The Distinction Between Positive and Negative Reinforcement: Some Additional Considerations

Because of the many constructive contributions by Alan Baron and Mark Galizio, I have learned to pay close attention to whatever they have to say. In the present instance (Baron & Galizio, 2005), however, I am puzzled—as I was when Jack Michael published his original paper suggesting that we abandon the distinction between positive and negative reinforcement (Michael, 1975). It is not clear to me whether they are recommending simply a new terminological convention or whether they are suggesting something more fundamental—a change in basic principles, or at least a new way to conceptualize our data.

I am sure that Baron and Galizio (2005) and Michael (1975) are not recommending that we ignore the data from experiments on (or even from daily observations of) escape and avoidance behavior, the phenomena that give rise to the concept of negative reinforcement. They are as committed as anyone to a behavioral science based on data, and would never suggest that we throw data away simply because some of those data are difficult to classify. I wondered, therefore, whether they were simply looking for another term that might encompass those data more precisely. I certainly agree with their observation that the concept of negative reinforcement has caused confusion. In my experience, this has been true especially of readers and audiences who are being introduced to behavioral science. For example, readers of my book on coercion (Sidman, 2000) who do not finish the book are most likely to have put it down permanently after they reached my presentation of the distinction between positive and negative reinforcement. (This remains true even after an extensive revision of that section.) I believe the difficulty here stems from conventions of ordinary speech, in which the term negative usually denotes the opposite of something positive; having accepted a definition of positive reinforcement, and, in the process, equating it at least roughly with the everyday term reward, many then find it sensible to equate negative reinforcement with punishment. This, of course, is not what we ask them to do, so we lose the ones who are not willing or who are unprepared to work out the problem.

And so, I would welcome a substitute for the term negative reinforcement, a new term that removes the confusion of negative reinforcement with punishment while retaining the notion that escape and avoidance are kinds of reinforcement. I have not been able to come up with a satisfactory term, but I would not object in principle to such a suggestion. Or perhaps we could drop the characterization of reinforcement as either positive or negative and just talk about different kinds of reinforcement: food, water, sex, escape, avoidance, and so on. Baron and Galizio (2005) however, seem to be unwilling to take the latter route, on the grounds of the uncertainty in specifying whether production or escape is the critical reinforcing event.

The tendency to equate negative reinforcement with punishment is not, however, a major source of Baron and Galizio's (2005) concern. Their main discontent seems to arise from a perceived difficulty in distinguishing between the production and the removal (or prevention) of stimuli as a basis for the classification of reinforcing events:

The argument … is that positive and negative reinforcement are changes from one stimulus condition to another, not the simple presentation or removal of a stimulus. Without this essential clarification, the claim that a reinforcer is exclusively positive or negative always can be challenged by the assertion that the alternative form is the true basis for the reinforcing effect. (p. 87)

One cannot, of course, claim that any event is exclusively a positive or negative reinforcer in the sense that it is always just one of these. Given an appropriate history and context, any event can function as either a positive or a negative reinforcer. I must assume, then, that Baron and Galizio are referring to an event that takes place at a particular time and in a particular context, and are questioning whether the reinforcement on that particular occasion is positive or negative. To illustrate this difficulty, they cite the classic Weiss and Laties (1961) experiment in which rats learned to press a lever that turned on a heat lamp and increased the temperature in their cold chamber. Is the reinforcement here the production of heat (positive) or the reduction of cold (negative)? Baron and Galizio then generalize this interpretive difficulty to all instances of reinforcement. They argue that production of a stimulus (positive reinforcement) always involves escape from a situation (negative reinforcement) in which the stimulus was absent; removal of a stimulus (negative reinforcement) always involves production of a situation (positive reinforcement) in which the stimulus is absent. I have two major difficulties with this argument.

First, even though Baron and Galizio raise cogent objections to notions that reinforcement involves changes in physiological or emotional states, they still find it necessary to buttress their principal argument by appeal to such states. For example, the presentation of food is also said to involve reduction in a state of deprivation (p. 87). Ingenious research designed to support exactly such an explanation for the effectiveness of food and water as reinforcers (e.g., Miller & Kessen, 1952; Miller, Sampliner, & Woodrow, 1957) failed to prove convincing, and the theory that reinforcement must involve drive reduction is no longer in vogue. Similarly, Baron and Galizio ask, “Is it better to speak of the consequence as increased attention or as relief from loneliness?” (p. 91). They ask this question in support of their thesis, in spite of their valid critique (p. 90) of the use of emotional processes as explanations.

Second, and in my view, more important, is the too easy acceptance by Baron and Galizio of the impossibility of determining whether presentation or removal is the reinforcer in any particular case. In their exemplar, the Weiss and Laties (1961) study, it is, of course, true that both antecedent and consequent temperatures must be taken into account in specifying the reinforcer. But does the existence of instances that cannot be classified solely as positive or negative reinforcement provide sufficient reason to do away with the distinction? Could it not still be true that often it is specifiable and that there is some value in doing so? Areas of definitional confusion exist at the edges of many, if not most, concepts, yet this does not necessarily make the concepts unusable. Is something large or is it small, fast or slow, high or low? In these and many other cases, one can only answer, “That depends.” It is doubtless true that the termination of a moneyless period may be critical in determining the reinforcing value of money, yet we see billionaires working mightily to get more. The effectiveness of some positive reinforcers depends less on their previous absence than on what behavior they make possible in the future. Such problems of definition should lead to research that will help to clarify the existing definitions. Even punishment, which Baron and Galizio rely on as a replacement for negative reinforcement, has its definitional problems, about which I shall have more to say below.

Very often, when an instance seems open to confusion about its status as a positive or negative reinforcer, ways exist to make the distinction clear, to determine whether a behavioral change that results from production or removal can be specified as something more than just the absence of an alternative. Responses that turn off electric shocks also take the organism into an environment in which shocks are absent, but would the animal work to produce (or to maintain) that same environment independently of any experience with shocks? That experiment can be done, but the results are so predictable that no one does so. The termination of shock will be reinforcing even if the environment produced by the escape behavior keeps changing. When a child who acts in such a way as to bring about his removal to a time-out room does so only when he is being required to engage in a particular activity, one can easily determine whether escape from that activity or whether the time-out room per se provides the reinforcers for his disruptive behavior. This experiment is often worth doing because production of the social interaction on the way to the time-out room may be found to be the reinforcer rather than escape per se. Would the production of an aspirin suffice to reinforce the opening of the bottle if the pill did not turn off the headache? I think we know what would happen to aspirin consumption if we filled aspirin bottles with falsely labeled placebo tablets. Might a child turn on a TV program even if it took her away from a favorite game or a much-wanted dessert? Such an event is certainly possible, and when it happens, it would be difficult to identify the reinforcement as escape from boredom. Do spouses usually seek divorces independently of their marital experiences? Surely, we do not need experimental data to classify many, if not most, divorces as escape behavior. The possibility and utility of identifying positive and negative reinforcers do not seem as discouraging to me as they apparently do to Baron and Galizio.

I suspect that one reason for Baron and Galizio's (2005) unease with the positive–negative reinforcement distinction is their feeling that it reflects differences in the reinforcement process itself. I am not sure where that feeling comes from. For anyone who does espouse such a view, their convincing discussion should settle the issue. I am, however, uneasy about their unquestioning substitution of the reinforcement–punishment differentiation for the positive–negative reinforcement distinction.

First of all, I have a problem with the definition of punishment as a consequence of behavior that reduces the future probability of that behavior. When a teacher's correction ensures that a student's wrong answer will never occur again, has the teacher punished that wrong answer? When a parent ends a child's eating between meals by hiding the cookie jar, has the child's cookie eating been punished? If a child stops misbehaving after being asked to stop, has the misbehavior been punished? When I ask for something and the answer is “no,” does my failure to make the request again signify that it has been punished? For such reasons, I do not believe that Baron and Galizio's definition of punishment is any more free of ambiguity than are most concepts.

The above definition of punishment was first proposed by Azrin and Holz (1966, p. 381), and it slipped into the mainstream of the operant culture without the extensive discussion one might have expected from such a radical change in B. F. Skinner's definition. There seem to be no more than a few today who even recognize that such a change took place. I do not, of course, quarrel with Azrin and Holz's definition simply because it differs from Skinner's, and perhaps the difference does not affect anything we might do in the laboratory or clinic, but it seems to me to possess a systematic status that would have been worth discussing before accepting it uncritically.

Skinner defined punishment as either the response-contingent presentation of a negative reinforcer or the removal of a positive reinforcer, a definition that was most clearly detailed in Holland and Skinner (1961, p. 245). There are two types of punishment simply because each type of reinforcement, positive and negative, has a symmetric counterpart. Just as Skinner claimed no differences in principle between positive and negative reinforcement, he claimed no such differences between the two types of punishment. As I have noted elsewhere, Skinner's definition of punishment “says nothing about the effect of a punisher on the action that produces it. It says neither that punishment is the opposite of reinforcement nor that punishment reduces the future likelihood of punished actions” (Sidman, 2000, p. 45). Unlike his definition of reinforcement, his definition of punishment appeals to no behavioral outcome. The effects of any punishment are to be empirically determined.

To specify a stimulus or an event as a reinforcer, one must observe a subsequent increase in the behavior on which it was contingent. Then, to explain the increase in behavior, one appeals to reinforcement. The circularity here is apparent, and although it can be obviated by a rather intricate chain of verbal reasoning, it remains an inelegant feature of reinforcement theory. Azrin and Holz's (1966) definition of punishment imparts the same inelegance to a second fundamental concept. Skinner's original definition avoids this awkwardness because it requires no particular behavioral outcome in order to identify a punisher. Perhaps further discussion of this matter would relieve my unease about appealing to the currently dominant definition of punishment to solve any perceived problems with the concept of negative reinforcement.

Finally, something more needs to be said about relating science to problems of everyday life. Baron and Galizio (p. 91) object to the notion that the positive–negative reinforcement distinction might help to inform applied behavior analysts about the undesirable aspects of negative reinforcement. In a sense, this is a “straw man” type of issue. First of all, undesirable here must refer to social rather than to scientific desirability. Whether negative reinforcement procedures are undesirable is not an empirical scientific question at all, because it is not a question that science can answer. What science can do, however, and has done, is to provide data that should be relevant to any such decisions—extensive data, for example, on escape and avoidance behavior, on punishment, on conditioned suppression, on extinction- and shock-induced aggression, and so on. Our science should present these data to our society not to tell it that those procedures produce undesirable effects but to tell it just what those procedures do. If a scientist wants to do more than just provide information and also to help make society's judgment call, it must be recognized that he or she is doing so not as a scientist but as a concerned citizen.

With respect to Michael's (1975) doubts about the wisdom of maintaining “a distinction at the level of basic science because of its possible social implications” (p. 42), those very doubts may, in turn, be questioned. If any science is to survive or is to be accepted by society as something worthwhile—that is to say, worth supporting—then it had better relate its content to problems of everyday life. Indeed, that is where sciences start (Lee, 1988). People see things happening and wonder about them, trying to understand and make use of them. Coercion, which I have defined as punishment and negative reinforcement (Sidman, 2000), is a topic that has held people's interest through the ages and up to now. I was interested in studying avoidance and escape because even without any training in behavioral science I saw those as the roots of what used to be called “abnormal” behavior. A science that fails to make a place for topics that are considered important in the world at large may not only fail to survive but may deserve that fate.

And so, I find myself in disagreement with colleagues for whom I have unbounded respect. If those disagreements could be resolved, I would consider myself in one of those situations in which positive and negative reinforcement (production of agreement and escape from disagreement) are inextricably mixed.