Economists, like other social scientists, are sometimes confronted with the charge that their discipline is not a science. Human behaviour, it is said, cannot be analyzed with the same objectivity as the behaviour of atoms and molecules. Value judgments, philosophical preconceptions, and ideological biases unavoidably interfere with the attempt to derive conclusions that are independent of the particular economist espousing them. Moreover, there is no realistic laboratory in which economists can test their hypotheses.

In response, economists are wont to distinguish between “positive economics” and “normative economics.” Positive economics seeks to establish facts: If butter producers are paid a subsidy, will the price of butter be lowered? Will a rise in wages in the automotive industry reduce the employment of automobile workers? Will the devaluation of currency improve a country’s balance of payments? Does monopoly foster technical progress? Normative economics, on the other hand, is concerned not with matters of fact but with questions of policy or of trade-offs between “good” and “bad” effects: Should the goal of price stability be sacrificed to that of full employment? Should income be taxed at a progressive rate? Should there be legislation in favour of competition?

Because positive economics in principle involves no judgments of value, its findings may appear impersonal. This is not to deny that most of the interesting economic propositions involve the addition of definite value judgments to a body of established facts, that ideological bias creeps into the very selection of the questions that economists investigate, or even that much practical economic advice is loaded with concealed value judgments (the better to persuade rather than merely to advise). This is only to say that economists are human. Their commitment to the ideal of value-free positive economics (or to the candid declaration of personal values in normative economics) serves as a defense against the attempts of special interests to bend the science to their own purposes. The best assurance against bias on the part of any particular economist comes from the criticism of other economists. The best protection against special pleading in the name of science is founded in the professional standards of scientists.

If the science of economics is not based on laboratory experiments (as are the “hard” sciences), then how are facts established? Simply put, facts are established by means of statistical inference. Economists typically begin by describing the terms they believe to be most important in the area under study. Then they construct a “model” of the real world, deliberately repressing some of its features and emphasizing others. Using this model, they abstract, isolate, and simplify, thus imposing a certain order on a theoretical world. They then manipulate the model by a process of logical deduction, arriving eventually at some prediction or implication that is of general significance. At this point, they compare their findings to the real world to see if the prediction is borne out by observed events.

But these observable events are merely a sample, and they may fail to represent real-world examples. This raises a central problem of statistical inference: namely, what can be construed about a population from a sample of the population? Statistical inference may serve as an agreed-upon procedure for making such judgments, but it cannot remove all elements of doubt. Thus the empirical truths of economics are invariably surrounded by a band of uncertainty, and economists therefore make assertions that are “probable” or “likely,” or they state propositions with “a certain degree of confidence” because it is unlikely that their findings could have come about by chance.

It follows that judgments are at the heart of both positive and normative economics. It is easy to see, however, that judgments about “degrees of confidence” and “statistical levels of significance” are of a totally different order from those that crop up in normative economics. Normative statements—that individuals should be allowed to spend income as they choose, that people should not be free to control material resources and to employ others, or that governments must offer relief for the victims of economic distress—represent the kind of value judgments associated with the act of disguising personal preferences as scientific conclusions. There is no room for such value judgments in positive economics.

Most assumptions in economic theory cannot be tested directly. For example, there is the famous assumption of price theory that entrepreneurs strive to maximize profits. Attempts to find out whether they do, by asking them, usually fail; after all, entrepreneurs are no more fully conscious of their own motives than other people are. A logical approach would be to observe entrepreneurs in action. But that would require knowing what sort of action is associated with profit maximizing, which is to say that one would have drawn out all the implications of a profit-maximizing model. Thus one would be testing an assumption about business behaviour by comparing the predictions of a theory of the firm with observations from the real world.

This is not as easy as it sounds. Since the predictions of economics are couched in the nature of probability statements, there can be no such thing as a conclusive, once-and-for-all test of an economic hypothesis. The science of statistics cannot prove any hypothesis; it can only fail to disprove it. Hence economic theories tend to survive until they are falsified repeatedly with new or better data. This is not because they are economic theories but because the attempt to compare predictions with outcomes in the social sciences is always limited by the rules of statistical inference.

It is not remarkable that competing theories exist to explain the same phenomena, with economists disagreeing as to which theory is to be preferred. Much has been written about the uncertain accuracy of economists’ predictions. While economists can foretell the effects of specific changes in the economy, they are better at predicting the direction rather than the actual magnitude of events. When economists predict that a tax cut will raise national income, one may be confident that the prediction is accurate; when they predict that it will raise national income by a certain amount in three years, however, the forecast is likely to miss the mark. The reason is that most economic models do not contain any explicit reference to the passage of time and hence have little to say about how long it takes for a certain effect to make itself felt. Short-period predictions generally fare better than long-period ones. Since its development in the 1990s, experimental economics has, in fact, been testing economic hypotheses in artificial situations—often by using monetary rewards and student subjects. Much of this innovative work has been stimulated by the ascendance of game theory, the mathematical analysis of strategic interactions between economic agents, represented in such works as Theory of Games and Economic Behaviour (1944) by John von Neumann and Oskar Morgenstern. Aspects of game theory have since been applied to nearly every subfield in economics, and its influence has been felt not just in economics but in sociology, political science, and, above all, biology. Because game theory fosters the construction of experimental game situations, it has helped diminish the old accusation that economics is not a laboratory-based discipline.