and other methods rather
crudely termed “qualitative.”
Smith brings up four distinct issues in these comments. First, he
questions the wisdom of searching for universal covering laws producing
“timeless behavioral constants.” Second, he suggests that
we must attend to historical and contextual processes, especially
political processes. Third, he notes that we must take meaning and
values seriously. Finally, he proposes that the way to do this is
through “contextual immersion, psychological empathy, and
creative imagination.” Each issue requires some discussion.
I have already suggested that Smith is right in thinking that science
does not require universal covering laws. Just as evolutionary theory
takes history and context seriously, so must social science. Smith
presents two examples of research that he believes rely too much on
trying to identify “abstract, ahistorical, and enduring
regularities in political behavior.” David Laitin's
Identity in Formation uses a rational
choice “tipping model” to explain the conditions under which
Russian speakers in those states that were formerly part of the Soviet
Union will decide to learn the non-Russian language of the dominant group
(the “titular nationality”) in those states. Henry Brady and
Cynthia Kaplan's “Categorically
Wrong” uses psychological models and data from Estonia to
explain when nominal measures of ethnic identity will suffice for predicting
attitudes and when gradations of ethnic identity will be necessary to
explain variations in attitudes.
Smith argues that Laitin's tipping model explains little
compared to measures of social distance, and these measures are
inherently the result of “the particularities of the different
languages and religions in question” and the degree to which
“they resemble each other in decisive respects.” But social
distance, says Smith, is the result of “different historically
shaped patterns of political institutions and state policies.”
Similarly, Brady and Kaplan argue that nominal measures of identity
suffice when an identity is highly salient, whereas graded measures
(based upon social distance) will be necessary when an identity is less
salient, but “they do not go on to explore the logical next
question: what determines whether ethnic identity is salient to certain
groups?” Yet, Smith notes, it seems likely that salience is the
result of political factors, some of which are even described by Brady
and Kaplan in their discussion of Estonia's history and
institutions.
As the co-author of one of these two works, my inclination is to
tread somewhat lightly here. But Smith is certainly right in saying
that historical factors matter in determining social distance and
salience, just as historical factors matter in determining the genetic
constitution of an organism or the environment that it faces. But he
seems to glide too quickly over the contributions of these two pieces
of research—namely, the identification of the major factors,
social distance and the salience of identity, that explain why Russians
choose to learn the titular language or why members of ethnic groups
can be treated categorically in some circumstances. The identification
of these factors makes it clear what history must explain. This, it
seems to me, is no small feat. To put it in perspective, Darwin's
achievement was not just to say that history mattered, it was to say
that history had to explain how natural selection and survival of the
fittest led to the evolution of an organism. Furthermore, beyond
identifying what history must explain, the trick is to say how it
matters. Smith seems to recognize this, and he ends with his own
proto-theory of the crafting of political identity through coercive
force and persuasive stories. He also suggests that we need “to
study processes of political identity formation largely through
interpretive, ethnographic, and historical methods.” I certainly
agree that we need to study these processes through a variety of
methods, but I wonder why Smith limits himself to these particular
ones. Why not quantitative and survey methods as well? And why does
Smith do so little to explain the comparative advantages of the methods
he favors over those he does not mention?
Perhaps the advantage of the methods he mentions is that at least
some of them, such as interpretive methods, take meaning and values
seriously. Certainly two of the most important things about human
society are that we use language and we construct cultures. Meaning is
very important to human beings. Social science theories that ignore
meaning, such as the stimulus-response behavioralism of the
psychologists B.F. Skinner and Clark Hull or the revealed preference
utility theory of the economists Paul Samuelson and John Hicks, seem
crippled from the start. Yet interpretive methods are by no means the
only techniques that take meaning and values seriously. Survey
researchers have puzzled over the meaning of survey questions;
conversed with people to get their responses; considered their values;
and measured their religious, ethnic, political, class, and social
identities for decades. In fact, perhaps the best known identity
concept in political science is party identification, which was
introduced in the early 1950s by Michigan survey researchers. As a
survey researcher myself, I'm a bit bemused by Smith's
clarion call for political scientists to take identities seriously.
Smith tells us that “the most crucial work in analyzing
political identities must often be done by immersing ourselves in
information about the actors in question, and using both empathy and
imagination to construct credible accounts of identities and
interests.” I have no quarrel with the rhetoric of this
assertion, but what does it really mean? Good survey researchers become
immersed in the cultures that they study; they use focus groups; they
contemplate what questions mean to those who are asked them; they
analyze pre-tests; they train interviewers using hundred-page
instruction booklets; and they employ experimental variations in
question wording to explore people's identities, attitudes, and
values. Empathy and imagination are certainly part of this method, but
there is also a great deal of technique that has been built-up over the
course of fifty years of survey research. It is a simple fact that the
literature on survey research is much more extensive than the
literature on interpretive methods. Indeed, the literature on
interpretive methods is very thin, and it often eschews technique as
detrimental to the method. One reads Clifford Geertz with wonder, but
also with questions about his technique. Does he use methods that
others might learn? If not, then how can we be sure that his
interpretation of a Balinese cockfight is something more than brilliant
literary invention? The problem here is not any doubt about the need
for interpretation and for taking meaning seriously. Meaning
matters—that is why survey researchers ask people what they think
on surveys and why they go to great lengths to interpret people's
responses. The problem is that we know some things about how to teach
students to formulate and interpret survey questions, but I am not sure
that we know much about how to teach them to do interpretive work.
Others have some ideas about how to do this, and I am associated with a
growing movement to improve the explication and teaching of qualitative
methods. I have also co-edited a book with David Collier on what
quantitative methodologists have to learn from qualitative methods (as
well as the reverse).23
But, and I lay this down
as a friendly challenge, interpretivists have a ways to go before they can
claim a body of methods as well defined and articulated as those available
to survey researchers.
Of course, one response to my call for better codification of
interpretive methods is to simply reject the notion that interpretive
methods should be scientific. This is not Smith's response. He
wants theories that that are, “if not universal, at least
falsifiable, replicable, and generalizable over some middle range of
cases. They should not be simply thickly descriptive case studies or
more literary narrative accounts of unique historical events.”
And I do not think it is the response that any social scientist should
offer. Just as modern biology proves that the organic world can be
studied scientifically, the lessons of modern psychology, linguistics,
and social science are that human behavior can be studied
scientifically as well. Let's not give up on the attempt before we
have really tried.
Linking Theories with Data and with
History, Context, and Meaning
Rogers Smith provides one direction that we could take to improve
political science. I think he is right in believing that we should take
political identity even more seriously (even though survey researchers
have been taking it seriously for a long time), and I agree with his
assertion that history, context, and meaning must be part of any
serious explanation of social behavior (even though I think that
philosophers of science long ago gave up the idea that there should be
universal covering laws that ignore them). The EITM style research
suggested by Granato and Scioli could, in my view, help move us in this
direction for the simple reason that the dialectic they endorse between
models and reality will inevitably lead us to the important factors
that explain political behavior. Indeed, their four examples of
EITM-Type research suggest that this is so. Their first example, Philip
Converse's model of adjustments to partisan identification, shows
how history explains weak partisan attachment in France compared to
America. Their second example, the Phillips curve, shows how
people's rational desire to understand the inflation-unemployment
tradeoff leads to their formation of rational expectations that destroy
the relationship. Their third example, quantal response equilibrium,
shows how the consideration of people's strategic calculations in
international relations lead to markedly different conclusions about
the role of military forces. In both of these last two cases,
acceptable models of human behavior require assumptions about how
people think, interpret, and act strategically. The fourth EITM example
shows how public concern, or at least discussion, of a disease affects
the Food and Drug Administration's approval of drugs. In each
case, history, context, and meaning play important roles.
The Granato-Scioli article is notable for its eclecticism. It finds
virtue in qualitative and quantitative research, case-studies and
large-n studies, and formal models formulated verbally or
mathematically. I view this as a great strength and in the tradition of
Darwin's eclecticism. And, like Darwin, Granato and Scioli want to
use these methods in concert to submit theory to empirical tests and to
structure reality with theoretical models. This EITM quest leads us
back to a fundamental question: What is a theory? Just as the question
is not easily answered for biology, it is not easily answered for
social science. Granato and Scioli's examples suggest that Markov
chains, rational expectations models, quantal response equilibrium, and
optimal stopping models are suitable theories. That seems right to me,
but what about James C. Scott's theory of domination and the arts
of resistance? What about Roger Smith's theory of the crafting of
political identity through coercive force and persuasive stories? I
called it a “proto-theory” earlier, but mostly because
Smith himself calls it a “sketch.” It would be useful if
Granato and Scioli could tell us what would be required to turn
Smith's sketch into a full-blown theory. More generally, we
probably need work that identifies a broad range of theoretical
exemplars. It would be a great contribution if EITM could produce this
kind of guidance.
Will the path suggested by Granato and Scioli and the path suggested
by Rogers Smith lead us to exactly the same place? The papers have
different emphases. At first blush, Smith is more substantive, with a
focus on the study of identity through attention to history, context,
and meaning; Granato and Scioli are more methodological, with their
focus on linking theories and data. But upon reflection, Smith is also
deeply methodological with his concern for finding ways to get at
meaning and context, and Granato and Scioli are deeply substantive with
their desire to make political science more capable of advising
policy-makers about political systems and public policy. I view the two
papers as complementary. Both endorse a scientific approach that
requires theory-building and theory-testing. Both endorse the use of
different methods for studying politics. Like Darwin, they believe that
good science consists of good arguments constructed from carefully
conceived theories that are tested in multiple ways using a variety of
methods. Once we embrace this view of shared standards, it seems to me
that different starting points and diverse methods are a virtue because
they help ensure that we have not missed any of the important terrain,
and our shared standards will lead us in the same general
direction—towards a better understanding of politics.
