I picked up a copy of Frederick Schauer's Profiles, Probabilities, and Stereotypes because it looked very promising. The book is about generalization in legal reasoning, and I thought I'd learn something about methods. Schauer is Frank Stanton Professor of the First Amendment at Harvard's Kennedy School of Government, and author of several books in legal philosophy. I started the book in happy anticipation, rapidly became puzzled, then a little put off, and finally wound up just plain irritated.
Schauer starts with an important premise: traditionally, there's a tendency in legal thinking to downplay the importance of generalization, and this is wrong, because generalization is (and should be) an important part of legal thinking. So, for example, there's little doubt that pit bulls as a breed are more likely to attack and injure people than other breeds-- but what does this fact license in the way of remedial policies, legislation, or interpretation of individual cases? Clearly, it should make some difference, but it's not easy to specify what those differences should be. Schauer discusses many variations of the theme that generalizations are important. Along the way, he makes several mistakes which cloud his argument and draws conclusions which are troubling in their implications.
One point that Schauer makes is relatively unimportant in the larger scheme of things, although it stands out to someone comparing different kinds of work. Schauer argues that there's something special about the way that lawyers use and rely on generalization:
Unlike particularity however, generality is not important to law in just the same way that is important to medicine, psychology, social work, architecture, and criminal investigation. All these endeavors use rules, but they use them less, they use them less strongly, and they do not train their practitioners in using rules, even if they may at times train their practitioners to use particular rules. So too with reason-giving. All sorts of professions and human activities require the giving of reasons, but none obsesses about it the way that law does, and few-- philosophy may be an exception-- consider reason-giving a central feature of the enterprise. (p. 273)
This is an extraordinary set of claims, made more so by the complete absence of data, or any reference to data. The training programs of all the professions Schauer names spend a great deal of time teaching how to think about rules and their usage, as well as particular rules. In medicine, for example, the process is called "differential diagnosis", and it is considered the central skill of the art. So too with reason-giving. In addition to the professions Schauer names, each of the scholarly disciplines considers reason-giving to be central features of the enterprise. Indeed, the argument is often made that the capacity to think about rules and give reasons effectively (often called "critical thinking") is precisely the point of general education in the liberal arts.
But this over-crediting of one's own team is a relatively harmless excess of natural loyalty. I am more troubled by Schauer's confounding of generalization with other properties of reasoning, because the confounding leads to confusion, and perhaps to bad policy as well. In order to show this, let me start by listing a few distinctions.
Abstraction. Abstraction means thinking about or recognizing only some of the properties which a thing or group of things exhibits. For the purposes of designing a bridge, for example, I consider the weight and speed of vehicles on the bridge, but I don't bother with their color, or the fabrics of their upholstery. In that sense, I can work with abstract vehicles, which have only the properties relevant to my work.
Idealization. Idealization means leaving open the value of some property or characteristic. When I design a bridge, I abstract by deleting color from consideration-- my vehicles don't have any color at all. When I design a car however, it will have a color-- but I don't know what it is yet. For the bridge designer, color is abstracted; for the car designer, it is idealized.
Delimitation. Delimitation is the complement of abstraction-- it is the process of choosing which properties to recognize as significant or relevant. This is the process of choosing what sorts of evidence to include.
Specification. Specification is the complement of idealization. It is the process of putting particular values on attributes such as color. Ordering from a restaurant menu is the specification of a meal.
Unification. Unification means recognizing two (or more) abstractions as sub-kinds of a more abstract kind.
And finally:
Generalization. Generalization is the assertion of a relationship between two (or more) abstractions or idealizations. For example: pit bulls are more likely to bite than other breeds of dog.
All of these distinctions apply to kinds, sub-kinds and their instances. Abstraction, for example, creates more inclusive kinds; delimitation creates more finely differentiated and less inclusive kinds.
This is a lot of hair-splitting, but it's necessary, and Schauer doesn't do it reliably. Instead, he tends to use "generalization" to refer to all of these properties of reason indiscriminately. This leads to a variety of problems, as he shifts among different parts of his argument. Take the pit bull example again. This generalization (which relates the abstraction "pit bull" to the abstraction "attack") leads to a public policy problem: should pit bulls be banned or otherwise singled out for regulation? Attempts to ban pit bulls have been fought on the grounds that banning the class of animals (as opposed to individual offenders) is discriminatory and unfair. Framed this way, the debate is about appropriate degrees of abstraction, not of generality. Nobody questions the correlation between breed and aggressive behavior; rather, the argument is about whether (or how much) abstraction is appropriate in identifying aggressors: dogs as a species? Pit bulls as a breed? Pit bulls trained to be aggressive?
Meanwhile there's another group of problems: the issue of explaining particular cases, rather than dealing with policy problems. Given that I've been bitten by a pit bull, what should be done? Here, the issue of pit bull aggressiveness takes on a different cast; the issue isn't a matter of comparison with other animals, but rather of relevant particular circumstances: has the accused pit bull been trained to be aggressive? Was I provoking it? Was I in the dog's normal territory? Coming to a decision certainly involves the use of generalizations, but they are used in the service of appropriate specification, not abstraction as in the public policy case.
But Schauer doesn't use pit bull attacks to treat these questions of specification. Instead he shifts to the example of a driver run off the road by a bus. If the driver is unable to describe the bus accurately, she cannot use the fact that a certain company operates most of the buses in the region as an argument for recovering damages from that company. Rather, there must be additional evidence about the particular situation that goes beyond the "generalization" (actually, a deductive inference) that the bus was probably operated by the company. The fact that the courts insist on such evidence is not hostility toward generalization; it is appropriate respect for specification. Schauer has been waylaid by his respect for decision theory, which is based on probability and statistics. But law and industrial engineering (the primary users of decision theory) differ in important respects, and transferring methods between them will not be a straightforward proposition.
All of these issues are interesting, but they don't invalidate Schauer's main point, which is that generalization is an important part of legal and policy reasoning, and we are well-advised to take it seriously. But there is one further confounding in Schauer's argument which disturbs me.
In his last chapter, Schauer applies his concept of generalization to the notion of "community". A community, for Schauer, generalizes by imposing a kind of similarity on its constituents. He uses the example of variations in European cheese-making practices. Some kinds of cheese (Roquefort and Camembert, for example) are made with raw (unpasteurized) milk. Regulations of the European Union require stringent controls over the processing of such milk, in order to reduce risks of infection. These controls are expensive to implement, and many small cheese-makers using raw milk (who are disproportionately French) object to the regulations, arguing that their methods are safe. Schauer argues that the creation of a larger European community means eliminating some of the traditional regional differences in Europe, and that this elimination is a kind of generalization.
Perhaps the differences must be eliminated, but the process of doing so is not generalization, because it is not about kinds, sub-kinds, and instances; it is about parts and wholes. Let us call the process homogenization in honor of the fallen milk-handling procedures. This confounding of generalization and community is disturbing. To see why, consider the heart of Schauer's argument on the matter:
Generality and particularity are relative terms, and any larger group that encompasses a smaller one is for that reason more general, just as the smaller group is more particular. The category of American is more general than the category of Texan, the category of European is more general than the category of German, and the category of pitcher for the New York Yankees (of which there are about ten at any given time) is smaller than the category of New York Yankee (of which there are about twenty-five at any given time). As a result, whenever it is thought important to create a community around the larger grouping, that grouping will serve as a generalization excluding otherwise germane variations among individuals or subgroups. (Pp. 287-88)
This passage contains two major logical mistakes. First (and less important) is the size issue, as embodied in the Yankee example. We may consider pitchers to be a sub-kind of players. Hence, "player" is more idealized than "pitcher"-- if we're told someone is a player, we know he has a position, but we don't know what it is. On the other hand, if we're told he's a pitcher we do. This is about the definitions of "player" and "pitcher" (what philosophers call intention) and has nothing to do with how many there are of each kind (what philosophers call extension). Idealizations and abstractions are about intensions, not extensions. Generalizations can be about either.
More important, Schauer here confuses the relationship between parts and wholes with the relationships among kind, subkinds, and instances. The notions of abstraction, idealization, and generalization do not apply to the relationship between parts and wholes. For example, a hand is a kind of organ. "Organ" is more abstract than "hand". In contrast, my hand is part of my arm, but both are equally concrete and particular. Arguments which relate parts to wholes are not generalizations, because they speak of concrete things.
When Schauer says that the category American is more general (meaning, larger in extension) than the category Texan, he is correct in the sense that instances of the kind "Texan" are instances of the kind "American" and there are other instances of the kind "American" that are not instances of the kind "American". But that's not what we mean when we talk about generalization and that's not what we mean when we talk about "community". When we speak of communities, we imply that there are relationships among the members, and between the members and the whole. In short, in communities the relationship is one of parts to whole, not instances to kinds. This is how we typically understand the idea that America "encompasses" Texas.
The confounding here is important, because the analogy goes as follows: more general (i.e., abstract) = fewer properties = fewer subkinds and/or less variation among instances; hence, community = increased homogeneity and less variation among members. Assuming that creating a community implies a kind of (abstracting, homogenizing) generalization is far too limited a view, one which is too easily used to justify arbitrary coercive restrictions on conduct. I doubt Schauer would go along with that, but nothing says that others can't misuse his ideas or ignore his good advice.
Besides, there's a better way of thinking about it. Community is not about kinds, it is about parts. Hence, creating or improving communities does not consist of eliminating inconvenient parts or differences among them, but of making the parts work together better.
Comments