You are on page 1of 22
Royce Singleton, Jr. and Bruce C. Straits, Approaches to Social Research, 3" ed., New York: Oxford University Press, 1999, pp. 40-62. 3 f The Logic of Scientific Reasoning Like any serious intellectual endeavor, science involves thinking and reasoning. In science, the reasoning process generally consists of attempts to draw conclusions about the order of the empirical world on the basis of observational evidence. As we study metheds of social research, we will be concerned mostly with how to gen- erate and how to judge the quality of scientific evidence. But ultimately, scientific assertions—in the form of theories and conclusions drawn from evidence—must be _ based on logical reasoning. To understand the reasoning that underlies theoretical explanations and the acceptance or rejection of hypotheses, we turn to logic, the study of the principles of reasoning. Most people recognize that certain ways of reasoning are acceptable or valid whereas others are unacceptable or invalid. The aim of logic is to state the princi- ples upon which we make this distinction. We will now introduce some of these principles for the purpose of applying them to scientific reasoning. Studying the principles of reasoning should further enbance your understanding and appreciation of how the game of science is played. It also should help you evaluate your own thinking as well as the persuasive messages that you encounter every day. Logic and Reasoning If we think of reasoning as an art or craft, then one of the true masters would have to be Sir Arthur Conan Doyle’s famous detective Sherlock Holmes. The logician Wesley Salmon (1973;1-2) relates how in one of Holmes’s adventures he comes across an old felt hat, whose owner is unknown to him. After carefully observing the hat, Holmes tells Dr. Watson several things about its owner, one of which is that he is highly “intellectual.” Puzzled, Dr. Watson asks Holmes about the basis for this assertion. In answer, Holmes puts the hat upon his head, where it comes down over his forehead and settles upon the bridge of his nose. “It is a question of cubic capacity,” he says. “A man with so large a brain must have something in it.” This example contains the basic ingredients for logical analysis. Holmes has Stated a conclusion along with supporting evidence. In his thinking he has pro- ceeded from certain beliefs that he holds to be true (the evidence) to a claim (a con- clusion) that he infers from them. Logic deals with the relation between evidence and conclusion (Salmon, 1973:3). Once an inference is made, logic can tell us whether the inference is correct, that is, whether the evidence properly justifies the conclusion. The Logic of Scientific Reasoning 41 Whenever we reason we proceed from certain information to conclusions based on that information. The point at which logical analysis begins is after the act of reasoning has taken place. Once an inference has been made, it can be transformed into an explicit set of statements and subjected to logical analysis. In this way, logic can clarify our thinking. However, logic does not provide a description of actual thought processes, with acts of thinking, imagining, and creating. It cannot tell us what to infer or how to think and reason, but it can reveal whether our reasoning is sound. Although Dr. Watson and Sherlock Holmes were inevitably in possession of the same information, it was always the clever Holmes who drew the inferences. But after Holmes had expressed his reasoning, Watson could, through his under- standing of the rules of logic, evaluate the correctness of the inference. In a simi- Jar manner, we may not be able to grasp how a particular scientific inference was made in the first place, but we can evaluate whether the inference is logically jus- tified. Elements of Logical Analysis We will now introduce some of the technical concepts and rules that logic supplies for investigating reasoning. This excursion into logic will be highly selective, as we identify a few logical skills that are essential for studying the particular forms of reasoning found in science. Of necessity we will sidestep many issues that would be dealt with in more advanced logical studies. As in the previous chapter, we do not intend to dwell on knotty issues, but rather to fill in some of the background in the picture of social research, to provide a context for understanding the specific guidelines and procedures presented in the ensuing chapters. The fundamental elements of logical analysis consist of terms, propositions, and arguments (Wheelwright, 1962). Let us examine the meaning of each of these elements as well as their relevance to science. Terms The simplest element of logical analysis is the term. A term is whatever is meant by a word or phrase. Some terms you will encounter in the following pages are an- tecedent, syllogism, valid argument, induction, and the fallacy of affirming the con- sequent. The distinctive feature of a term is its meaning. A term is neither true nor false. We can understand a term but cannot affirm or deny it. Terms are analogous to concepts in science. Propositions A proposition is an expression of a judgment about some term or terms. In the grammatical sense, it is a declarative sentence. Unlike terms, propositions are, by definition, either true or false. Logicians are interested in determining precisely what a given proposition says, that is, what is true or false according to the propo- sition, and in examining the logical role of propositions in reasoning. 42 SCIENTIFIC AND LOGICAL FOUNDATIONS OF SOCIAL RESEARCH There are several types of propositions in logical analysis. The traditional logic that stems from Aristotle deals primarily with what are called “categorical” propo- sitions; for example, “All men are mortal,” and “No events are uncaused.” In this chapter we will concentrate on what are known as “conditional” or “hypothetical” propositions.! A conditional proposition consists of two simple statements joined by the words “if” and “then.” For example, 1. If today is Friday, then tomorrow is Saturday. 2. If the reader has taken a course in logic, then this chapter is largely a review. In a conditional proposition, the part that is introduced by the “if” is called the an- tecedent and the part that follows the “then” is called the consequent. “Today is Friday” is the antecedent of statement 1; “the reader has taken a course in logic” is the antecedent of statement 2. “Tomorrow is Saturday” is the consequent of state- ment 1; “this chapter is largely a review” is the consequent of statement 2. A conditional proposition asserts that the antecedent implies the consequent, that the consequent is true if the antecedent is true. Beyond this common meaning, however, there are different kinds of relations between antecedent and consequent that can be asserted by conditional propositions. One kind of relation is definitional. In the conditional proposition “If the figure is a triangle, then it has three sides,” the consequent follows from the antecedent by the very definition of the word “tri- angle” (Copi, 1973:15). Another sense of the “if-then” connection is causal. “If this metal is immersed in nitric acid, then it will dissolve” asserts that the antecedent (immersing the metal in nitric acid) causes the consequent (the metal’s dissolution). Since scientists are concerned precisely with this kind of relation, the truth of which must be established empirically, our discussion of reasoning deals primarily with conditional propositions. Arguments In logic, an argument is a set of two or more propositions of which one is claimed to follow either necessarily or probably from the other(s). This technical definition of argument should not be confused with use of the word in everyday discourse to mean some sort of disagreement, quarrel, or debate. Constructing arguments is the logician’s way of making explicit a person’s reasoning from evidence to conclu- sion. Once an act of reasoning has taken place, the reasoning can be transformed into a group of verbal or written statements that constitute an argument. The state- ment that is claimed to follow from the others in an argument is, in fact, referred to as the conclusion. The other statements that supply evidence for accepting the conclusion are known as premises. The principal kind of argument studied by logicians is the syllogism. Syilo- gisms are arguments composed of three propositions: two premises and the con- clusion that the premises logically imply. A classic example of a syllogism, used by logicians for the past 2000 years, is The Logic of Scientific Reasoning B 3.1 All men are mortal. (PREMISE) Socrates is a man. (PREMISE) Therefore, Socrates is mortal. (CONCLUSION) Although the arguments implicit in any act of reasoning will often contain more than three propositions, longer arguments can always be broken down into a series of syllogisms. Hence, the syllogism is the basic unit of logical analysis. To analyze the logic of scientific reasoning we must first identify the syllogistic structure of that rea- soning and then examine the syllogisms with respect to their validity or invalidity. Whereas terms are judged as to their meaning, and propositions primarily as to their truth, syliogisms are judged primarily as to their validity (Wheelwright, 1962:14). The validity of a syllogism depends solely on the relation between its premises and its conclusion. In a valid syllogism the premises have the following relation to the conclusion: if the premises are true, then the conclusion must also be true. Notice from this definition that to assess the validity of a syllogism we do not need to know whether the premises or conclusions are actually true. We only need to know whether the conclusion would be true if the premises were true. Thus, it is possible to have a valid syllogism that consists entirely of false propositions: 3.2 All students are seniors. (FALSE) Some robots are students, (FALSE) ‘Therefore, some robots are seniors. (FALSE) It is also possible to have an invalid syllogism that contains exclusively true state- ments. One should be able to recognize, by common sense, the invalidity of the fol- lowing syllogism: 3.3 All butterflies can fly. (TRUE) All crows are birds. (TRUE) Therefore, all crows can fly. (TRUE) These two examples do not exhaust the possible combinations, but they should clar- ify that the validity or invalidity of a syllogism is independent of the truth of its premises. That is, we can logically analyze arguments without knowing whether the premises are true. In fact, example 3.2 shows that we can analyze syllogisms that we know have false premises (although we, as social scientists, would ordinarily leave this task to logicians). - Validity and Truth: Logic and Science With the aim of evaluating reasoning, logic can analyze relations among proposi- tions irrespective of whether the propositions are factually true or false or whether they say anything at all about reality. Science, by contrast, has the broader goal of establishing knowledge about the empirical world. To justify their conclusions about reality, scientists must evaluate both the adequacy of their reasoning and the actuality of their statements, that is, both validity and truth. In other words, to jus- tify conclusions, (1) the premises must be properly related to the conclusion so that 44 SCIENTIFIC AND LOGICAL FOUNDATIONS OF SOCIAL RESEARCH the argument is logically correct or valid, and (2) the premises must be true (Salmon, 1973:3-4). Logic is concerned with the first question only. Logic can tell us if, given X and Y, we can reasonably infer Z, but it cannot tell us if X and Y are true. It is up to scientific observation to make that determination. To clarify these two aspects of justifying conclusions, recall our Sherlock Holmes example. Holmes was interested in establishing as fact that the wearer of the hat was highly intellectual. To substantiate this assertion he presented an argument. Although we need not spell out his argument in complete detail here, let us exam- ine one key syllogism that the argument contained (adapted from Salmon, 1973:3): 3.4 People with large brains are highly intellectual. ‘The owner of this hat has a large brain. Therefore, the owner of this hat is highly intellectual. The premises of this argument are alleged statements of fact that supply evi- dence for the conclusion. Therefore, if either of the premises is false, then the evi- dence does not justify the conclusion. But even if the premises are factually true, they still would not justify the conclusion unless the argument itself is correct. In point of fact, the argument is logically valid, and therefore Holmes’s reasoning is sound. However, scientific research has shown the first premise to be false: the size of human brains is not directly related to intelligence. And the truth of the second premise rests upon the assumptions that hat size indicates skull size and that skull size indicates brain size. Hence, Holmes has not justified his conclusion. This does not mean that the conclusion is false. It may well be true, but Holmes has not pro- vided adequate justification for accepting it as true. Deduction and Induction The two major types of reasoning are deduction and induction. The primary dis- tinction between deduction and induction rests with the strength or certainty of the claim that is made about the conclusion on the basis of the premises. When a per- son uses deductive reasoning, or presents a deductive argument, he or she is claim- ing that the conclusion absolutely must be true if all the premises are true. (All the syllogisms that we have considered heretofore are deductive arguments.) When a person argues inductively, he or she is claiming that the conclusion is probably true but not necessarily true if all the premises are true. Consider the following examples of each argument type: 3.5 Deductive: If Joan belongs to the Union, then she votes Democratic. Joan belongs to the Union. Therefore, Joan votes Democratic. 3.6 Inductive: Hubert, Walter, and Joan, who are all union members, vote Democratic. Therefore, all union members vote Democratic. — — — The Logic of Scientific Reasoning 45, in the first argument it would be inconsistent to deny the conclusion if the premises are true because the conclusion says nothing that was not already stated by the premises. The conclusion merely makes explicit the information contained in the premises (Salmon, 1973:14-15). Jn the second argument, on the other hand, it is possible for the conclusion to be false when the premise is true. The premise does not provide certain support for the conclusion because the conclusion goes be- yond the information contained in the premise. Thus, a deductive argument, judged as to its validity, is either valid or invalid—there is no in-between. An inductive ar- gument, by contrast, is judged as more or jess sound depending on how probable it is that the conclusion is true, given the evidence provided in the premises. It is sometimes said that deduction moves from general principles to particular instances, while induction moves from the particular to the general. For example, the above deductive argument might have begun with the general proposition that “all union members vote Democratic,” from which we deduce that a particular union member votes Democratic. Ja our inductive argument, on the other hand, we observe that several union members vote Democratic and infer from this the gen- eral conclusion that they all do.? This way of distinguishing between deduction and induction highlights the role of these logical processes in science. Recall from chapter 2 that science is a cycli- cal process. At one point in the cycle, the scientist reasons inductively, starting with particular observations and deriving general propositions and theories. Then the sci- entist reasons deductively, deducing certain observable consequences from his or her theory. Both of these applications are discussed as we consider different forms of deductive and inductive arguments in science. Deductive Reasoning in Science In chapter 2 we showed how the propositions of scientific explanations can be or- ganized into deductive arguments. It is obviously important that such explanations represent valid deductive reasoning, for it is pointless to investigate a conclusion deduced from a theory if the deduction is invalid. To evaluate an act of deductive reasoning we must determine the validity or invalidity of the argument that repre- sents the reasoning. So far we have done this informally, relying on the reader's common sense or intuition to recognize the validity or invalidity of an argument. However, such a procedure is notoriously fallible. In this section we first introduce a more rigorous technique for testing the validity of arguments containing condi- tional propositions. Then we show how one can apply this technique to analyze the logical validity of scientific explanations. Argument Forms An argument is deductively valid if the truth of its premises necessarily implies the truth of its conclusion. We can recognize an invalid argument, then, as one in which — ———— 46 SCIENTIFIC AND LOGICAL FOUNDATIONS OF SOCIAL RESEARCH the premises are true, but the conclusion is false. By this definition the following syllogism is clearly invalid: 3.7 If a person is President, then he or she is famous. (TRUE) Tiger Woods is not President. (TRUE) Therefore, Tiger Woods is not famous. (FALSE) From the content of this syllogism we know that the premises are true and the conclusion is false; hence, it is invalid. However, since questions of truth and validity are independent of one another, we should be able to assess valid- ity without reference to the content of the premises and conclusion. Recall that validity refers to the relation between premises and conclusion; this is some- thing that we can check merely by examining the logical structure or form of the argument. The form of an argument is identified by substituting letters for each of the sen- tences or phrases it contains. Thus, in argument 3.7 above, if we let the letter p stand for the proposition “a person is President” and q stand for the proposition “he or she is famous,” then the form of this argument may be represented as follows: 3.8 If p, then q- Not p. Therefore, not q. We can read 3.8 as “If p is true, then q is true; p is not true; therefore, q is not true.” This particular argument form is invalid because we have shown that it is possible to have an argument of this form with true premises and a false conclusion. It fol- lows, then, that any argument having this form is invalid. In general, it is possible to determine the validity of a given argument by checking the validity of its form. Any argument with a valid form is a valid argu- ment; any argument with an invalid argument form is an invalid argument. This method has its limitations, which we will not go into here (see Copi, 1973:18-23; Salmon, 1973:20-21). More sophisticated methods for testing validity such as Venn diagrams, syllogistic rules, and truth tables are introduced in textbooks on symbolic or formal logic. For our purposes, it will suffice to learn some common valid and invalid argument forms. Once you are able to recognize these forms you can un- derstand and evaluate the soundness of much scientific reasoning. Letting the letters p, g, and r stand for any statement whatever, the three sim- plest valid argument forms containing “if-then” statements (i.e., conditional propo- sitions) are as follows: 1. Affirming the antecedent If p, then g. If this metal is gold, then it will not dissolve in nitric acid. P. This metal is gold. Therefore, q. Therefore, it will not dissolve in nitric acid. The Logic of Scientific Reasoning 47 2. Denying the consequent If p, then q. If he can do long division, then he can add and subtract. Not g. He cannot add and subtract. Therefore, not p. ‘Therefore, he cannot do long division. 3. Chain argument (or “hypothetical syllogism”) If p, then q. If money gets tight, then interest rates will rise. Tf q, then r. If interest rates rise, then the volume of loans will be low. Therefore, if p, then r. If money gets tight, then the volume of loans will be low. Note how the name of each of these forms is derived. In the form called “af- firming the antecedent,” the first premise is a conditional proposition, and the sec- ond premise affirms the antecedent of this conditional; in the form called “denying the consequent,” the first premise is a conditional proposition, and the second premise denies or negates the consequent of this conditional (Salmon, 1973:25). Fi- nally, in the chain argument form, the two premises are linked, as in a chain, by a common statement (symbolized by q) that is the consequent of the first premise and the antecedent of the second premise. It is possible to construct longer chain arguments with any number of premises. The only requirements for a valid chain argument are (1) that the consequent of each premise be the antecedent of the next premise, and (2) that the conclusion have the antecedent of the first premise as its antecedent and the consequent of the last premise as its consequent. Invalid argument forms are called fallacies. There are two common fallacies that are deceptively similar to the first two valid argument forms above: 1. Fallacy of affirming the consequent If p, then q. Tf he can do long division, then he can add and subtract q- He can add and subtract. Therefore, p. Therefore, he can do long division. 2. Fallacy of denying the antecedent If p, then g. If you study hard, then you will pass the test. Not p. You did not study hard. Therefore, not g. Therefore, you will not pass the test. Each of these arguments and argument forms is invalid because it is possible for the premises to be true and the conclusion false. You can see this in the first ex- ample if you realize that long division can be performed (without calculators, of course) only if one can add and subtract, but that those who can add and subtract (e.g., third graders) cannot necessarily do long division. It is also easy to imagine the same situation—that is, true premises and a false conclusion—in example 2, es- pecially if you have had the experience of passing tests without studying much. (The reader should note that syllogism 3.7 above is another example of denying the antecedent.) 48 SCIENTIFIC AND LOGICAL FOUNDATIONS OF SOCIAL RESEARCH Fallacies are identified by one particular combination of truth and falsity among the premises and conclusion of an argument. That is, an argument is invalid or fal- lacious if it is possible for the premises to be true and the conclusion false. It is im- portant to note, however, that a valid argument can contain any other combination of true and false premises and conclusion. For example, a valid argument with a true conclusion may have one or more false premises, as in the following: 3.9 If Harvard University is in California, then it is in New England. (FALSE) Harvard University is in California. (FALSE) Therefore, Harvard University is in New England. (TRUE) 3.10 If this animal is a whale, then it is a fish, (FALSE) If it is a fish, then it lives in water. (TRUE) Therefore, if this animal is a whale, then it Jives in water. (TRUE) The above arguments show that it would be a logical error either to infer true premises from a true conclusion or to infer a false conclusion from one or more false premises. A useful summary of the various combinations of true and false premises, valid and invalid arguments, and true and false conclusions is contained in the following three statements (Manheim, 1977:35): 1. If all the premises are true and the argument is valid, the conclusion must be true. 2. If all the premises are true and the conclusion is false, the argument must be in- valid. 3. if the argument is valid and the conclusion is false, at least one premise must be false. These statements should help you to evaluate deductive reasoning. However, lest one be fooled by merely examining the truth or falsity of propositions, the best practice is to assess the validity of an argument strictly by its form. To identify the form of an argument, as we have shown, one must simply substitute letters for each of the simple statements in an argument. If an argument’s apparent form is valid, the argument is valid; if the form is invalid, the argument is invalid. The Deductive Pattern of Scientific Explanation Now that we have the tools with which to evaluate deductive reasoning, let us ap- ply them to scientific explanation. Recall from chapter 2 that a particular event or phenomenon is explained in science by showing how it can be subsumed under a general empirical proposition. Suppose, for example, that we want to explain why your friend John, who runs 2 miles daily, runs faster when people are watching him run than when he runs in isolation. We might say that because John runs daily, run- ning is apparently not a terribly strenuous or difficult behavior for him, and people perform simple, well-learned tasks better in the presence of others than in isolation. Thus, we explain the phenomenon by showing that it is an instance of a general em- The Logic of Scientific Reasoning 49 pirical (or causal) rule, which we will call a hypothesis (but which might also be a law, depending on its degree of empirical support). Closer examination will show that the inference from hypothesis to a specific event can be put into the form of a deductive argument. The major premise of the argument consists of the hypothesis. This can be stated in the form of a conditional proposition that asserts a causal link between the antecedent and the consequent. Stated in general terms, the hypothesis does not logically imply any particular event. A second premise (or set of premises) therefore specifies the antecedent con- ditions that must be met to logically deduce the event to be explained. The conclu- sion is then the event to be explained. For example, the above explanation can be represented as follows: General law or hypothesis (premise 1): If a person performs a simple or well- learned task in the presence of others, then he or she will do better than when performing alone. Antecedent conditions (premise 2): John ran 2 miles, a well-learned task, as others looked on. (When John runs alone 2 miles his time is 12 minutes.) Event explained (conclusion): John ran faster (i.e., in less than 12 minutes) in the presence of others than he did alone. ‘The reader should recognize that this argument has a valid deductive argument form—affirming the antecedent. Scientific explanations of this kind thus follow a de- ductive pattem. Just as the conclusion is an expected consequence of the premises in log- ical deduction, the given phenomenon or event to be explained (or predicted) logically follows from a general proposition or hypothesis under certain specified conditions. In chapter 2 we also pointed out that laws are explained deductively by theories. However, the logical form of a theoretical explanation is not quite the same as in the explanation of events. The simplest theory has the form of a chain argument, with premises consisting of lawlike propositions from which a given law can be deduced (or explained). For example, Nikolas Cottrell’s evaluation apprehension theory of the social facilitation effect (1972) may be organized into the following set of propositions: Proposition (premise 1): If a person performs a well-leamed task in the presence of others, then the person will think that others are evaluating his or her performance. Proposition (premise 2): If a person thinks that others are evaluating his or her performance, then the person's response drive will increase. Theory Proposition (premise 3): If response drive increases, then dominant or well-learned responses are strengthened. Proposition (premise 4): If a person’s dominant responses are strength- ened, then the person’s performance on well-leamed tasks will be en- hanced. Law or hypothesis explained (conclusion): Therefore, if a person per forms a well-learned task in the presence of others, then the persons’ per- formance will be enhanced (i.e., will be better than the person’s perfor- mance in isolation from others). 50 SCIENTIFIC AND LOGICAL FOUNDATIONS OF SOCIAL RESEARCH This argument represents an interpretation of Cottrell’s theory. As far as we know, the theory has not appeared elsewhere in this form. Indeed, most theories in the social sciences are presented informally in ordinary sentences and paragraphs, albeit couched in technical language. To analyze a theory logically, it is necessary first to break it down into a series of propositions and then to organize the propo- sitions into an argument. The form of the argument then may be examined as to its deductive validity. Expressing a theory in standard logical form facilitates logical analysis by en- abling one to examine the form of the theory and the relations among theoretical propositions apart from their substantive meaning.* This makes it easier to detect implicit propositions and assumptions as well as possible contradictions that are of- ten hidden in ordinary language. It also clarifies the distinction between proposi- tions that are directly testable and those that are not. For example, because the term “response drive” in the above theory refers to a hypothetical entity that has no im- mediate empirical referent, any proposition containing this term cannot be directly verified or tested. However, such propositions are logically connected within the theory to testable propositions, or hypotheses, containing terms with observable properties. It is this logical connection that makes propositions without immediate empirical referents, as well as the theory as a whole, testable. Since scientific theories ideally explain events by showing how the phenome- non to be explained logically follows from a set of propositions, it is easy to see the relevance of deductive logic to scientific inquiry. The reasoning from theories to hypotheses should be deductively valid, for if the argument by which a testable conclusion is deduced from a theory is invalid, then it is pointless to investigate the truth of the conclusion. Inductive Reasoning in Science Deductive reasoning is either valid or invalid; either the truth of the conclusion nec- essarily follows when the premises are true, as in a valid argument, or it does not, as in an invalid argument. To meet the criterion of validity, the conclusion of a de- ductive argument cannot go beyond the content of the premises. Deduction there- fore “only tells us things we knew already,” but that we may not have realized that we knew prior to the deduction (Kemeny, 1959:113). Induction, on the other hand, involves the drawing of conclusions that exceed the information contained in the premises. Because science seeks to establish gen- eral knowledge that goes beyond particular observations, it inevitably uses inductive reasoning. Indeed, the truth of a generalization, which is assumed in assessing the deductive validity of explanations, ultimately rests upon induction. That is, at some point we say that a generalization is justified because it is based on one or more past observations. Our justification of the social facilitation effect, for example, is our ob- servation that in past situations people have performed better in the presence of oth- ers than alone. The fact that this effect has been observed frequently makes it rea- sonable to assume that it will happen again under similar conditions; however, we can never be absolutely certain of its recurrence. That is the nature of induction. The Logic of Scientific Reasoning st As with deductive reasoning, in inductive reasoning we are interested in the re- lations between premises and conclusion. However, analyzing inductive arguments is more complicated than analyzing deductive arguments because we must take into account the degree to which the premises support the conclusion. When the premises of a valid deductive argument are true, then the conclusion is always true; however, when the premises of a correct inductive argument are true, the conclusion is only probably true. Correct inductive arguments vary in the strength, or degree of probability, with which the premises provide evidence for accepting the conclusion. Further, the de- gree of support for an inductive conclusion can be increased or decreased by adding premises to the argument. For example, by inductive reasoning an actuary would claim it highly probable that a healthy, 20-year-old man will survive to age 65. This probability would increase somewhat if it is learned that the man exercises regu- larly and is not overweight. But if it is learned that the man smokes two packs of cigarettes per day or that he loves to practice hang-gliding, then the chances of his survival to age 65 are reduced considerably. We now examine two types of inductive reasoning common in the social sci- ences: inductive generalization and hypothesis testing. As each type of induction is considered, we look at the kinds of additional evidence that increase or decrease the strength of the argument. Inductive Generalization An inductive generalization says something about an entire class of objects or events on the basis of information about only part of the class. Suppose that each of the five union members you know is a registered Democrat. You may argue in- ductively from this evidence that all union members are Democrats. This is the sim- plest type of inductive generalization. It has the following form: 3.11 All observed members of p are q. ‘Therefore, all p are q. A variation of this argument asserts that if a certain percentage of observed p are q, then the same percentage of the entire class of p are q. For example, you might observe that four of five (80%) of the union members whom you know are registered Democrats and from this evidence infer that 80 percent of all union mem- bers are Democrats. The only difference between this argument and the above is that “80 percent” has been substituted for “all” in both premise and conclusion. Since “all” means “100 percent,” the general form of inductive generalization may be written as follows (Salmon, 1973:83). 3.12. X percent of the observed members of p are q. Therefore, X percent of p are q. Like all inductive inferences, inductive generalizations go beyond the observed facts. Hence, the best that can be said for them is that they are probably true. But 52 SCIENTIFIC AND LOGICAL FOUNDATIONS OF SOCIAL RESEARCH some inductive conclusions are more likely to be true than others; that is, some ar- guments are stronger than others. The strength of an inductive argument depends on how reasonable it is to suppose that the observed instances of a class (e.g., union members) are representative of the entire class. Several factors affect the reason- ableness of this supposition (cf. Barker, 1974). 1. Similarity of observations. The more alike the observed instances of p are in ways other than q, the weaker the argument. If, for example, all the observed union members were over age 40, lived in Detroit, and worked for an automobile manufacturer, then our conclusions about their Democratic political preference would be based on a relatively weak argument. The shared characteristics reduce our confidence that our observations are representative of the whole class of union members. It is possible that only those members over age 40, or those who reside in Detroit, and so forth, are Democrats. 2. Dissimilarity of observations. The more ways that the observed instances of p differ from one another, the stronger the argument. Thus, if we observe union members of all ages, from several cities, who are employed in a number of indus- tries, then our argument is strengthened. We have increased the probability that our observations are representative of all union members, for we have excluded the pos- sibilities that it is only members of a certain age, or residence, or occupation, who are Democrats. 3. Scope and precision of generalization. “The more sweeping the generaliza- tion that we seek to establish, the less is its probability relative to our evidence, and the weaker is our argument” (Barker, 1974:228). We can alter how much an in- ductive generalization says in two ways. First, we can alter the specificity of p, the class that is described. A conclusion about “all union members” is more sweeping than a conclusion about “all union members over 40 years of age.” Because it is more limited and specific, we would be more confident in drawing conclusions about the latter generalization than the former one. Similarly, we could change the likelihood of a conclusion by changing the precision of q, our description. Thus the conclusion that “all are Democrats” or that “80 percent are Democrats” is not only more precise, but is also less probable, than the conclusion that “most are Democ- rats.” 4, Number of observations. Ordinarily, the greater the number of observed in- stances, the stronger the argument. It is reasonable to assume that a conclusion based on a thousand observations is more probable than one based on just five. Gen- erally this will be the case, since larger numbers of observations are likely to have fewer characteristics in common and therefore to be more representative of an en- tire class. However, if the additional observed instances are all alike (as in 1 above), then the probability of the conclusion will not change. 5. Known relevance. The greater the relevance of the generalization to prior knowledge, the stronger the argument. The question that must be asked here is whether the connection between one property and another makes sense in light of other generalizations. For example, the expectation that members of labor unions are Democrats is consistent with the sociologist’s knowledge that political ideology is often related to economic position in society. In light of such knowledge, the in- — - TO The Logic of Scientific Reasoning 53 duction is more reasonable. If, however, the induction were incompatible with well- established knowledge, then it would be less reasonable. In evaluating the strength of inductive generalizations, all five of these factors must be taken into account simultaneously. Thus, one must consider the number of observations (4) in conjunction with their degree of similarity (1) or dissimilarity (2) as well as the scope and precision (3) of the inductive generalization and its re- lation to prior knowledge (5). Larger numbers of observations produce stronger in- ductive arguments if the observations are dissimilar or if the generalization is lim- ited in scope and precision. And generalizations consistent with established knowledge are more probable than those that are not consistent. Reasoning by inductive generalization is common in science. The idea of strengthening conclusions with dissimilar observations directly parallels the idea of replicating studies with dissimilar procedures, which we discuss in chapter 13. Con- siderations about the representativeness of observations are central to the sampling procedures discussed in chapter 6. Moreover, this form of induction underlies the logic of many statistical techniques, such as those designed to estimate characteris- tics of whole populations from a sample of observations. Perhaps even more commonly, scientists and nonscientists alike apply the logic of inductive generalization in their attempts to perceive order in the world around them. For example, a sociologist finds that 60 of 100 students observed in a given school have smoked marijuana and concludes that the majority of students in the school have smoked marijuana. Upon entering college you notice that all your pro- fessors have doctorates and infer that all professors do. You find that your first course in sociology is interesting, have similar experiences in the second and third courses, and conclude that all sociology courses are interesting. Of course, none of these conclusions may actually be true. As with all inductive generalizations, they merely possess greater or lesser probability, depending on the strength of the evi- dence from which they are derived. Comparing the above conclusions about mari- juana smokers, professors, and sociology courses, which inference do you think is strongest? Be sure to consider each of the five factors. Testing Hypotheses: The Hypothetico-Deductive Method The underlying logic of testing hypotheses in science is sometimes called the hypo- thetico-deductive method. The first part of this term signifies that the scientific ex- planations tested are hypothetical in the sense that they are assumed to be true. Once this assumption is made about an explanation, then its observable consequences are deduced for testing—hence, the second part of the term. Forming a hypothesis and deducing consequences from the hypothesis are actually the first two of four steps in the hypothetico-deductive (h-d) method. Step 3 involves checking by observation to see whether the deduced consequences are true, and step 4 entails making infer- ences about the hypothesis on the basis of one’s observations. We shall forego a dis- cussion of step 3 here because it is considered in the remaining chapters in the book. For now we confine our attention to the logical characteristics of steps 1, 2, and 4. 54 SCIENTIFIC AND LOGICAL FOUNDATIONS OF SOCIAL RESEARCH In the section on deduction we saw how scientific explanations explain facts deductively. However, while focusing on the logical form of explanations, we ig- nored the inductive process by which the explanations are derived in the first place. All scientific explanations—theories, generalizations, hypotheses—are inductive conclusions. To impose an order on observations and make them meaningful, an explanation must go beyond the information contained in the observations that it explains. Consider, for example, Emile Durkheim’s generalization that high social solidarity is related to low rates of suicide. This proposition was derived inductively from several observations, such as that divorced and widowed persons committed suicide more than married persons, childless persons more than people who had children, and city dwellers more than rural people. To explain such facts, Durkheim had to formulate an inductive conclusion—a hypothesis that went beyond the facts available.* Once formulated, a hypothesis explains the facts from which it is de- rived in that when it is assumed to be true, the facts logically follow. More pre- cisely, like all explanations, the hypothesis serves as a premise in deductive argu- ments in which the facts are conclusions. By deduction, Durkheim’s hypothesis explains the fact that the incidence of suicide is lower among married people than among divorced or widowed persons: 3.13 If the social solidarity of one group is higher than that of another, then its sui- cide rate will be lower. (HYPOTHESIS) Social solidarity is higher among married people than among widowed or di- vorced persons. Therefore, the suicide rate is lower among married people than among wid- owed or divorced persons. (OBSERVED FACT) Explaining a set of facts in this fashion indicates that we have a legitimate hy- pothesis, but it does not mean that the hypothesis is true, for some other hypothe- sis might explain the facts just as well. Considering Durkheim's data, for example, perhaps widowed, divorced, and childless persons as well as city dwellers tend to share the kind of social or religious philosophy that finds suicide morally accept- able. Or, it is possible that these same groups have higher rates of mental illness, which may be the cause of suicide. Because such alternative explanations exist, Durkheim’s hypothesis should be thought of as more or less probable in light of the facts that it explains. Like all inductive conclusions, it is drawn and accepted “in a spirit of tentativeness” (Runkle, 1978:261). Because any set of facts can have several explanations, we gain confidence in a particular explanation by testing its ability to explain additional observable facts. Basically, this means that we deduce other consequences from the hypothesis and then check to see if they are true. Thus, while formulated to explain past or exist- ing events, hypotheses must be tested in terms of their capacity to predict future events. The second step in the h-d method therefore involves deducing conse- quences from the hypothesis. Let us now see what conclusions can be drawn when the hypothesis is confirmed and when it is disconfirmed. The logic of confirming hypotheses. One of the consequences that Durkheim derived from his hypothesis was that Catholics in France will have lower suicide rates than Protestants in France. This follows from his hypothesis and from the as-

You might also like