Introduction: The difficulties that may be barriers

From Parts I and II to the Conclusion

We considered recent social “experiments,” such as the four income maintenance studies, as experiments in two senses. In Part I we viewed them as the sort of random-assignment experiments that R. A. Fisher introduced into agronomy in the 1920s. In Part II we considered them as the kind of tests of hypotheses-deduced-from-theory that Karl Popper argued, beginning in the 1930s, were the hallmark of physics and science more generally. The first sense of experiment is that of what one could call an empiricist vision of science, whereas the second is that of what one might label a rationalist vision of science. We argued that these visions of science were problematic accounts of agronomy and physics experiments, that the problems could be catalogued, and that reconstructions of recent social experiments in accordance with the two visions of science possessed the same catalogue of problems and then some. If the social experiments were not science, we argued, the reasons they were not might be found in the “and then some” portion of their reconstructions' catalogue of problems.

At the base of the empiricist vision of science, we argued, was the drive to know the object of inquiry on the basis of observation alone.

Introduction: Microeconomics as “normative science”

Chapter 5 catalogued the ways in which a standard set of classical mechanics experiments failed to measure up to Popper's rationalist ideal notion of science. Chapter 6 argued that hypothetical microeconomic experiments of necessity fail in the same ways as the classical mechanics experiments to be science in Popper's sense. But Chapter 7 argued that a set of actual microeconomic studies – the income maintenance experiments – fail in a whole additional set of ways to be science in Popper's sense. So if the income maintenance experiments are the best scientific studies microeconomists can offer, then microeconomics appears to be at best an undeveloped or underdeveloped science.

The nineteenth-century political economist John N. Keynes distinguished two senses of the term “science” in economics:

On Keynes's distinction, our argument in Part II so far has been that microeconomics is an undeveloped or underdeveloped positive science. In Chapter 8 we argue that microeconomics is also an undeveloped or underdeveloped normative science.

Conceptual Anomalies in Economics and Statistics synthesizes close to thirty years of study, research, and teaching in a variety of academic areas. Teachers, students, family members, an editor, schools, and funding institutions contributed indirectly to the writing of this book and deserve recognition.

Though I didn't realize it then, I started preparing to write this book at Chicago's Hyde Park High School during 1956–60, where Mrs. Eva Shull sparked the interest in mathematics evident in nearly every chapter.

During 1960–5, I studied humanities and science at the Massachusetts Institute of Technology (M.I.T.), concentrating in literature (with some philosophy) and mathematics (with some physics). Literature courses with Benjamin DeMott, Norman Holland, Louis Kampf, and William Harvey Youngren helped me develop the textual analytic writing style that permeates this book. A philosophy course with John Rawls introduced me to the work of David Hume, a discussion of which opens Chapter 4, and inspired me to study Mill and the modern philosophers on whose writings the book draws. Gian Carlo Rota's course in combinatorial analysis underpins Chapter 2. Courses in elementary classical, advanced classical, and quantum mechanics from Alan Lazarus, R. H. Lemmer, and Irwin Pless, respectively, suggested some of my examples in Chapter 5.

During 1965–7, I was a graduate student in mathematics at Northwestern University, and during 1967–8, I taught mathematics at Chicago City College, Southeast Branch.

Introduction: An ancient example

Writing in 1949, the psychologist Richard L. Solomon observed: “The history of the idea of a controlled experiment is a long one. Usually one goes back to J. S. Mill's canons for the concept of experimental controls” (p. 137). Recent historical research, however, reveals that one has to go a good deal further back than Mill for the origin of controlled experimentation. Consider the following passage from the Old Testament:

Introduction: Hume's problem in the modern approach

In the first three chapters we have given considerable attention to the notion of causality. But Hume originally came to the notion of causality by posing another question. He first asks himself, “What is the nature of all our reasonings concerning matters of fact?” and answers, “they are founded on the relation of cause and effect.” He then asks, “What is the foundation of all our reasonings and conclusion concerning that relation [of cause and effect]?” and answers, “in one word, experience” (1955, p. 46). Thus, in arguing that our reasoning about matters of fact is based on “in one word, experience,” Hume appears to adopt an empiricist perspective.

But Hume's empiricism is not as thoroughgoing as that of his successor Mill. Hume continues his argument:

Hume then anticipates and rejects as circular Mill's later argument that Induction can be based on induction (see Section 1.2.2).

Introduction: Addressing some possible objections

In the next two chapters we seek some tentative answers to questions about the nature of microeconomics as science. Is microeconomics science in the same sense as classical mechanics? If not, is it science in some other sense? Were the income maintenance experiments science experiments in some important sense? If microeconomics is not presently science, could it become science? What are the barriers to its becoming science? Does microeconomics face possibly insurmountable barriers to becoming science? Our approach to answering these questions will be first to attempt to reconstruct microeconomics, while noting differences between the developing reconstruction and the previous chapter's account of classical mechanics, and then try to decide if the noted differences demarcate a science from a nonscience. At least two sorts of objections can be raised to our questions and approach to answers.

The first sort of objection runs roughly as follows: Because microeconomics is in fact science like classical mechanics, our first question has been answered in the affirmative, and so our other questions don't apply. Consider, for example, a remark by Popper: “It must be admitted, however, that the success of mathematical economics shows that one social science at least has gone through its Newtonian revolution” (1964, p. 60). Yet Popper never details “the success of mathematical economics”; so his remark is merely a claim, without argument or supporting evidence, that mathematical economics is a science on a par with classical mechanics.

Introduction: Some philosophers on the logic of science and classical mechanics

From Mill to Popper

Our ultimate aim in Part II is to understand the sense in which microeconomic studies such as the income maintenance experiments may perhaps be science experiments, or realizations of a rationalist program of social inquiry. To achieve that aim, it will help to first develop a concept of physical science with which to compare our income maintenance examples. Just as in Chapters 2 and 4 we focused on the logics of statistical inference, here we shall consider the logic of classical mechanics experiments. We develop our view of the logic of such experiments by distillation of the views of J. S. Mill, Karl Popper, and Thomas Kuhn on science generally and through close scrutiny of some examples from classical mechanics.

In explaining a feature of his Deductive Method, Mill offers an embryonic rationalist account of science:

On Mill's account, then, one begins (at least on reconstruction) with a set of propositions, deduces some conclusions, and compares the deduced conclusions with observations or perhaps with observed regularities (= “empirical laws”). So on Mill's view, a science experiment seems unproblematic.

Introduction: Fisher and the emergence of the modern approach

Assessing Fisher's historical role

In a 1976 paper on R. A. Fisher's work, the statistician L. Savage remarks: “Fisher is the undisputed creator of the modern field that statisticians call the design of experiments” (1976, p. 450). What does it mean, though, to say that Fisher “created” a field? Work on techniques of agricultural field experimentation began with Bacon in 1627. By the end of the nineteenth century, that work included such modern techniques as matched-pairs designs, use of (at least particular) Latin-square layouts, and replication. And the use of probabilistic methods in astronomical observations began with the work of Legendre (in 1806) and Gauss (in 1809) on the theory of errors. By the end of the nineteenth century, these probabilistic methods included such staples of modern analysis of variance as fixed-effects and variance-components models. Finally, Darwin, in 1859, made the roughly modern notion of variation and variability central to subsequent work in evolutionary biology. By the dawn of the twentieth century there were three independent strands of research (in agriculture, astronomy, and biology), each containing elements of the statistician's modern experimental design.

In the early part of the twentieth century the three independent strands of research began to come together. In 1908, “Student” proposed his t test, worked out some of its probability distributional properties (thus extending the theory of errors), and illustrated its use with a matched-pairs design.

Introduction: The nature of the object of inquiry and difficulties for causal inference

Fisher first employed his technique of randomization, together with the theory of causal inference formalized in the previous chapter, in agricultural experiments, where the group members were fields, or sections of fields, of crops, and the treatments were various sorts or amounts of soil enrichment or fertilizer. Only with concepts of “field” and “crop” radically at odds with common sense, then, could we say that Fisher's objects of inquiry were conscious or willful. Within a few decades of its introduction, though, randomization became a standard feature of biomedical drug-testing experiments, where the group members are human subjects and the treatments are various sorts or amounts of drugs. Thus, the human subjects of such drug-testing experiments are certainly conscious, in any ordinary sense of the term, and also creatures of will.

Physicians had early recognized that subject awareness in a medical experiment in which subjects had to be relied on to report results might lead to difficulties. For example, knowledge that they were taking a new aspirin tablet might lead subjects to report greater relief from their headache conditions than had actually occurred. To counteract such misreporting, physicians employ a so-called double-blind experiment, in which a treatment group receives the aspirin, and the control group gets a “placebo,” and neither subjects nor physicians know which group subjects are in. Recent historical research illuminates the origin of such techniques:

Introduction

In previous chapters we considered the role of social information in such matters as the response to surveys or polls. The possession of information about the attitudes of others did not, in this work, affect the underlying individual attitudes, which were assumed to be immutable with respect to social information. Suppose, however, that knowledge of the attitudes or intentions of others affects our own intentions. Then the purveyors of social information evidently acquire considerable power to influence events. Such a contingency has long been noted by political pollsters, for instance. The bandwagon influence of surveyed intentions upon election outcomes is one such effect. Moreover, campaign money and activity is heavily dependent upon the performance in current opinion polls of the competing candidates, and the effects of such campaigning will in turn materially affect voter intentions. That many commentators – and indeed legislators – have expressed grave disquiet at the consequential effects of polls as news events is ample testimony to the pervasiveness and importance attached to the nonneutral effects of surveys of voter opinion. More recently it has been realized (see Section 4.3) that opinion polls may have a social function in transferring information about candidates to the electorate and vice versa so that the results of a cycle of polling and publication may be socially beneficial, a cause for encouragement rather than disquiet.

The behavior-modifying influence of published predictions is by no means confined to the political arena.

Introduction

Consider the following questions, all taken from the 1986 Australian Census:

1. Q.17: Does the person speak a language other than English at home?

2. Q.18: How well does the person speak English?

3. Q.24: What is the gross income (including pensions and/or allowances) that the person usually receives each week from all sources?

Q.33: In the main job held last week, how many hours did the person work?

Legal strictures notwithstanding, each of these questions is subject to the suspicion of untruthful responses for strategic reasons. Such questions have to be considered in their social context. Questions 17 and 18, which refer to aspects of migrant assimilation, have to be set against a continued and at times strident debate on immigration and its ethnic and social composition. Households continuing to speak Chinese or Italian at home might well choose not to reveal this. Question 24 should be set against prospective legislation on taxation policy or pension availability and general concern by social reformers with economic inequality as well as the traditional antipodean obsession with “tall poppies.” Finally, those in the public service, including academics and schoolteachers, have a motive to overstate their hours of work in question 33 since salaries and other benefits are ultimately tied to the public's perception of the demands on the time of their servants.

Getting people to truthfully reveal their preferences or intentions has long been recognized as a problem in certain fields of economics, in particular in the theory of the provision of public goods.

Introduction

In the previous chapter we described the idea of a rational-expectations equilibrium. The reader will recall that such an equilibrium is characterized by equality between the subjective expectations of participants and mathematical expectations given their respective information sets. All predictions are in this sense self-fulfilling. One might suspect from the anticipatory nature of such systems that some problems might arise for the theory of social or economic policy. This does turn out to be the case, and the early part of this chapter is concerned with the resulting problems. To begin with, a degree of care must be exercised that policy rules are based on the real invariants of the system and not upon the kind of reduced form that is usually taken to govern the system's evolution. Moreover an anticipatory system implies a fair measure of interdependence between the action of the policymaker and the actions that he or she is trying to influence, introducing the possibility of game-theoretic interactions between policymaker and subjects. The additional complexity of the resulting control problems together with the necessity of identifying those parameters that remain invariant under different controls impose an estimation task of considerable magnitude for the policymaker. All the preceding issues are discussed in Sections 6.2 and 6.3. Even if these problems can all be satisfactorily resolved, an additional difficulty arises for the application of standard control-theoretic methods: the problem of time inconsistency, discussed in Section 6.4.