Thank you to the readers who helped me with this paper. Any mistakes are my own.
Some terminology in behavior science is notoriously hard to get one’s head around. One of these terms is negative reinforcement. Not only is this learning process itself a challenge to understand, but the terminology itself is counterintuitive. Behavior scientists specialize in training, teaching, and learning, so naturally, if a term from their own field trips people up, they are going to analyze the problem. The terminology for negative reinforcement has already been changed once, in the 1950s to early 1960s. There has been more discussion since then. This post is about the article that started the more recent discussion, and how it is often misunderstood in the animal training community.
In 1975, psychologist Dr. Jack Michael published an article named, “Positive and negative reinforcement, a distinction that is no longer necessary; or a better way to talk about bad things” in the journal Behaviorism.
This journal article is widely mischaracterized, in my opinion. It is commonly quoted by people who use aversives in training and seek to minimize that when discussing or defending their methods. And certainly, the title sounds very promising for just that purpose. But only if you ignore the last phrase about “bad things.”
Some people claim the article says that the distinction between the learning processes of negative reinforcement and positive reinforcement doesn’t exist or is immaterial. They say that the difference between positive and negative reinforcement is blurred and can’t always be determined. Some say that Dr. Michael dismisses all the possible reasons for maintaining a distinction between the two. This is false (see page 43 in the paper).
Michael’s paper centers on better ways to make descriptions of and determinations about the contingent processes of operant learning. The claim that Michael states that there is little difference between positive and negative reinforcement is false. This claim misrepresents both the focus and the conclusions of the article. Note again the last part of the title: “A better way to talk about bad things.”
In the article, Michael asks whether we need to make the distinction between what we call positive and negative reinforcement. His final answer is yes, that we need the distinction. He concludes, “We need to make the distinction in order to have a name for the bad things in our world.” (page 43)
Dr. Michael is concerned about terminology on two fronts:
- He wants to get rid of positive/negative and present/remove in the descriptions for different types of reinforcement.
- He wants to find a better nomenclature to indicate when an aversive is involved.
He proposes a solution, which I will describe below.
There are four major parts to the paper: a history of the usage of the terms for reinforcement and punishment, a critique of the current terminology, a section that explores whether we need the distinction or not (his answer: yes), and a proposed solution. I’ll summarize each briefly. The following four sections are headed with the subtitles used in the paper.
1. A Brief History of the Distinction Between Positive and Negative Reinforcement
This section comprises 75% of the paper and is devoted to a retrospective of the usage of the terms for reinforcement and punishment, starting with Skinner in 1938. As some people know, what Skinner initially called “negative reinforcement” is what we now call punishment.
A textbook published in 1950 by Keller and Schoenfeld (1950) used different terminology, and in 1953 Skinner reversed his usage of the terms in his own textbook, defining them as we know them today. There was a period of transition—Michael mentions that it had to have been especially tough for the students who attended courses at the same time that employed different textbooks—and by the 1960s Skinner’s revised usage, what we use today, was in common use.
I am not going into detail here, but Dr. Michael did. Eight and a half of the eleven pages of the article are dedicated to the changes in definitions and usages of the terms and the resultant confusion. This is a major focus of the article and a major part of his criticism of the use of “positive” and “negative” with regard to reinforcement.
2. What is Wrong with the Present Usage?
In this section, Michael says, “Since 1953 there must have been thousands of man-hours spent in the attempt to prevent the learner of behavioral terminology from equating [negative reinforcement] with punishment…”
Even though I am not credentialed in that field, I know what he means. I have spent many hours myself figuring out the processes of operant learning, with bonus time on negative reinforcement, and many hours as well trying to pass on my basic understanding to others. Reinforcement, punishment, and the plusses and minuses can be confusing, especially since almost all the words used have other meanings or common metaphorical uses.
Michael goes on to describe another problem that includes semantics, in this often-quoted section:
Another difficulty with current usages is that the critical distinction between positive and negative reinforcement depends upon being able to distinguish stimulus changes which are presentations from those which are removals or withdrawals, and these latter terms are not very satisfactory descriptions of changes. The circumstances under which we have a tendency to say “present” certainly seem to differ from those where we say “remove” in vernacular usage, but some of these differences are irrelevant to a science of behavior, and there are a number of circumstances where the distinction is not easily made.Michael, 1975, p. 40
Note that he is not saying that it’s difficult to detect the differences between aversive and appetitive stimuli. The issue he objects to is the use of the terminology of presenting and removing stimuli.
…In other words, from the point of view of the behaving organism presentations and removals are both simply types of environmental changes. If they differ, the difference must not be based upon the variables controlling the person who causes the change.Michael, 1975, p. 40
This section merits careful reading. His major objections to the concepts of “presenting” and “removing” are that they focus unnecessarily on the actions of the environment or a third party and that they have societal and linguistic baggage (e.g., he mentions that removal can sound negative). He says what is really important to the subject organism is simply that something changed, and it is the point of view of the subject that we should be concerned about describing. We don’t need to talk about adding or removing stimuli, we need to describe bad changes and good changes from the standpoint of the subject. (Michael uses the terminology of “bad” and “good” throughout the article, which is also a deviation from standard practice.)
In other words, it appears that “present” and “remove” are abbreviations that can sometimes stand in place of a more complete description of both the pre-change and post-change condition. The abbreviation is usually possible in the case of unconditioned reinforcements, although even here it must always be possible to infer the characteristics of both pre- and post-change conditions if we are to imply behavioral significance.Michael, 1975, p. 41
An interesting point: he states “present” and “remove” are incomplete descriptions. He is not arguing to ignore the nature of the circumstances the organism finds itself in. He is arguing against shorthand. He is arguing that we need to describe the state of the environment and the nature of the change more accurately in order to determine the learning process in play.
There is much more in this section about presentations and removals not being specific enough for scientific usage, and it is in this section one really gets a sense of Dr. Michael’s concerns.
He also addresses an argument that has been going on for a long time in behavior science. It goes like this:
You can’t tell the difference between positive and negative reinforcement if you train using food because you don’t know if you are adding food or removing hunger.Various people
Experts in the field discuss this question earnestly and with goodwill. But you will also see it glibly thrown into arguments by trainers who seek to mask their use of aversives (it’s is a favorite among force trainers). I humbly offer my own study of this question, but here’s a surprise. Dr. Michael addresses this very situation in his paper.
When we say that we present a food pellet to the rat the listener can always assume that the pre-change condition is one in which no food is available. We could say that we remove the “no-food” condition, but then the behaviorally important aspect of the change would remain to be described. When we say that we terminate a 50 volt electric shock, the subsequent “no-shock” condition can generally go without further description, but if it were described alone little information would be provided.Michael, 1975, p. 41
He is saying that only one description is typically accurate in a particular situation because the other one fails to describe crucial parts of the situation. Again, he is arguing that we need to analyze the reinforcement situation with information about the environment before and after the change, not by focusing on one stimulus and whether someone “presented” or “removed” it. What is happening from the animal’s point of view? Is it a “good” change or a “bad” change, and does it involve a bad thing (aversive)?
3. Why Do We Bother?
In this section of the paper, Michael examines possible differences between negative and positive reinforcement and discusses whether each particular aspect could or should be the reason we need to make a distinction.
As we find ourselves applying behavioral analysis to more and more complex human situations we find it increasingly difficult to distinguish between presenting and removing, or we find an increasing number of situations that seem to involve both. A fairly common response to this situation is to avoid making the distinction, and simply refer to the relevant environmental change as “reinforcement,” without attempting to determine whether a positive reinforcer is being presented or a negative removed. One might well ask, then, why we bother making the distinction even in those cases where it can easily be made.Michael, 1975, p. 41
If your goal is to generally minimize fallout of the use of negative reinforcement, you can cherrypick the above paragraph without continuing and make it look like Michael is saying the distinction between positive and negative reinforcement is unnecessary. On the contrary, this is the section where he specifically rejects that interpretation. He considers four reasons for making the distinction. He discounts the first three as follows.
- Are the (behavioral) strengthening effects of R+ and R- different? He answers that they are not any more different than the differences between different forms of R+.
- Do R+ and R- involve different physiological structures or processes? He doesn’t think trying to make this distinction is a good idea in view of the changing field, but he leaves room for future research. This article was published in 1975, before most of the current discoveries that showed exactly that: that different physiological processes are likely involved (Overall, 2013, p. 69).
- Should we keep the current terminology so as to warn people only to use “positive,” not “negative”? He appears to be asking whether we should actually appeal to the double meaning of positive. Again he answers no, that we shouldn’t base a scientific definition on a social distinction. (Note that in using the term “social distinction” he is referring to the words “positive” and “negative,” not to the actual learning processes.)
So he rejects three reasons for making the distinction between positive and negative reinforcement. Then, in a section that is rarely quoted, he goes on to answer his original question in the affirmative, saying that we do need a way to distinguish the difference. He says:
The layman frequently finds it necessary to identify an environmental event or condition as one which he doesn’t like, which he attempts to escape, or avoid. He may refer to such an event as “bad” (without the moral implications of this term), “undesirable,” “unfavorable,” etc., and he also has “punishment” to use as a contrast with “reward.” A science of behavior also needs a way of identifying such events.Michael, 1975, p. 42
We need to make the distinction in order to have a name for the bad things in our world…Michael, 1975, p. 43
He is arguing that we need the distinction between what is currently called negative and positive reinforcement so as to be able to specify when a “bad thing” is involved. So it is incongruous that this paper is cited in support of arguments to blur and erase the use of aversives.
4. The Solution
Michael spends so much time focusing on confusing terminology in the paper that it is strange he doesn’t devote more space to making his solution clear.
But here is what he wrote.
So, the solution to our terminological problem is to refer to the good things as reinforcers and reinforcement and call the bad things punishers and punishment. One set of terms refers to changes which have a strengthening effect on the preceding behavior; the other to changes which have a weakening effect. The distinction between two types of reinforcement, based in turn upon the distinction between presentation and removal simply can be dropped.Michael, 1975, p. 44, bold added by Eileen
The last sentence of that quotation can also be taken out of context in a misleading way. A hasty reader, or one with an agenda, can claim Michael is saying that there is no difference between the learning processes we call R+ and R-. But he has already said that we need to specify when there is a bad thing involved. He is arguing not to base the distinction on the terminology of presentation and removal.
Finally he writes:
The arguments set forth above convinced me about 6 years ago to stop making the distinction between negative and positive reinforcement and to refer to the bad things as punishers and punishment.Michael, 1975, p. 44
That is the way he achieves his goals of getting rid of the terminology of presentations and removals and finding a better way to describe the “bad things.”
It’s a shame that Dr. Michael doesn’t give some examples of applying his terminology. But I would suggest a couple of examples, following his lead. Both of these are what we would now call negative reinforcement.
- In a shock experiment with the goal of increasing behavior, the learning process could be called using reinforcement with the punisher of shock.
- In an escape protocol where an animal’s behavior is reinforced by giving them more distance from a scary thing, the learning process involved could be called using reinforcement with the punisher of a feared stimulus.
It seems clunky at first, but once you realize a bad thing (“punisher”) can be involved in reinforcement in only one way, escape/avoidance, it falls into place.
Dr. Michael makes it clear that we need to stipulate when there is a bad thing included as part of the learning process. He also states that what we call negative reinforcement includes a bad thing, and presents a cogent argument that the differences between what we currently call R+ and R- are important and are distinct from each other in real-life situations.
In 2013 (yes, I’ve been working on this post for seven years), I tried to contact Dr. Michael to ask for some examples of how he applied his terminology: how and when he made the distinction that a punisher was involved. I reached his wife, who said he was not able to discuss such things any longer due to dementia. He passed away this year: November 13, 2020.
Dr. Michael’s paper prompted several others in the same vein, questioning the terminology of “positive” and “negative” with regard to reinforcement. In my reading, the arguments had some of the same flavor but were not exactly the same. I’ve included those articles in the references below. My arguments above apply to Michael’s 1975 article alone.
These papers usually get a footnote in behavior science textbooks, but the standard nomenclature hasn’t changed to reflect the ideas put forth, which Michael himself later noted (Michael, 2005). I recently heard a behavior analyst being interviewed in a podcast voice a similar concern with “presentations and removals.” She mentioned that in her work it is most important to observe whether behavior is under aversive or appetitive control, and those are the classifications she uses.
And permit me one moment of editorializing: I don’t know any trainers who don’t use negative reinforcement. Even the kindly act of letting an animal leave or take a break from a difficult procedure means that R- is a planned part of a training plan. Most of us would agree that allowing escape is less intrusive than flooding, but we also try mightily to train with enough skill that the animal doesn’t want to leave in the first place. So my aim here is not to preach purity, although I try to avoid the use of R- in every possible way. My argument is with people who are disingenuous about their use, and who cherrypick quotes from this paper to attempt to obfuscate the contingent processes of operant learning.
Rest in peace, Dr. Michael, and I hope my efforts here have done this famous paper justice.
Baron, A., & Galizio, M. (2005). Positive and negative reinforcement: Should the distinction be preserved?. The Behavior Analyst, 28(2), 85-98.
Baron, A., & Galizio, M. (2006). The distinction between positive and negative reinforcement: Use with care. The Behavior Analyst, 29(1), 141-151.
Chase, P. N. (2006). Teaching the distinction between positive and negative reinforcement. The Behavior Analyst, 29(1), 113.
Iwata, B. A. (2006). On the distinction between positive and negative reinforcement. The behavior analyst, 29(1), 121.
Keller, F. S., & Schoenfeld, W. N. (1950). Principles of psychology: A systematic text in the science of behavior.
Lattal, K. A., & Lattal, A. D. (2006). And Yet…: Further comments on distinguishing positive and negative reinforcement. The Behavior Analyst, 29(1), 129.
Michael, J. (2006). Comment on Baron and Galizio (2005). The Behavior Analyst, 29(1), 117.
Michael, J. (1975). Positive and negative reinforcement, a distinction that is no longer necessary; or a better way to talk about bad things. Behaviorism, 3(1), 33-44.
Nakajima, S. (2006). Speculation and Explicit Identification as Judgmental Standards for Positive or Negative Reinforcement: A Comment on. The Behavior Analyst, 29(2), 269.
Overall, K. (2013). Manual of Clinical Behavioral Medicine for Dogs and Cats. Elsevier Health Sciences.
Skinner, B. F. (1938). The behavior of organisms: an experimental analysis.
Copyright 2019 Eileen Anderson