The notion of a theory is controversial in social science. A single and simple conception of theory is unlikely to apply across all fields, from mathematical economics to cultural anthropology. Still, construing theory broadly as any attempt to systematize and explain certain phenomena, it is clear that theories play a central role in social science. Many social-science pioneers, for example, Karl Marx, Emile Durkheim, Talcott Parsons, Sigmund Freud, Clark Hull, and Paul Samuelson, developed ambitious theories intended to explain a wide range of social phenomena. Today the tendency is toward more modest theories with narrower scope.
Philosophers of science have also adopted a more flexible and eclectic account of theorizing. For much of the twentieth-century, philosophers (and many social scientists) accepted the logical-positivist view that a theory is an axiomatized deductive system consisting of a few basic principles or laws (e.g., Newton’s laws of motion constitute his theory of universal gravitation). These principles contain a theoretical vocabulary describing entities that are often unobservable (e.g., electron, utility, social role), and also bridge laws that link the theoretical vocabulary with observable things. Such theories are tested by deriving predictions from basic principles, bridge laws, and statements describing the test situation, and then determining whether those predictions come true.
By the 1960s, every aspect of the positivist view was under attack, and today few philosophers accept it. The notion of laws has come to play a smaller role in philosophical discussion and that of models a larger one. But there remains no consensus about the nature of theories in the social sciences. Some are still expressed in formal, mathematical terms, with basic principles (axioms) from which predictions are deduced. But many are less formal, with a looser connection between theory, on the one hand, and explanation and prediction, on the other. However, even relatively modest theories can be illuminating. To the extent that a theory allows one to make predictions, it provides some measure of control over the social world. Moreover, some theoretical assumptions are needed to guide exploratory research, or even mere observation, since otherwise there are potentially an infinite number of things that might be relevant.
GOALS OF SOCIAL SCIENCE
Views about the nature of theories in social science go hand in hand with views about the nature of social science itself. Naturalists (e.g., in cognitive psychology) see the social and natural sciences as continuous in their goals and methods. They aim to explain human behavior by uncovering its causal mechanisms. As objects of study, however, people are distinctive because they think about, and so guide, their own actions. Given this sort of agency, some social scientists (e.g., in cultural anthropology) hold that mechanistic theories are inappropriate for studying humans. The point of social-scientific theories is, on this view, to explain actions by interpreting them so that they are intelligible.
Interpretive scientists hold that we already have a scheme for making sense of human action: People act to get what they want (feel obliged to pursue, and so forth), given what they believe. This common-sense pattern is so fundamental that it is considered to be more than a behavioral hypothesis—it is the standard of intelligibility for action. Interpretation is a sort of translation project—people read the text of other people’s actions. To understand their language one must empathetically immerse oneself in their concerns and situations. Like languages, conceptual schemes have conventions and norms, and fluency in both is the capacity to follow (and exploit) those conventions and norms.
Naturalists need not reject common-sense psychological principles, though historically some (e.g., behavior-ists) have. Now, however, most think that it is desirable to integrate reason-based behavior into the natural-scientific picture of the world, and so they seek to find the mechanisms that underwrite common-sense psychology. It now seems clear that a person’s reasons for acting are the causes of her behavior. Causal mechanisms may include preferences, values, habits, reinforcement histories, and the like.
Theories often include terms that ostensibly refer to entities involved in causal mechanisms (e.g., utilities, norms, information flows, social roles). So-called instrumentalists do not interpret such talk literally, but see it simply as a tool for predicting the behavior of individuals (organizations, institutions, societies, and others). They are agnostic about mechanisms—tools should pinpoint what will happen, but they need not say why. Interpreted instrumentally, however, theoretical constructs are fictions that can do no causal work. Hence, many philosophers and social scientists are realists about the entities posited by more successful theories.
Contrary to the views of many early-twentieth-century philosophers, causal explanation does not require appeal to, or even the existence of, precise causal laws. One can explain, for example, that the glass broke because someone dropped it, without knowing any general laws about glass or fragility. Of course some general knowledge about how things break is needed, but it need not add up to a precise exceptionless absolute (or precise probabilistic) law. Such generalizations as we do possess typically describe how a given construct or tendency would operate in isolation (e.g., rational choice theory describes how agents would act if they had clear preferences, coherent subjective probabilities, and acted to maximize their expected utility).
Behavioral mechanisms rarely operate in isolation. Outside the artificial environment of the laboratory, most social phenomena are produced by a wide variety of causal factors that differ from situation to situation and interact with each other in very complicated (frequently nonlinear) ways. Prediction in such cases is usually difficult. Even in physics, prediction is often next to impossible, because many systems of interest are nonlinear and hence chaotic. Pure, idealized cases can provide explanatory insight into the role mechanisms play in the generation of behavior despite their predictive impotence.
With the rejection of the positivist view that prediction and explanation are symmetrical (though with prediction occurring before a predicted event and explanation after the event explained), it is now clear that it is possible to have a theory that is good at one but not the other. For instance, theories that simply extrapolate correlations may be good at predicting without offering much explanatory insight into why the predictions come true. By contrast, many theories in the social sciences can offer explanations after the fact, even though they could not have predicted the phenomena beforehand. Typically they do this by pinpointing (or at least conjecturing about) the causal mechanisms that led to the thing to be explained (e.g., the bear market in U.S. stocks in the early 2000s). Even theories that are better at explaining than predicting are typically tested by their predictive power, however, and both explanatory power and predictive power are clearly desirable.
Because social-scientific theories rarely underwrite precise predictions (or postdictions), they are often difficult to test. In many cases theories yield only rough qualitative predictions, and these are typically compatible with a number of competing theories. Studies in the real world often allow us to discover correlations among variables (like years of education and income), and a theory gains some support when it makes reasonably precise predictions about the strengths of such correlations. But scientists are often interested in isolating causes, and, as the truism goes, correlation is not causation. Causes are traditionally discovered by carefully designed experiments whose outcomes are evaluated using null-hypothesis significance testing. Such testing, however, has come under increasing fire from methodologists such as Lisa Harlow, Stanley Mulaik, and James Steiger.
Social-scientific theories are useful when they help isolate the mechanisms that subserve behavior. Without some idea about when such mechanisms will be triggered and how they will interact, however, predictive ability is limited. Recent work, such as that of Judea Pearl, attempts to distill causal claims from real-world correlation data (plus a few assumptions about causation). Work on path analysis and other approaches to causal modeling will probably play an increasing role in testing theories in the future.
ADMISSIBLE THEORETICAL CONSTRUCTS
Debates in the social sciences often center around identifying appropriate theoretical constructs. One classical division distinguishes the Homo sociologicus and Homo economicus pictures of human beings. A third image— Homo psychologicus—might be added. The first picture holds, roughly, that people are shaped by their cultures and (mostly) act in accordance with social norms. Central constructs here include norms, values, and social roles. By contrast, Homo economicus is a rational calculator who (normally) acts so as to secure desired outcomes. Central constructs here include preferences and subjective utilities. Thus a sociologist might explain crime in terms of the norms of a criminal’s peer group, while an economist might argue that, in the agent’s milieu, crime is the most sensible behavior. Those favoring a Homo psychologicus account are more likely to employ information-processing constructs.
No single approach has been notably successful in explaining behavior on its own. Many social scientists agree that each account supplies part of the picture: Norms can influence preferences and beliefs; economic outcomes can influence the evolution of norms and social roles. In practice, however, cooperation among social sciences has proven elusive. Integrating the insights of sociology, economics, psychology, and other social sciences is difficult because each appeals to different properties and processes. Rather than attempt any integration, different theories are often offered as competitors. As a result, disciplines often tend to be identified more with their explanatory approaches than with specific areas of interest. Cross-disciplinary theorizing is an important frontier for social science.
Another historically important debate over appropriate constructs concerns the level at which theoretical concepts and generalizations should be framed. Social scientists disagree about whether their investigations should focus on individuals or groups, but clearly they often do care about the dynamics of social groups. Scientists want to know, for example, how price levels change with world events, not just how individual consumers react. When they look for patterns at this level of aggregation they often find them. Supply shortfalls in crucial commodities result in price increases. Researchers can often detect such regularities without considering the idiosyncrasies of individuals.
French philosopher Emile Durkheim went so far as to argue that there are autonomous social facts and that social explanations should cite social properties rather than appeal to the psychology of individuals. This view has led to much fruitless debate over reductionism and the existence of social facts. Recent philosophy may help to clarify these issues. Higher level patterns are best understood as equilibria resulting from a number of actions. In a market system, for example, general price increases result from the predictable decisions of numerous agents. Accounts of supervenience, such as those by American philosopher Jaegwon Kim, make clear that the truth of generalizations at one level of analysis (e.g., biology) may depend on that of generalizations at a lower level (e.g., chemistry) without being translatable into the latter. These accounts permit reconciliation of the claim that social phenomena would not exist without the actions of individual agents with the claim that many important generalizations can only be framed in vocabulary that does not involve specific individuals.