Boston Colloquium for the Philosophy of Science
Boston
University
(Fall
1992)
Abstract
This
article disputes the common view that social science explanations depend on
discovery of lawlike generalizations from which descriptions of social outcomes
can be derived. It distinguishes
between governing and phenomenal regularities, and argues that social
regularities are phenomenal rather than governing. In place of nomological-deductive arguments the article
maintains that social explanations depend on the discovery of causal mechanisms
underlying various social processes.
The metaphysical correlate of this argument is that there are no social
kinds: types of social entities that share a common causal constitution giving
rise to strong regularities of behavior.
The article turns next to a consideration of the character of social
causation and argues for a microfoundational interpretation of social
causation: social causal powers are embodied in the constraints and
opportunities that institutions present to individual agents. Finally, it is noted that these
arguments have consequences for the credibility of social predictions: it is
argued that predictions in social science are generally unreliable.
Is
the social world law-governed? Are
there social laws, analogous to laws of nature, that give rise to social
phenomena? And is it a central
task of the social sciences to discover law-like regularities among social
phenomena? These are the questions
to be discussed in this paper.
Some
social scientists write as though the scientific credentials of their
disciplines rise or fall on the strength of the law-like generalizations and
regularities that they are able to identify.[1] The task of social science research is
to discover the laws that govern social processes. And if a given level of analysis and description fails to
produce such laws, then we need to probe more deeply until we discover the underlying
order.[2] At the other extreme, some social
scientists write as though generalizations have nothing at all to do with
social knowledge. All knowledge is
"local knowledge" (Geertz 1983): historical, culturally specific,
unique, particular, singular. On
this approach, there are no interesting regularities among social phenomena,
and causal explanation is an inappropriate model of explanation for the social
sciences.[3]
I
disagree with both these views.
The general view I will defend is that there are regularities to be
found within the social sciences, at a variety of levels of social description;
that these regularities derive from features of individual agency in the
context of specific social arrangements; and that discovery of such
regularities is one important goal of social science research. But I also maintain that these
regularities are substantially weaker than those that obtain among natural
phenomena. They are phenomenal
regularities, not governing regularities; and--in comparison to typical
phenomenal regularities among natural phenomena--they are substantially less
reliable. Moreover, I maintain
that these regularities have a much more limited role within good social
explanations than they are often thought to do. The upshot of these arguments is that we should have little
confidence in the projectability of social regularities as a basis for
prediction, and must therefore pay more attention to the specifics of the
social and individual-level mechanisms that produce the regularities as well as
the exceptions. And this analysis
implies as well that some social scientists have drawn the wrong lessons from
the legacy of logical positivism.[4]
Subsumption under law-like regularities is not the appropriate model for social
explanation. Rather, the
corresponding goal ought to be premised on causal realism: discovery of
mechanisms and processes that derive from agents and institutions, and that in
turn produce regularities.
Regularities derive from underlying causal properties, and it is the discovery
of these that is the explanatory business of social science research.[5]
Much
of the impulse toward emphasizing the explanatory importance of regularities in
the social sciences derives from an unhelpful analogy with the natural
sciences. The successes of the
natural sciences have given natural scientists confidence that natural systems
operate in accordance with a strict set of laws, that these laws may be given
precise mathematical formulation, that they derive from the underlying real
properties of constituent physical entities, and, finally, that these facts
entail that the future behavior of physical systems is in principle (though
perhaps not in practice) predictable.
And this conception of the nature of physical systems in turn gave rise
to a paradigm of scientific explanation: to explain a phenomenon is to derive
the explanandum from a set of general laws and a description of the initial
conditions of the system.
Prediction and explanation go hand in hand, and both depend on the
availability of empirically supportable general laws.[6] For reasons that will be developed
below, however, I do not believe that this is a good way of understanding
social phenomena.
A
related reason social scientists have thought that the search for
generalizations and regularities should be central for their research and
theorizing is the influence of the covering-law model of explanation. On this approach, to explain a
phenomenon is to show that the phenomenon (or empirical generalization) may be
inferred from one or more general laws.
If the covering-law model is a correct theory of all scientific
explanation, then social scientists, if they are to produce explanations at
all, must arrive at generalizations.[7] However, in Varieties of Social
Explanation I argue for a view that places
causal analysis, not subsumption under general laws, at the core of social
explanation. Here the general idea
is that explanation of a phenomenon or regularity involves identifying the
causal processes and causal relations that underlie this phenomenon or
regularity.[8] On this approach, the central
explanatory task for social scientists is to uncover causal mechanisms, not to
formulate explanatory regularities that permit the deduction of observed
phenomena. There are regularities
that correspond to causal mechanisms, to be sure; but these may not be
discernible (because of the difficulty of isolating causal factors), and they
are unlikely to take the form of strong high-level regularities across social
contexts (e.g. strong regularities of behavior of certain types of states, trading
regimes, or popular movements).[9]
So
I will argue for a position that is neither positivist nor
anti-positivist. Against the main
line of positivist philosophy of science, I will dispute the idea that law-like
generalizations are fundamental to successful scientific explanation. But against current anti-positivist
criticisms among some social scientists, I will argue for causal realism in
social explanation: causal explanation is at the core of much social research,
and causal hypotheses depend on appropriate standards of empirical confirmation
for their acceptability. Finally,
successful causal analysis permits us to arrive at statements of social
regularities--this time, however, based on an understanding of the underlying
processes that give rise to them.
And this understanding permits us to assess the limits and conditions of
the regularities we affirm, and the likelihood of failure of these regularities
in various social circumstances.
It
is important to emphasize that my reasons for these conclusions are not a
priori arguments to the effect that the
social sciences must have this or
that feature (as is common in much discussion of methodological individualism,
for example). Philosophy of
science needs to be done in close proximity to the scientific disciplines that
are its subject. The arguments
here respect this requirement. The
conclusions I advance depend, first, on features of the generalizations available
in numerous areas of the social sciences, and second, on an empirically
informed analysis of the metaphysics of social causation. The generalizations that are available
in political science, economics, or sociology are weak and exception-laden, and
they permit only tentative predictions about future developments. The burden of this paper is to offer a
theory about why this is so. And
the theory that I offer involves an analysis of social causation and individual
agency: I argue that the only form of causal influence that social entities have
is through their effects on individual action, and that this leads quite
understandably to generalizations of the sort we find in the social sciences.
It
might be thought that the central conclusions here have anti-naturalistic
implications.[10] But I do not believe that this is
so. I, for one, certainly regard
social phenomena as natural: they are the result of the actions and states of
human beings, who are themselves natural organisms. Social phenomena supervene upon physical systems which are
themselves regulated by laws of nature.
But this does not entail that there are strong law-like regularities at
the social level. There are many
reasons why this is so. First,
multiple causation with complicated INUS conditions and probabilistic causation
may result in higher-level regularities being indiscernible given feasible data
sets.[11] Second, turbulence, chaos, and
sensitivity to parameter change may mean that outcomes are in principle
unpredictable.[12] Third, social causation is commonly
probabilistic; this means that long causal chains accumulate uncertainties, so
that we cannot predict outcomes based on initial conditions. And finally there is the problem of
specification. Even if our theory
correctly describes the causal mechanism, it is often the case that the
hypothesis must be specified in terms of a mathematical model, and this can be
done in diverse ways--with different predictive consequences. Likewise, the predictions of the model
depend on the specification of its parameters; but there is generally a range
of plausible empirical estimates of the parameters of a model, upon which
substantial differences of the model's performance may depend. So one may consistently affirm that
social phenomena are natural while at the same time deny that there are laws
governing social phenomena.
What
is a law-like regularity? It is a
universal generalization about empirical phenomena. It is one that conveys necessity. It is a generalization that supports counterfactual
judgments. It is a regularity that
is grounded in the real causal properties of the entities in question. These are reasonably familiar ideas
from the philosophy of science.
Let
us begin by distinguishing between what I will call "governing
regularities" and "phenomenal regularities."[13] The notion of a law of nature
represents a paradigm of a governing regularity: a description of the laws that
generate the behavior of a given kind of thing. A phenomenal regularity, by contrast, is a regularity of
behavior that emerges from the real causal properties of a thing, but which
does not itself give rise to or constrain the thing's behavior. It is a governing regularity that
protons and electrons are attracted by the forces described by electrodynamics. It is a phenomenal regularity that
glass flows slowly; given the real constitution of glass, it emerges that glass
has many of the phenomenal properties of a liquid. (Note that the distinction between governing and
phenomenal regularities is not the same as that between law-like
generalizations and accidental generalizations. Phenomenal regularities support counterfactuals, so they
qualify as law-like, not accidental; but neither are they essential,
determining, or regulative.
Likewise, the distinction does not collapse onto that between
theoretical generalizations and inductive generalizations. A governing regularity could be
identified inductively, and a theoretical account may give rise to a phenomenal
regularity.)
Having
made this distinction, the central topic of this paper divides into two
questions. Are there governing laws of social phenomena? And are there phenomenal
regularities? My answer to the
first question is negative. The
closest thing available are laws that describe common features of agency: given
that persons want such and so in a given environment, thus and so emerges. It has been observed, for example, that
land-tenure systems with a particular structure create common incentives for
individuals wherever they are implemented; it is then a regularity of these
systems that they have common features (e.g. underinvestment in capital
improvements). But these
regularities are strictly derivative from features of individual agency, and
they do not represent governing regularities of a certain kind of social
institution.
Consider
an example. Are there laws of the
modern state as an institution?
Social scientists have discerned a variety of regularities concerning
the state: states maximize revenues (Levi); state crises cause revolutions
(Skocpol); states create entrenched bureaucracies; and the like. These statements represent
generalizations across a number of cases, and they are intended to have
counterfactual import. But it is
plain that these generalizations are subordinate to our hypotheses about the
underlying institutional and individual-level circumstances that give rise to
the regularities of state behavior identified. States conform to regularities because they are the product
of a number of agents whose purposes, powers, and opportunities are similar in
many different social contexts; this leads to a regularity of state behavior.
Consider
for a moment a different sort of regularity. Hand tools weigh less than 20 pounds. This is a strongly confirmed empirical
regularity over its domain. But
the existence of the regularity does not lead us to imagine that there are
governing laws of hand tools.
Instead, this regularity emerges from facts about human constitution,
powers, and intentions. Tools are
designed to be used by human beings in transforming nature. It is a common fact about human beings
that there are limits to their strength and dexterity. Tools are designed by humans in light
of these limits. It results from
these facts that there will be regularities across tools.
Likewise,
I suggest, for social regularities. Social phenomena are the consequence of intentional human
actions (sometimes in vast numbers).
Some social regularities are the intended consequence of individual action--e.g.
the revenue-maximizing property of typical states follows from the interest that
government officials have in maximizing the power and income of the state. Others are the unintended consequence
of purposive individual action--e.g. the falling rate of profit is the
unintended consequence of profit-maximizing strategies by large numbers of
individual capitalists (according to Marx). But in either case the regularity is strictly derivative
from the constitution and powers of the individuals whose actions give rise to
the social phenomenon in question.
This
leads to a fairly clear conclusion.
Social regularities emerge rather than govern. The governing regularities are regularities of individual
agency: the principles of rational choice theory or the findings of
motivational psychology. Social
regularities are strictly consequent, not governing. They obtain because of the lower-level regularities; they
have no independent force (unlike a common interpretation of the force of the
laws of gravitation).
Turn
now to the second question: are there phenomenal regularities among social phenomena,
and are these explanatory? Here
the answer is, yes and no. It is
evident that there are phenomenal regularities among social phenomena; but
equally, these regularities turn out not
to be explanatory. States tend to
maximize revenues (Levi 1988); low income states tend to have high infant
mortality rates; bureaucracies tend to resist change. These are regularities that can be discerned through
empirical investigation--chiefly large multi-case studies and small comparative
studies. And these are
regularities that to some degree support counterfactual judgments: If India's GNP were to double, its
infant mortality rate would fall.
But I take the view that phenomenal regularities are not
explanatory. If we want to know
why windows are thicker at the bottom, it is not explanatory to offer the
argument that windows are made of glass and glass flows like a liquid. Rather, we want to know what it is
about the fine structure of window glass in virtue of which it flows. That is, we need to know what the
causal mechanism is that gives rise to this phenomenal regularity.
Likewise,
the fact that India has a high infant mortality rate is not explained by the
circumstance that India's per capita GNP is low and the phenomenal regularity
that "countries with low per capita GNPs tend to have high infant
mortality rates." The
bare discovery of a stable statistical relationship between these two
characteristics is not by itself explanatory of a given country's high infant
mortality rate. Instead, we need
to know what it is about developing countries in typical circumstances such
that low GNP is associated with high infant mortality. That is, we need to know what the
causal connection is through which low GNP leads to circumstances in which
infant mortality is likely to be high.
But once we have identified the typical causal mechanism, we no longer
need to use the phenomenal regularity to explain; we can then directly
ascertain whether that mechanism is in place in India and can account for
India's high infant mortality rate as the consequence of the presence of the
specified causal factors. (The
mechanism is presumably something like this. Low GNP causes low personal income and low state
revenues. Low personal income
entails low ability to pay for nutrition and health care. This leads to poor average maternal
health. Low government revenues
entail low ability to pay for publicly funded health and nutrition programs. This also leads to poor average
maternal health. But if the state
devotes a substantial fraction of its resources to public health, these causal
connections do not go through, and we should expect an exception to the
rule. Thus the exception of Sri
Lanka, which is a country whose per capita GNP is roughly that of India, but
whose infant mortality rate is comparable to that of many European countries.)
So:
there are phenomenal regularities among social phenomena, and these can be
discerned through familiar forms of empirical investigation; but they do not
serve an important explanatory function within the social sciences.
Another
way to put the thrust of my position is to consider whether there are social
kinds, analogous to natural kinds.
A natural kind is a set of entities which share a common causal
structure, and whose behavior can therefore be predicted on the basis of the
laws that govern the behavior of such entities (Putnam 1975b). A social kind, then, would be a class
of social entities that share a common causal structure; this common structure
would give rise in turn to one or more governing laws. Candidates for social kinds include
"riot", "revolution", "class",
"religion", "share-cropping land-tenure system", "constitutional
monarchy", "market economy", "nationalist political
movement", "international trading regime", and "labor
union". Note, to begin, that
it is not the case that all scientifically useful concepts and categories
designate natural kinds. The
concept of a predator is useful within evolutionary theory; likewise, the
concept of an acid in chemistry, an earthquake in seismology, or a gas in
physics. None of these represents
a genuine natural kind, however, since there is no homogeneous causal structure
that is shared by all members of the class.
I
deny that any social concepts serve to identify social kinds in the strong
sense outlined above. Instead,
social concepts function as ideal types or cluster concepts, permitting us to
classify a range of diverse phenomena under a single concept. The notion of a cluster term captures
many scientific concepts--terms that encompass a variety of phenomena that
share some among a cluster of properties (Putnam 1975a, pp. 50-54). An ideal-type concept is a complex
description of a group of social phenomena that emphasizes some features and
abstracts from others (Weber 1949).
It is apparent that generalizations and predictions based on cluster
concepts and ideal types demand a great deal of care. Since the entities that fall under such concepts do not
share a homogeneous causal structure, we cannot infer that instances of the
concept will behave in the typical way.
Thus market economies have many properties in common. However, particular market economies
also have causally significant differences; for example, the purported
willingness of Japanese investors to consider a longer time horizon in their
investments than their American counterparts could be expected to give the
Japanese market economy different characteristics than the U.S. economy.[14]
The
value in making use of cluster concepts and ideal type concepts in the social
sciences is that it permits us to group social entities together in ways that
emphasize their common features.
This serves to suggest hypotheses about the dynamic properties of such entities. But at the same time, the fact that
there are wide differences in important causal features among the entities that
fall under a given concept, means that we cannot simply project the future
behavior of the entity on the basis of the general features that it shares with
other instances.
The
metaphysical counterpart, then, to my view that there are no governing social
regularities among social phenomena, is that there are no genuine social
natural kinds.[15]
In
place of the goal of discovering governing regularities, I maintain that
explanations in social science typically involve efforts to uncover the causal
properties of social entities and processes. In Varieties of Social Explanation I argue that the central idea of causal ascription is
the idea of causal powers and causal mechanisms: to assert that A causes B is
to assert that A in the context of typical causal fields brings about B (or
increases the probability of the occurrence of B). This view may be elaborated in terms of the idea of a causal
chain: A causes B just in case there is a series of causal mechanisms linking
the occurrence of A to the occurrence of B. This may be called "causal realism."[16] Second, I argue for a microfoundational
approach to social causation: the causal properties of social entities derive
from the structured circumstances of agency of the individuals who make up
social entities--institutions, organizations, states, economies, and the
like. The mechanisms through which
social causation is mediated turn on the structured circumstances of choice of
intentional agents, and nothing else.
(This is not equivalent to methodological individualism or reductionism
because it admits that social arrangements affect individual action. This is the structuring to which I
refer in this formulation.) This
means that social science research that sheds light on the individual-level
mechanisms through which social phenomena emerge have a foundational place
within the social sciences: rational choice theory, theory of institutions and
organizations, public choice theory, analytical Marxism. What these fields have in common is a
commitment to providing microfoundations for social science.
My
argument so far is that causal mechanisms are more fundamental than
regularities of association between causal variables. But there is an apparently straightforward objection to this
approach: don't causal ascriptions depend on or at least implicate
regularities? Two points can be
made here. First, if A has the
causal power to bring about B in a wide range of causal fields, then it will be
true that there will be an observable regularity of association between A and
B. But this does not imply that
the causal power reduced to a description of the corresponding association. Consider the statement that magnets
cause forces in iron filings. This
is a causal fact that gives rise to regularities, and it is easily discerned by
observing regularities. But it
does not reduce to a bare regularity, and it is not necessarily discovered or
verified on the basis of observed regularities. More profoundly, however, it is possible that A has the
causal power to bring about B in some fields and not others, with the result
that it is practically impossible to observe the corresponding regularity,
given data limitations. Causal stories
involving complex causal diagrams, complex sets of INUS conditions,
probabilistic causation, and incomplete causal fields give rise to situations
in which two things may be true: A is a cause of B (in that A is an
ineliminable part of the underlying causal diagram or INUS conditions), and
there is no observable correlation between A and B.[17]
What,
then, can we say about the causal properties of social entities? Do social entities have causal
properties? Does a given state,
labor organization, bank, or political party have causal properties? And do types of social entities have
common causal properties? That is,
do states, labor organizations, banks, or political parties have common causal
properties? Consider the causal
powers of the U.S. government with respect to U.S. economic activity. Various agencies have instruments of
action that produce changes in economic activity. The economic variables of interest include the inflation
rate, the rate of employment, and the growth rate. Changes in money supply, changes in federal spending, and
changes in interest rates are all actions that government agencies can
undertake that have effects on economic activity. Do these constitute causal powers in the sense described
above? They do, but this judgment is
attenuated by the fact that the relation between cause and effect is often
highly contextual in the case of social causation. In some contexts lowering the interest rate may stimulate
growth while dampening inflation; in other contexts it leaves both growth and
inflation unchanged. This implies
that an adequate causal analysis will not take the causal properties of the Fed
as basic, but will rather involve a large number of causal factors (including
the Fed's actions) which jointly produce given outcomes.
Second,
we can say a great deal about the metaphysics of social causation. The discussion of microfoundations
above gives the clue; the causal properties of a social entity consist in the
structures that it embodies that affect the actions of individuals (through
incentives, opportunities, powers, information). I assert that certain social entities have causal
relevance--e.g. centralized bureaucratic states have greater capacity to
collect revenues from the periphery than decentralized feudal states (Mann). What this capacity consists in,
however, is not merely the observed regularity that corresponds to it. And it is not some mysterious social
force inhering in the social entity itself. It is rather the specific features of these states in virtue
of which the agents of the state have both the interest and the means to
effectively extract revenues from actors distant from them.
The
idea that a social entity has causal powers suggests that the inner
constitution of the entity is such as to lend necessity to certain transitions
and not others. Is there a
relation among social states of affairs that conveys necessity? What features of social life would
support such a judgment? What
connections between states of affairs are available that would provide an
interpretation of social necessity?
Elsewhere I use the idea of a "logic of institutions" to
attempt to capture the idea that a set of social circumstances brings about
certain types of outcomes (Little 1986).
I describe an institutional-logic explanation as an analysis that is
concerned with determining the results for social organization and development
of an entrenched set of incentives and constraints on individual action (Little
1986, p. 34). Given the stylized
arrangements described in the explanans, we can expect the outcome. I maintain that this is the sole form
of necessity that can be discerned among social phenomena. Note, however, that the necessity that
attaches to an institutional logic is solely grounded in the intentionality of
the individuals whose actions are affected by the arrangements under
scrutiny. The explanatory force of
an institutional logic depends fundamentally on facts about individual agency.
Consider
an example. Transport systems have
the causal capacity to influence patterns of settlement; settlements arise and
grow at hubs of the transport system.
Why so? It is not a brute
fact, representing a bare correlation of the two factors. Instead, it is the understandable
result of a fuller description of the way that commerce and settlement
interact. Agents have an interest
in settling in places where they can market and gain income. The transport system is the structure
through which economic activity flows.
Proximity to the transport system is economically desirable for agents:
they can expect rising density of demand for their services and supply of the
things they need. So when a new
transport possibility emerges--extension of a rail line, steamer traffic
farther up a river, or a new shipping technique that permits cheap
transportation to offshore islands--we can expect a new pattern of settlement
to emerge as well. This is an
instance of an institutional-logic explanation.[18]
So
far, then, my argument is two-fold: social entities have causal influence, and
these causal capacities are to be explained in terms of the structuring of
incentives and opportunities for agents.
The causal powers or capacities of a social entity inhere in its power
to affect individuals' behavior through incentives, preference-formation,
belief-acquisition, or powers and opportunities. The micromechanism that conveys cause to effect is supplied
by an account of the actions of agents with specific goals, beliefs, and
powers. Social entities can exert
their influence, then, in several possible ways.
1. They can alter the incentives
presented to individuals.
2. They can alter the preferences
of individuals.
3. They can alter the beliefs of
individuals. (constraints on
knowledge; ideology)
4. They can alter the powers or
opportunities available to individuals.
Lowering
the prime interest rate has the causal capacity to reduce the rate of
inflation. Why is this? Because rational investors lower their
rate of investment in the face of lower interest rates; demand for producer
goods falls; incomes for workers remain steady; and demand for goods remains
flat. So prices tend to stay
constant. This story accounts for
the causal powers of the intervention in terms of the incentives created and
strategies available to the relevant agents.
The
conclusion of this line of thought is that institutions have effects on
individual behavior (incentives, constraints, indoctrination, preference
formation), which in turn produce aggregate social outcomes. Some social regularities follow
immediately from these effects--e.g. increasing the tariff on imported running
shoes leads to an increase in consumption of domestic running shoes. (This regularity derives from the fact
that consumers are price-sensitive; so increasing the cost of imports leads to
a shift in typical consumer's behavior.)
(Naturally, we need another story to tell to indicate how institutions
are embodied in the current beliefs, preferences, and behavior of existing
individuals.)
We
can say also, derived from this first point, that certain institutions have
specific causal powers with respect to given social outcomes as a consequence
of the common constitution and circumstances of individuals. The Fed has the causal power to dampen
inflation, in that it can tighten the money supply; this creates an individual
disincentive to purchase; this leads to reduced demand for goods; and this
lessens the upward pressure on prices.
This causal power is entirely derivative, however, upon facts about typical
consumers. The Fed has the power
to alter the environment of choice for consumers; the result of this new
environment is a pattern of consumption in which demand is shifted downward.
The
upshot, then, is this. Social
entities possess causal powers only in a weak and derivative sense: they possess
characteristics that affect individuals' behavior in simple, widespread
ways. Given features of the common
constitution and circumstances of individuals, such alterations at the social
level produce regularities of behavior at the individual level that eventuate
in new social circumstances. S1 {structured environment of
individual choice} S2. Theda Skocpol's
causal analysis of the state and revolution, then, does legitimately attribute
causal powers to the state. But
these causal powers derive entirely from the ways in which the institutions of
the state assign incentives, powers, and opportunities to various individuals.
Do
causal powers depend metaphysically on the existence of law-governed regularities? And does our knowledge of causal
properties depend on the discovery of strong regularities? I argue that the answer to both
questions is negative. The causal
powers of a thing rather give rise to whatever regularities are observed; and
the discovery of regularities is only one out of a number of methods by which
we can identify causal relations and powers. Regularities are symptomatic rather than criterial of causal
powers and relations.
Social
causal ascriptions depend on regularities; but these are not generally social
regularities, but rather lawlike characteristics of agents (e.g. the central
axioms of rational choice theory).
I would assert, then, that the rock-bottom causal stories--the governing
regularities for the social sciences--are stories about the characteristics of
typical human agents. The causal
powers of a particular social institution--a conscription system, a revenue
system, a system of democratic legislation--derive from the incentives, powers,
and knowledge that these institutions provide for participants.
Moreover,
these phenomenal regularities are weak, tendential, conditional, and
unreliable. So social explanation
does not rest on discovering regularities, or deriving outcomes from statements
of regularities. Instead, analysis
of the underlying causal mechanisms, and particularly the microfoundations, is
central. Social scientists would
do better to embrace causal realism instead.
I
have argued for three central ideas to this point: there are social
regularities, but they are weak and not the central component of social
explanations; second, that the explanatory work of social inquiry commonly
takes the form of a search for causal relations and causal powers; and third,
that causal relations among social phenomena derive their necessity through
features of structured individual agency, and nothing else. Let us turn finally to the issue of
prediction: what consequences do these arguments have for the reliability of
predictions in the social sciences?
Some
philosophers and scientists maintain that it is essential to an adequate
science that it support predictions about the phenomena with which it is
concerned. The thrust of much that
has gone before in this paper points in the direction of rather narrow limits
on the feasibility of predictions in the social sciences. The fact of the complexity of causal
fields among social phenomena; the fact that social causal hypotheses are more
than usually burdened by extensive ceteris paribus conditions; and the inexactness of social causal
hypotheses; all these militate toward the conclusion that we must be skeptical
about the predictions advanced by social scientists.
What
is the status of prediction in the social sciences, given the analysis of social
regularities sketched out above?
There are generally three avenues through which scientific predictions
are generated. First, there are
predictions based on simple induction.
We note that low-income countries usually have high infant mortality, and
we predict that the as-yet unexamined low income country will have a high
infant mortality rate as well.
This sort of prediction depends on identifying a phenomenal regularity. Second, there are predictions based on
a theory of the governing regularities of the system in question. On my account, there are no such
regularities among social phenomena, so this form of prediction is unavailable
in the social sciences. Finally,
it is possible to support predictions in novel circumstances on the basis of an
analysis of the causal mechanisms that we can identify in the circumstance,
along with a model that permits us to attempt to estimate the aggregate effects
of these causal mechanisms. This
is essentially the strategy of attempting to work out the institutional logic
implicit in a given set of social arrangements. A special case of this last alternative is the construction
of abstract models (e.g. economic models) designed to capture certain social
mechanisms and permitting prediction of the future behavior of the system in
question.
So
social predictions may be based on phenomenal regularities; they may be based
on analysis of specific causal pathways (institutional-logic explanation); and
they may be based on abstract models of the workings of social sub-systems. What is the epistemic status of these
sorts of predictions in the social sciences?
Begin
with phenomenal regularities as a basis for prediction. According to the analysis provided
above, phenomenal regularities are inductively discernible patterns that derive
from the underlying causal mechanisms.
The causal properties of social phenomena give rise to regularities, and
it is reasonable to conclude that they are law-like (in the sense of supporting
counterfactuals). The causal
properties of social institutions, and the micro-mechanisms that underlie them,
give rise to phenomenal laws, and these are the chief regularities identified
by social scientists--not governing regularities. E.g.:
bureaucratic states collect revenues efficiently
low GNP is correlated with high infant mortality
political development produces political
instability
What is the scientific interest of
such regularities? To begin, they
are not fundamental. Rather, we
are always well-served by seeking an account of the causal mechanisms that
produce them; and we will better understand the scope, reliability, and
variance of such regularities when we have a true theory of the underlying
causal mechanisms. And second, the
predictions that these sorts of regularities support are weak, unreliable, and
subject to ceteris paribus
conditions.
Second,
phenomenal regularities do support predictions. If we are confident that there is a strong association
between GNP per capita and infant mortality, then we are justified in
predicting that if India's GNP per capita were to rise, its infant mortality
rate would fall. But predictions
based on such regularities are limited in just the way that any other purely
statistical association is limited.
The fact that 75% of philosophy professors enjoy Chinese food gives
ground to a prediction that I, a philosophy professor, enjoy Chinese food. But we would be more confident in such
a prediction if we knew more about the circumstances that account for the brute
generalization. (If, for example,
the explanation was that 75% of professors received their graduate education in
cities with excellent Chinese restaurants, this would directly causally account
for the statistic. And it would
likewise allow us to account for the failures of the generalization; professors
who were self-taught would not be subject to the same causal influence.)
Turn
now to the second possible source of predictions in social science: predictions
that depend on a hypothesis about the full set of important causal factors in a
given situation, and an analysis of the likely aggregate causal consequences of
these factors. A common source of
failures of prediction in the social sciences stems from the fact that causal
hypotheses and models are generally subject to ceteris paribus conditions.
Predictions and counterfactual assertions are advanced conditioned by
the assumption that no other exogenous causal factors intervene; that is, the
assertive content of the hypothesis is that the social processes under analysis
will unfold in the described manner absent intervening causal factors. But if there are intervening causal
factors, then the overall behavior of the system may be indeterminate. In some cases it is possible to specify
particularly salient interfering causal factors (e.g. political
instability). But it is often
necessary to incorporate open-ended ceteris paribus conditions as well. And in real situations it is all too common that the ceteris
paribus conditions of a given analysis turn
out not to be satisfied. This
means, in turn, that predictions based on such analysis must be understood as
representing tendencies rather
than probable outcomes.
A
related problem stems from the general point that causal hypotheses and models
unavoidably make simplifying or idealizing assumptions about the populations,
properties, and processes that they describe. Consumers are represented as possessing consistent and
complete preference rankings; firms are represented as making optimizing
choices of products and technologies; product markets are assumed to function
perfectly; and so on. Suppose that
an economic model makes the assumption that the coefficients of production are
constant. This implies that
producers do not alter production technologies in the face of different price
schedules for inputs. This
assumption abstracts from producers' substitution behavior. But the model-builder may argue that
this is a reasonable approximation in a static model; whatever substitutions
occur from one period to the next will be small and will have little effect on
aggregate input-output relations.
A
third reason why social predictions often fail has to do with the complexity of
causal fields in social phenomena.
This point may be put in terms of the idea of a set of INUS conditions
(or a causal diagram); the true INUS conditions for a given social phenomenon
are generally very complex, with various kinds of conjunctural causation
leading to complex and conditional relations between causes and effects. To the extent that a given causal
hypothesis has only identified some of the causal conditions included in the
true underlying causal diagram, it is foreseeable that predictions based on the
hypothesis will often go wrong.
Putting the point in another way, to the extent that a given causal
analysis does not provide a complete representation of the full causal field,
its predictions may be expected to fail on occasion. But it is highly implausible to suppose that we have ever
arrived at a complete causal field.
Fourth,
the uncertainties implied by incomplete causal fields are increased by the fact
that the underlying governing regularities are not deterministic. They are--as argued above--regularities
of agency. If an economist is
presented with a market in which there are substantial price differentials, he
or she will predict that normal competitive processes will bring about an
equilibrium price which is equal throughout the market. This prediction depends on the idea
that rational consumers will take advantage of lower-price opportunities, thus
forcing high-price providers to lower their prices. But if consumers are imperfectly rational--if, for example,
they are satisficers rather than optimizers--then this prediction will not
materialize.
In
short, then, there are numerous reasons why we should be cautious in assessing
the reliability of predictions based on a causal analysis of a social
situation. This circumstance does
not undermine the scientific value of the analysis, it should be noted; but it
substantially undermines our confidence in the predictions the analysis gives
rise to.
Turn
now to predictions based on models and theories. The overall strategy is to arrive at a formal representation
of what we think (some of) the underlying causal processes are within a given
social context, and then use deductive and mathematical tools to predict future
states of the system. This approach
suggests that we can ask two sorts of questions about a model. We can ask whether the model is a good
approximation of the underlying social reality--that is, the approximate truth
of the theory or model. Likewise,
we can ask whether the theory or model gives rise to true predictions about the
future behavior of the underlying economic reality (subject to the time frame
of the analysis). Each of these
questions falls on the side of the truth value of the model. Another set of questions concerns the
warrant of the model: the strength of the evidence and theoretical grounds
available to us on the basis of which we assign a degree of credibility to the
model: does available evidence give us reason to believe that the model is
approximately true, and does available evidence give us reason to expect that
the model's predictions are likely to be true? These questions are centrally epistemic; answers to them
constitute the basis of our scientific confidence in the truth of the model and
its predictions.
It
is important to see that the question of the approximate truth of a model is
separate from that of the approximate truth of its predictions. It is possible that the model is
approximately true but its predictions are not. This might be the case because the ceteris paribus conditions are not satisfied, or because low
precision of estimates for exogenous variables and parameters leads to
indeterminate predictive consequences.
Therefore it is possible that the warrant attaching to the approximate
truth of the model and the reliability of its predictions may be
different. It may be that we have
good reason to believe that the model is a good approximation of the underlying
economic reality, while at the same time we have little reason to rely on its
predictions about the future behavior of the system. The warrant of the model is high on this account, while the
warrant of its predictions is low.
Points
considered above concerning the need for ceteris paribus conditions, the need for simplifying assumptions,
and the weak necessity associated with the postulated governing regularities
entail that the predictions based on formal models may have substantially lower
warrant that the models themselves.
But there is another important consideration that is more specific to
economic and social modeling techniques that undermines the warrant of
predictions in the social sciences.
Theories give rise to models and models produce predictions. But there is a specification problem at
two levels: the model can specify the theory in various ways, leading to
different predictions. And the
parameters of the model themselves can be estimated in various ways--again
producing different predictions.
Consider the example of general equilibrium theory. General equilibrium theory represents
the general hypothesis underlying applied general equilibrium models. But the application of the theory to a
particular economy or policy problem is not straightforward. There is no canonical mode of
representing the central economic quantities and processes. Thus utility functions can be
represented in a variety of ways, and likewise with consumption and production
functions. (The linear expenditure
system is commonly used in computable general equilibrium models to represent
consumer demand, in large part because this is a highly tractable
formulation. But there are
alternative non-equivalent formulations available.) So a given model represents one out of many different
possible ways of implementing the general theory; and in order to arrive at an
overall judgment of the credibility of the model we need to assess the adequacy
of its particular implementation of supply, demand, savings behavior, and the
like.
It
follows from this observation that the specifics of a given model are not
deductively entailed by the economic theory that underlies it. Different model-builders can have equal
commitment to the general theory, while providing very different formulations
of the central economic processes (e.g., utility functions, production functions,
and demand functions). And the
resulting models may have significantly different properties, giving rise to
different predictions about the behavior of the economic system in
question. (It is for reasons of
this sort that Daniel Hausman refers to general equilibrium economic theory as
an "inexact science;" Hausman 1992.)
Another
reason why social phenomena often do not admit of confident prediction derives
from the possibility of multiple equilibria and hysteresis. There are multiple equilibria in a
problem of rational choice in circumstances where there are two or more
positions in which each player's strategies remain unchanged, even in full
knowledge of other players' strategies.
In circumstances where there are multiple equilibria, game theory does
not entail which equilibrium will emerge. Hysteresis refers to the possibility that the actual
equilibrium is path-dependent: factors extraneous to the rational
decision-making of the participants determine which point of equilibrium is
selected. An example of a social
situation in which there may be multiple equilibria is the dynamics of
population size, technical change, and standard of living. Rapid population growth and rapid
technical change produce an equilibrium in which there is a stable and high
standard of living (a high-level equilibrium); whereas rapid population growth
and slow technical change may lead to a low-level equilibrium. The rate of technical change, however,
depends on the rate of population growth, the current level of the standard of
living and exogenous variables; likewise,
the rate of population growth may be affected by the other two variables. So it is entirely possible that wholly
extraneous historical circumstances--an external economic shock, a navigational
discovery, a well-timed period of agricultural abundance, or a new type of
food--may determine which path is selected--thus determining as well the
eventual equilibrium outcome. (See
Elvin 1973, for an argument to the effect that traditional China was caught in
a high-level equilibrium trap.)
These
considerations suggest that the scope of confident prediction in the social
sciences is rather limited. We can
often be much more confident in the approximate truth of the hypotheses,
theories, and causal models that we put forward of underlying social processes
than in the predictions about the future that these hypotheses give rise to.
If
we were to adopt either the hypothetico-deductive model of confirmation or
Popper's falsifiability requirement on theory acceptance, this would be a very
serious blow to the scientific standing of social science, since on those
accounts, the chief source of empirical evaluation for scientific theories is
to be found in the predictive consequences of these theories. This is not the place for me to
elaborate the theory of empirical evaluation that accompanies this view in any
detail. Briefly, however, I
maintain that the empirical status of a social theory does not depend on its predictive consequences (Little 1986,
pp. 156-58). Instead, social
scientists are able to empirically evaluate their hypotheses and theories
piecemeal. The unreliability of
social predictions, then, does not mean that social science hypotheses cannot
be empirically evaluated. Instead,
social scientists often make use of a technique of empirical evaluation that
Mill described as the "deductive method": evaluation of social
science theories in terms of the independent support available for their
central hypotheses (Mill 1950, books III:XI and VI: IX).[19]
The
general point of this section, then, is that there are diverse but converging
reasons for being skeptical of the precision and reliability of the predictions
produced by social science research.
The
upshot of these arguments is relatively clear. My central conclusion is that there are no governing social
regularities underlying social phenomena.
There are governing regularities of sorts, but they are not social
(rather, they are regularities representing features of rational agency). And there are social regularities, but
they are phenomenal. Therefore the
social world is not a system of interrelated variables, concerning which we
might aim to discover the state laws.
It is rather a complex of processes subject to various causal
influences, conveyed by individual agency, onto diverse and rarely predictable
outcomes.
Inductive
regularities among social phenomena can be discovered. But these are distinctly phenomenal
laws rather than governing regularities; they have little explanatory import;
and they are not particularly reliable as a basis for prediction. Better on both counts is a theory of
the underlying causal mechanisms that produce them. These theories in turn need to be supported empirically. We can be most confident in statements
of lawlike regularities when we have an account of the mechanisms that underlie
them. This means that we need
microfoundational accounts drawn from rational choice theory, theory of
institutions, collective action theory, game theory, or microeconomics and microsociology.
Another
important consequence of this analysis is that the predictive capacity of the
social sciences is very limited.
It is certainly possible to make predictions based on "logic of
institutions" analysis, causal modeling, and the like, and it is
reasonable to do so for a variety of purposes. However, we should not have a great deal of confidence in
the resulting predictions. Causal
analysis is conditioned by ceteris paribus
clauses, incomplete causal fields, and other problems--with the result that
predicted outcomes of a given analysis may well fail to obtain because the
conditions are violated. And crude
inductive generalizations--e.g. "recessions during election years are
usually followed by change of party in office"--have limited applicability
to particular cases because of the degree of variance in the evidence that
supports them. So I conclude that
the search for lawlike generalizations that permit confident predictions is not one of the most central tasks for the social
sciences.
My
most central conclusion, however, is metaphysical. We ought not think of the social world as a system of
phenomena in which we can expect to find a strong underlying order. Instead, social phenomena are highly
diverse, subject to many different and cross-cutting forms of causation. So the result is that the very
strongest regularities that will be ever be discerned will remain the
exception-laden phenomenal regularities described here and the highly qualified
predictions of regularities that derive from institutional-logic analyses. There is no more fundamental
description of the social world in which strong governing regularities drive
events and processes.
Notes
References
Adelman,
Irma, and Cynthia Taft Morris:
1967, Society, Politics, and Economic Development: A Quantitative
Approach, Johns Hopkins Press,
Baltimore.
Boyd,
Richard, Philip Gasper, and J. D. Trout (eds.): 1991, The Philosophy of
Science, MIT Press, Cambridge.
Braybrooke,
David: 1987, Philosophy of Social Science, Prentice-Hall, Englewood Cliffs, N.J.
Cartwright,
Nancy: 1983, How the Laws of Physics Lie, Oxford University Press, Oxford.
Cartwright,
Nancy: 1989, Nature's Capacities and their Measurement, Oxford University Press, Oxford.
Coleman,
James S.: 1990, Foundations of Social Theory, Harvard University Press, Cambridge.
Davidson,
Donald: 1980, Essays on Actions and Events, Oxford University Press,
Oxford.
Elster,
Jon: 1989, Nuts and Bolts For the Social Sciences, Cambridge University Press, Cambridge.
Elvin,
Mark: 1973, The Pattern of the Chinese Past, Stanford University Press, Stanford.
Geertz,
Clifford: 1983. Local Knowledge, Basic,
New York.
Hausman,
Daniel: 1981, "John Stuart Mill's Philosophy of Economics," Philosophy of Science 48, 363-85 .
Hausman,
Daniel: 1992, The Inexact and Separate Science of Economics, Cambridge University Press, Cambridge.
Hempel,
Carl: 1942, "The Function of General Laws in History", in C. Hempel, Aspects
of Scientific Explanation, The Free
Press, New York.
Hobbs,
Jesse: 1991, "Chaos and Indeterminism", Philosophy of the Social Sciences 21, 141-164 .
Huntington,
Samuel P.: 1968, Political Order in Changing Societies, Yale University Press, New Haven.
King,
Gary, Sidney Verba, and Robert O. Keohane: 1991, Scientific Inference in
Qualitative Research, unpublished manuscript.
Kincaid,
Harold: 1986, "Reduction, Explanation, and Individualism", Philosophy
of Science 54, 492-513 .
Kincaid,
Harold: 1990, "Defending Laws in the Social Sciences", Philosophy
of the Social Sciences 20, 56-83.
Kitcher,
Philip, and Wesley C. Salmon (eds.): 1989, Scientific Explanation, Minnesota Studies in the Philosophy of Science, Vol. 13,
University of Minnesota Press, Minneapolis.
Knowles,
Dudley (ed.): 1990, Explanation and Its Limits, Cambridge University Press, Cambridge.
Levi,
Margaret: 1988, Of Rule and Revenue,
University of California,
Berkeley.
Little,
Daniel: 1986, The Scientific Marx,
University of Minnesota Press,
Minneapolis.
Little,
Daniel: 1989, Understanding Peasant China: Case Studies in the Philosophy of
Social Science, Yale University Press, New Haven.
Little,
Daniel: 1991, Varieties of Social Explanation: An Introduction to the
Philosophy of Social Science,
Westview Press, Boulder, Colorado.
MacIntyre,
Alasdair: 1973, "Is a Science of Comparative Politics Possible?", in
A. Ryan (ed.), The Philosophy of Social Explanation, Oxford University Press, Oxford
Mackie,
J. L.: 1974, Cement of the Universe,
Oxford University Press, London.
Mann,
Michael: 1986, The Sources of Social Power. A history of power from the beginning to A.D. 1760, Vol. one,
Cambridge University Press, Cambridge.
McIntyre,
Lee: 1991, Problems in the Philosophy of Social Science: Towards a Defense
of Nomological Explanation in the Social Sciences, Ph.D. dissertation, University of Michigan.
Mill,
John Stuart: 1950, Philosophy of Scientific Method, Hafner,
New York.
Miller,
Richard: 1991, "Fact and Method in the Social Sciences", in R. Boyd,
P. Gasper and J. D. Trout (eds.), The Philosophy of Science, MIT Press, Cambridge.
Nelson,
Alan: 1990, "Are Economic Kinds Natural?", in C. W. Savage (ed.), Scientific
Theories, University of Minnesota
Press, Minneapolis.
Putnam,
Hilary: 1975a, "The Analytic and the Synthetic", in H. Putnam, Mind,
Language, and Reality, Cambridge
University Press, Cambridge.
Putnam,
Hilary: 1975b, "Explanation and Reference", in H. Putnam, Mind,
Language, and Reality, Cambridge
University Press, Cambridge.
Railton,
Peter: 1981, "Probability, Explanation, and Information", Synthese, 233-56 .
Railton,
Peter: 1989, "Explanation and Metaphysical Controversy", in P.
Kitcher and W. Salmon (eds.), Scientific Explanation, University of Minnesota Press, Minneapolis.
Roemer,
John: 1982, "Methodological Individualism and Deductive Marxism", Theory
and Society 11, 513-520.
Rosenberg,
Alexander: 1988, Philosophy of Social Science,
Westview, Boulder, Colorado.
Ruben, David-Hillel: 1990, "Singular Explanation
and the Social Sciences", in D. Knowles (ed.), Explanation and its
Limits, Cambridge University
Press, Cambridge.
Rudner,
Richard: 1966, Philosophy of Social Science, Prentice-Hall, Englewood Cliffs, N.J.
Ryan,
Alan (ed.): 1973, The Philosophy of Social Explanation, Oxford University Press, Oxford.
Sahlins,
Marshall: 1976, Culture and Practical Reason, University of Chicago, Chicago.
Salmon,
Merilee H.: 1989, "Explanation in the Social Sciences", in P. Kitcher
and W. C. Salmon (eds.), Scientific Explanation, University of Minnesota, Minneapolis.
Salmon,
Wesley C.: 1984, Scientific Explanation and the Causal Structure of the
World, Princeton University
Press, Princeton.
Savage,
C. Wade (ed.): 1990, Scientific Theories, Minnesota Studies in the Philosophy of Science, Vol. 14,
University of Minnesota Press, Minneapolis.
Skocpol,
Theda: 1979, States and Social Revolutions, Cambridge University
Press, Cambridge.
Taylor,
Charles: 1985a, Philosophy and the Human Sciences: Philosophical Papers 2,
Cambridge University Press, Cambridge.
Taylor,
Charles: 1985b, "Interpretation and the Sciences of Man", in C.
Taylor, Philosophy and the Human Sciences, Cambridge University Press, Cambridge.
Thomas,
David: 1979, Naturalism and Social Science, Cambridge University,
Cambridge.
Turner,
Victor: 1974, Dramas, Fields, and Metaphors: Symbolic Action in Human
Society, Cornell University
Press, Ithaca, NY.
Weber,
Max: 1949, The Methodology of the Social Sciences, Free Press, New York.
Zuckerman,
Alan S.: 1991, Doing Political Science: An Introduction to Political
Analysis, Westview Press, Boulder, CO.
Author:
Daniel
Little
Department
of Philosophy and Religion
Colgate
University
Hamilton NY 13346
* I am grateful for comments and criticisms received from Harold Kincaid, Lee McIntyre, and Jonathan Jacobs on earlier versions of this paper.
[1] Social scientists who have taken this view include Huntington (1968); Adelman and Morris (1967); King et al; and Zuckerman (1991). Philosophers supporting the idea that lawlike generalizations must undergird social explanations include Hempel (1942), Thomas (1979), M. Salmon (1989), Kincaid (1990), and McIntyre (1991).
[2] Debate about the foundations of the social sciences among philosophers has largely divided between empiricist and interpretivist positions. Philosophers of social science proceeding from an empiricist perspective have generally taken the view that generalizations are essential; whereas interpretivists have maintained instead that singular interpretation of agent's meanings is essential. Thus Richard Rudner regards the discovery of prediction-supporting generalizations as the sole means of social explanation (Rudner 1966, pp. 59-67). David Braybrooke generally endorses this conclusion as well (Braybrooke 1987, pp. 21-29). The line of argument here is squarely within the empiricist tradition, but emphasizes causal realism rather than law-like generalizations.
[3] In addition to Geertz, the interpretive anti-generalization approach includes Charles Taylor (1985b), Alasdair McIntyre (1973), Turner (1974), and Sahlins (1976).
[4] What, then, is the right legacy? My position is that what is most worthwhile within the positivist program is epistemological, not metaphysical. The insistence on appropriate empirical controls on knowledge, the broad distinction between observation and theory, the emphasis on coherence and deductive closure--these are all abidingly important features of scientific knowledge. What does not endure, on my account, is a family of metaphysical doctrines implied by the doctrine of the unity of science.
[5] More extensive argument for this conclusion may be found in my Varieties of Social Explanation (Little 1991, chapter 2). Jon Elster takes a similar position in Nuts and Bolts for the Social Sciences (1989).
[6] Note, however, the powerful arguments against this general view put forward by Nancy Cartwright in How the Laws of Physics Lie (1983).
[7] For a superb review of the development and criticism of the covering law model of explanation see Wesley Salmon's "Four Decades of Scientific Explanation" in Kitcher and Salmon (eds.) (1989). David-Hillel Ruben offers an interesting account of the applicability of the covering-law model for the social sciences in "Singular Explanation and the Social Sciences" (Ruben 1990).
[8] Other philosophers of science have argued for a similar position in the past decade. See particularly Salmon (1984) and Cartwright (1983, 1989).
[9] There is of course still a question about the relation between simple causal relations and associated regularities. I place priority on the causal powers of the thing; nonetheless, I also recognize that if A has the causal power to bring about B, then there will be some regularity of association between A and B. But this regularity may not be observable, given data limitations. And the regularity is derivative, not constitutive, of the causal power.
[10] David Thomas's Naturalism and Social Science (1979) provides an extensive explication of naturalism. However, I find that his account gives too much emphasis to the idea of subsuming social phenomena under general laws.
[11] See note 17 for definition of an INUS condition.
[12] See the recent application of the central results of the theory of chaos to theories of explanation and prediction in the philosophy of science in Hobbs (1991).
[13] This distinction parallels Nancy Cartwright's distinction between fundamental laws and phenomenological laws (Cartwright 1983). Another term that might be used for the latter idea is a behavioral regularity; I have chosen not to use this alternative because of its suggestion of individual behavior.
[14] Further discussion of this point can be found in my Understanding Peasant China (1989, chapter 6).
[15] Alan Nelson argues a similar point with respect to economic kinds (Nelson 1990).
[16] I make the case for this view at greater length in Varieties of Social Explanation (1991, chapter 2). Richard Miller has advocated a similar conception of social explanation; he writes that "an adequate explanation is a true description of underlying causal factors sufficient to bring about the phenomenon in question" (Miller 1991, p. 755).
[17] The concept of an INUS condition is the centerpiece of John Mackie's analysis of causation. It is an "insufficient but necessary part of a condition which is itself unnecessary but sufficient for the result" (Mackie 1976, p. 62).
[18] Similar examples of arguments about the logic of power relations in pre-modern societies may be found in Mann (1986).
[19] See also Daniel Hausman's useful discussion of Mill's deductive method in "John Stuart Mill's Philosophy of Economics" (1981).