B. Concepts and Principles

Behavior includes the activities of an organism, such as breathing, turning a page or silently rehearsing a word. Some behavior is publicly observable; private behavior is accessible mainly to the person engaging in it. One particular tap on a screen is a response. To identify a functional response class, group responses by what they accomplish: pointing, speaking and handing over a cup may all obtain more water.

Stimuli include detectable events outside or inside the body, such as a light turning on, pressure on the skin or stimulation from internal organs. A stimulus class groups stimuli using a specified criterion: their physical characteristics, their timing relative to behavior or their behavioral function. Feature classes use physical resemblance. In arbitrary classes, learned relations establish membership rather than a required physical resemblance.

Respondent behavior is elicited by antecedent stimuli: an air puff to the eye can elicit a blink. In respondent conditioning, a stimulus initially neutral with respect to a particular response can acquire an eliciting function through pairing with an unconditioned stimulus or an established conditioned stimulus. Conditioning depends on the arrangement, and the conditioned response need not resemble the unconditioned response. Operant behavior changes through its consequences; antecedent stimuli can occasion responding because of that history, rather than elicit it as a reflex.

Reinforcement occurs when a consequence produced by a response makes that kind of response more likely in similar conditions in the future. Whether the consequence feels pleasant is beside the point; only the effect on future behavior counts. Positive means a stimulus was added after the behavior. Negative means a stimulus was removed, reduced or postponed. Ending something already present is escape; preventing or delaying something that has not arrived yet is avoidance.

Punishment occurs when a consequence produced by a response makes that kind of response less likely in similar conditions in the future. Positive punishment adds a stimulus. Negative punishment removes one, as in response cost (losing points, paying a fine) or time-out from reinforcement. As with reinforcement, the name depends on the effect: if the behavior does not decrease, the consequence was not working as punishment, however unpleasant it looked.

In a socially mediated contingency, another person delivers the consequence: attention, an item handed over, a task taken away. In an automatic contingency, the behavior produces its consequence by itself, with no one involved. Automatic reinforcement can be positive (stimulation the behavior produces) or negative (discomfort the behavior relieves), and it is operant behavior like any other.

An unconditioned reinforcer does not require learning to acquire its reinforcing function: food when hungry or warmth when cold can be examples. A conditioned reinforcer acquires its function through a history linking it with already effective reinforcers. A generalized conditioned reinforcer has a history involving several different reinforcers, making its effectiveness less dependent on any single motivating operation. Its effectiveness can still change.

The same logic applies to punishers. Unconditioned punishers, such as pain or extreme cold, need no learning. A conditioned punisher acquires its effect by being paired with a punisher that already works. A generalized conditioned punisher has been paired with many different punishers.

Operant extinction discontinues the reinforcement previously produced by a response. Respondent extinction presents a conditioned stimulus without the unconditioned stimulus. The operation is the change in these arrangements; the process is the resulting reduction in responding across exposure. Operant extinction can be accompanied by a temporary increase in responding or response variation, and responding can briefly return after a break. An extinction burst and spontaneous recovery are possible effects, not requirements for identifying extinction.

A schedule of reinforcement is the rule that says which responses are reinforced. Continuous reinforcement reinforces every response and is useful for building a new behavior. In intermittent schedules, ratio schedules count responses, while interval schedules reinforce the first response after some time has passed. Fixed means the requirement stays the same; variable means it changes around an average. Each tends to produce a typical pattern: on fixed ratio, a pause after each reinforcer followed by a fast run; on variable ratio, a high and steady rate; on fixed interval, little responding right after reinforcement and an acceleration toward the end of the interval (the "scallop"); on variable interval, a moderate and steady rate.

Compound schedules combine simple ones, and they are named by how the parts are arranged. In a multiple schedule, components alternate and each one is signaled by its own stimulus. A mixed schedule alternates in the same way with no signal. In a chained schedule, components must be completed in order: finishing one produces the signal for the next, and the reinforcer comes after the last. In a concurrent schedule, two or more schedules are available at the same time for different responses, and the person chooses between them.

Behavioral momentum theory distinguishes response rate from resistance to disruption, assessed relative to baseline responding. Resistance is related to reinforcement obtained from all sources in a stimulus context, including reinforcement not contingent on the target response. Richer contexts often produce greater resistance under comparable conditions. High-probability request sequences were inspired by this account: several requests reliably followed by the person, with reinforcement for responding, precede a lower-probability request. Their effectiveness does not by itself demonstrate a momentum mechanism.

Under concurrent schedules, especially concurrent variable-interval schedules with comparable reinforcers, relative responding often approximates relative obtained reinforcement. In ideal proportional matching, an option producing three quarters of the obtained reinforcers receives three quarters of the responses; actual allocation also depends on factors such as reinforcer quality, delay, response effort and bias. Real allocation deviates in known ways: undermatching is less extreme than the reinforcement proportions, overmatching is more extreme, and bias is a steady preference for one option that the reinforcement rates do not explain.

A motivating operation is a variable that does two things at once. It changes how effective a consequence is as a reinforcer (or punisher) right now: the value-altering effect. And it changes how often behavior that has produced that consequence in the past occurs right now: the behavior-altering effect. For reinforcement, an establishing operation increases consequence effectiveness and evokes behavior previously reinforced by it; an abolishing operation decreases effectiveness and abates that behavior. For punishment, establishing a consequence as more effective generally abates behavior previously punished by it, while abolishing its punishing effectiveness can evoke that behavior. Deprivation and satiation commonly illustrate establishing and abolishing operations for particular reinforcers. In unconditioned motivating operations, the value-altering effect needs no learning history; conditioned ones acquire it through learning.

Both a discriminative stimulus and a motivating operation come before behavior, and both can make it more likely, but for different reasons. A discriminative stimulus signals that a reinforcer is available now if you respond. A motivating operation changes how much that reinforcer is worth now. Conditioned motivating operations include surrogate, reflexive and transitive functions. A surrogate CMO acquires motivating effects through pairing with another MO. A reflexive CMO changes the effectiveness of its own termination: a warning can make termination reinforcing, while a signal whose termination cancels an improvement can make termination punishing. These functions respectively evoke or abate behavior that previously terminated the signal. A transitive CMO changes the effectiveness of another stimulus through a learned dependency, such as a locked door making access to a key reinforcing.

Operant stimulus control is demonstrated when an antecedent stimulus reliably affects responding because of a relevant learning history, commonly differential reinforcement. The difference may involve response frequency, latency, duration or another dimension. Reinforcement can differ between conditions without being completely absent in one of them. Equal rates in two conditions provide no evidence of control in that measure, but do not rule out control of another dimension or by another feature. Control can develop over unintended features, such as position, an adult glance or a mark on a card.

Discrimination is demonstrated when responding differs reliably across stimuli because of the relevant learning history. An SD is correlated with greater availability of reinforcement for a particular response than comparison conditions; an S-delta is correlated with the absence of reinforcement for that response. In a simple discrimination, one stimulus decides whether a response pays off. In a conditional discrimination, which choice is correct depends on another stimulus, the sample, as in matching-to-sample. A generalization gradient shows how responding changes as test stimuli move away from the trained SD: a steep gradient means tight control, a flat one means little control along that dimension.

Stimulus generalization: the trained response occurs in the presence of stimuli that were not part of training, such as new people, places or materials. Response generalization: new responses that were not trained, but do the same job as the trained one, occur. Overgeneralization is responding to stimuli outside the intended class, and it calls for discrimination training.

To assess maintenance, reduce or remove specified intervention supports and check whether the behavior continues over time. Planning a transition to everyday reinforcement and gradually thinning programmed reinforcement can help, but neither guarantees maintenance. Follow-up observations should identify which supports were withdrawn and which contingencies still operate.

Contingency-shaped behavior is acquired through direct contact with consequences. Rule-governed behavior is controlled by a verbal description of a contingency, such as an instruction, a piece of advice or a rule a person tells themselves, often before the person has met the consequence. Rules are especially useful when consequences are delayed, small or improbable. Their cost is that rule-governed behavior can be slow to adjust when the real contingencies change.

Classify verbal operants by their controlling variables and reinforcement history, rather than by the word alone. A mand is controlled by an MO and a history of obtaining a characteristic consequence. A tact is controlled by nonverbal stimulation and typically a history of generalized conditioned reinforcement. An echoic involves a vocal verbal stimulus and vocal responding with point-to-point correspondence and formal similarity. An intraverbal is controlled by verbal stimulation without point-to-point correspondence. Textual behavior reads written verbal stimuli aloud; taking dictation, called transcription here, writes what is heard. Both have point-to-point correspondence without formal similarity. These nonmand operants are typically established through generalized conditioned reinforcement; every occurrence need not receive praise. An autoclitic depends on other verbal behavior of the speaker and changes its effect on a listener. Fetching an object when asked illustrates nonverbal listener responding.

Most verbal responses have more than one source of control. In convergent multiple control, one response is controlled by several variables at once. In divergent multiple control, one variable controls several different responses. Joint control is a special case: the same response form is evoked at the same moment by two sources, for example a word being rehearsed and the tact of an item on a shelf, and that coincidence can control selecting the item or ending a search.

Training some stimulus relations can be followed by the emergence of derived relations that were not directly trained; relevant learning history is still required. Reflexivity is demonstrated by untrained identity matching. Symmetry is demonstrated when training A→B is followed by untrained B→A responding. Transitivity is demonstrated when A→B and B→C training is followed by untrained A→C responding. Equivalence requires evidence that the relevant relations satisfy reflexivity, symmetry and transitivity. Generative performance also includes correct responses to untaught recombinations after selected combinations have been trained, as in matrix training. Derived relations can involve comparison or other relations beyond equivalence; relational frame theory analyzes such contextually controlled patterns.

Imitation has three parts: a model sets the occasion for the behavior, the behavior follows the model closely in time, and it looks or sounds like the model (formal similarity). Observational learning is broader: behavior changes after watching someone else's behavior and its consequences, and the new behavior need not resemble what was observed. Generalized imitation is imitating new models whose imitation was never reinforced, after imitation of many different models was.