Behavior-Change Procedures
The procedures that change behavior, one task at a time: how each one is set up, what it reinforces or removes, and how to tell from data whether it works.
- Concepts
- 19
- Sections
- 8
- Diagrams
- 7
In this chapter
- 01 Reinforcement as a procedure
- 02 Differential and time-based reinforcement
- 03 Teaching discriminations, and the prompts that help
- 04 Models, instructions, teaching formats and groups
- 05 Building new behavior: shaping and chaining
- 06 Generalization and maintenance
- 07 Punishment procedures and side effects
- 08 Emergent relations and generative learning
Reinforcement as a procedure
Knowing what reinforcement is (area B) is not the same as running it well. Here the questions are practical: what to deliver, when, under which motivating conditions, and how to tell from data that it is working.
Figure 1
A token economy, part by part
Target behavior
Reads a page aloud correctly
Token
One sticker per page, given right away
Exchange
Stickers are traded for an item on the menu
Backup reinforcers
- Drawing time
- A board game
- A snack
- Classroom helper job
Gradual schedule thinning
First week
Price: 1 · exchange right away
End of term
Price: 8 · exchange at the end of the day
Reinforcement procedures: adding or removing stimuli
A reinforcement procedure arranges a consequence after a target response and is judged by one thing: whether that response increases. In positive reinforcement something is added or made stronger; in negative reinforcement something is removed, reduced, delayed or prevented. Delivery works best when it is immediate, clearly tied to the response, of enough size and quality, offered while the person is motivated for it and, during acquisition, frequent. Negative reinforcement can be built in on purpose, for instance by letting finished work or a polite request end a task early.
Example
A student who completes the first five problems of a worksheet may skip the rest of it. Over the next week she finishes the first five problems faster and more often. Removing the remaining work after the target response has strengthened it: programmed negative reinforcement.
Common confusion
Name a consequence by its effect on future behavior, not by what staff intend or call it. Soothing words after a tantrum can reinforce tantrums, however kind the intention. Removal does not mean punishment either: when taking something away makes a response more frequent, that is negative reinforcement. And a liked item is not a reinforcer until the data show the response it follows going up.
Key points (5)
- A reinforcement procedure is identified by its effect: the behavior increases when the consequence follows it. In positive reinforcement a stimulus is added or intensified; in negative reinforcement a stimulus is removed, reduced, postponed or prevented (escape and avoidance).
- Preference assessments identify candidate reinforcers; whether a stimulus actually reinforces the target behavior is shown only by data collected while it is delivered contingently.
- Effectiveness depends on immediacy, a clear response–consequence contingency, magnitude and quality of the reinforcer, the current motivating operation and the schedule (dense schedules for acquisition).
- Negative reinforcement can be programmed deliberately, for example by letting an appropriate request or completed work end or postpone a task or other aversive event.
- Evaluating a reinforcement procedure means checking the target behavior's data and watching for consequences that unintentionally reinforce problem behavior, whatever staff call those consequences.
Establishing conditioned reinforcers and token systems
A stimulus can acquire a conditioned reinforcing function through a learning history involving effective reinforcers. Presenting it shortly before an established reinforcer is one common arrangement, not the only possible one. Pairing alone does not establish the function: check whether contingent delivery strengthens responding. A token with an effective history involving several backup reinforcers may function as a generalized conditioned reinforcer and be less dependent on any single motivating operation. Specify what earns tokens, the backup menu and prices, and the exchange procedure. Begin with readily attainable exchanges; change earning requirements or exchange delays gradually only when data and learner needs support the change.
Example
In a reading group, each correctly read page earns a sticker. In the first week, one sticker buys a minute of drawing right away. By the end of the term, eight stickers buy a choice from a menu (a game, a snack, a classroom job) at the end of the day.
Common confusion
Tokens, points and praise do not strengthen behavior on their own; their value comes from what they are paired with or exchanged for. When a token system stops working, check the backups, the prices, the earning rate and the delay to exchange before adding more tokens. Response cost is an optional add-on, not a required part of a token economy.
Key points (5)
- A neutral stimulus can become a conditioned reinforcer when it reliably comes just before an established reinforcer, which is confirmed when presenting it contingently strengthens responding; delivering it afterward, unrelated to the reinforcer or without it weakens or prevents conditioning.
- A generalized conditioned reinforcer is backed by many different reinforcers, which makes it less dependent on any single motivating operation.
- A token economy defines the target behaviors, the tokens, the backup reinforcers and their prices, and when and how exchange happens; response cost is an optional addition.
- Tokens start with frequent, immediate exchange; the number required and the delay to exchange are increased gradually, and the system is thinned toward naturally occurring reinforcers.
- When a token system loses effect, check the value of the backups, earning rates, prices and delays to exchange before adding more tokens.
Building motivating operations and discriminative stimuli into procedures
Two different antecedent variables can be planned. An establishing operation makes a consequence more valuable right now and evokes behavior that has produced it; it can be caught when it happens or set up on purpose. An abolishing operation, such as having just had plenty of something, makes that consequence less valuable, which weakens teaching that relies on it but can also be used to reduce problem behavior before it starts. A discriminative stimulus is different: through a relevant learning history, it signals greater availability of reinforcement for a particular response than in comparison conditions; reinforcement need not be completely absent elsewhere.
Example
To teach a girl to ask for juice, the Instructor sets a clear bottle of juice that she cannot open on the table before snack, when she has not had a drink for a while. Running the same lesson right after lunch, with a full cup already finished, would make juice worth much less and the requests would be rare.
Common confusion
A motivating operation changes how much a consequence is worth; a discriminative stimulus signals whether it is available. Showing, naming or highlighting a stimulus does not make it discriminative; differential reinforcement does. Prompting a request for an item the learner does not want at the moment teaches very little.
Key points (4)
- Establishing operations raise a reinforcer's value and evoke behavior that has produced it; they can be captured or contrived (for example, a wanted item in view but out of reach) to teach mands.
- Abolishing operations such as satiation lower a reinforcer's value; they weaken teaching that relies on that reinforcer and can be used deliberately as antecedent interventions for problem behavior.
- A stimulus becomes a discriminative stimulus only through differential reinforcement: the response is reinforced in its presence and not in its absence (S-delta).
- Planning separates the two variables: a motivating operation changes how valuable a reinforcer is, whereas a discriminative stimulus signals whether that reinforcer is available.
Differential and time-based reinforcement
These procedures reduce or reshape behavior mainly through reinforcement. Their names are exam favorites because each one reinforces something different: an alternative, an absence, a low rate or nothing in particular.
Figure 2
What each procedure reinforces
-
DRA Alternative
Reinforce a specified alternative behavior.
-
DRI Incompatible
Reinforce an alternative that cannot co-occur with the target.
-
DRO Other behavior
Reinforce when the interval passes with no target response.
-
DRL Low rate
Reinforce a target response that comes after a long enough gap.
-
FT Fixed time (NCR)
Deliver by the clock, whatever the person is doing.
Time in session →
Differential reinforcement procedures
DRA arranges reinforcement for a specified alternative; DRI uses an alternative physically incompatible with the target. Functional communication training selects an alternative that obtains the maintaining reinforcer. Whole-interval DRO requires no target response throughout the interval; momentary DRO checks absence only at the scheduled observation moment. Select and adjust the initial interval using baseline data and feasibility. DRL arranges reinforcement for a lower rate or sufficiently spaced responses, while DRH arranges it for a higher rate. DRA can be implemented with or without extinction for the target; when target reinforcement continues, adjust relative quality, delay, magnitude or effort to favor the alternative. Measure the behaviors and criteria relevant to the selected procedure.
Example
A student answers so often that classmates rarely get a turn. His teacher wants lower participation without eliminating it. He earns a point for one to five appropriate spoken answers during a lesson; zero answers do not earn the point. The criterion is adjusted from baseline and monitored. This full-session DRL arrangement reinforces a reduced but nonzero amount of participation.
Common confusion
DRO specifies absence of the target, not a particular replacement response, so other behavior occurring at delivery may also contact reinforcement. Whole-interval DRO requires absence throughout its interval; momentary DRO permits earlier target responses if the target is absent at the observation moment. DRL targets reduced responding rather than its elimination. Extinction is a separate contingency decision, not part of the definition of every DRA procedure.
Key points (5)
- DRA reinforces a specific alternative behavior; DRI is the form of DRA in which the alternative cannot occur at the same time as the target. An alternative that produces the same reinforcer as the problem behavior (as in functional communication training) competes best.
- DRO delivers reinforcement when the target behavior is absent throughout an interval (whole-interval) or at the moment the interval ends (momentary); intervals can be fixed or variable, resetting or not; the initial interval is usually set at or below the baseline interresponse time and then lengthened.
- DRL reinforces responding at or below a criterion to lower a rate without eliminating the behavior (full-session, interval and spaced-responding versions); DRH reinforces responding at or above a criterion to raise a rate.
- With extinction, the problem behavior no longer produces its reinforcer; without extinction, it still does, so the alternative must earn a richer, faster or higher-quality reinforcer to compete.
- Differential reinforcement is evaluated with data on both the problem and the alternative behavior, and its criteria are adjusted gradually as performance stabilizes.
Time-based (noncontingent) reinforcement schedules
Fixed-time and variable-time schedules deliver a stimulus after a fixed or varying amount of time, independently of responding. Variable-time values vary around a programmed average. Fixed- and variable-interval schedules instead require a qualifying response after the interval. Time-based delivery of a maintaining reinforcer can reduce problem behavior by reducing its current value; it can be arranged with or without extinction of the problem behavior. Specify separately whether problem behavior still produces that reinforcer. Use baseline data to select an initial schedule and thin it gradually when outcome and safety data support doing so.
Example
A man in a group home calls out for staff every few minutes. Staff begin stopping by for a short chat every three minutes on a timer, whether or not he has called out, and calling out drops. Over several weeks the timer is stretched to ten minutes, and he is taught to press a call button for anything urgent.
Common confusion
NCR follows elapsed time, whereas DRO makes delivery depend on the absence of a target response. Response-independent delivery can accidentally follow and strengthen behavior. It does not establish a contingency for a specified replacement response, so teach and reinforce that response separately when needed.
Key points (4)
- Fixed-time and variable-time schedules deliver a stimulus after a set or average amount of time, independent of responding; interval schedules, by contrast, require a response after the interval.
- Noncontingent reinforcement with the maintaining reinforcer can reduce problem behavior by lowering the motivating operation and by breaking the relation between the behavior and the reinforcer.
- Design: use the functional reinforcer (or one shown to compete with it), start with a dense schedule informed by baseline, and thin the schedule gradually while the data stay low.
- Limitations: deliveries can coincide with problem behavior and reinforce it adventitiously, and a time-based schedule does not teach a replacement behavior, so it is often combined with other procedures.
Teaching discriminations, and the prompts that help
Most teaching aims to bring a response under the control of the right stimulus. Prompts get correct responses started; fading moves control from the prompt to the stimulus that should be in charge.
Figure 3
Prompting from the top or from the bottom
Most-to-least
Begins with strong help and lowers it across trials as the learner succeeds.
- Full physical Start
- Partial physical
- Model
- Gesture
- Verbal
- Independent
Least-to-most
Waits for an independent response, then adds more help within the trial only if needed.
- Full physical
- Partial physical
- Model
- Gesture
- Verbal
- Independent Start
Teaching simple and conditional discriminations
In a simple discrimination, responding pays off when one stimulus is present and does not pay off when another stimulus is present, or when the first one is missing. In a conditional discrimination, which choice is correct depends on another stimulus, the sample: matching-to-sample and picking an item when it is named both work this way. The match can be by identity or arbitrary. Rotate positions, vary the order of trials, present several comparisons and let every target also appear as a distractor, so that only the intended feature can predict reinforcement.
Example
A child is shown three pictures and asked "Where is the shoe?" Which picture is right depends on the word spoken, so this is a conditional discrimination. If the shoe always sat on the left, the child could score perfectly by touching the left picture without listening at all.
Common confusion
Do not confuse the sample, the stimulus that sets the occasion, with the comparisons, among which the learner chooses. Touching the only available picture does not demonstrate selecting among comparisons conditional on a sample; a simple go/no-go discrimination would require separate evidence of responding differently across stimulus conditions. And high accuracy needs checking: an error pattern tied to position or order reveals control by the wrong thing.
Key points (4)
- In a simple discrimination, a response is reinforced in the presence of one stimulus (SD) and not in the presence of another or in its absence (S-delta).
- In a conditional discrimination, which comparison is correct depends on a sample stimulus, as in matching-to-sample or selecting an item when it is named; matching can be by identity or arbitrary.
- Procedural safeguards reduce the risk of faulty stimulus control: rotate positions, vary trial order, use several comparisons and have each target serve as a distractor; probes check the actual controlling stimuli.
- Accurate responding is evaluated for the intended control: error patterns and probes reveal control by position, order, an irrelevant feature or a single available comparison.
Choosing and evaluating prompts
Response prompts are things the Instructor does to help the response: a verbal hint, a gesture, a model or physical guidance. Stimulus prompts change or add to the materials: a within-stimulus prompt alters the relevant feature itself, while an extra-stimulus prompt adds a separate cue. Most-to-least prompting starts with strong help and reduces it across trials; least-to-most gives the learner a chance first and adds help only when needed. Errorless teaching arranges prompts so that errors are unlikely from the start. Use the least intrusive prompt that works, and plan its removal before you start.
Example
A boy is learning to put socks in the drawer matching their color. Pointing at the right drawer is a gestural response prompt. A colored dot stuck on the drawer is an extra-stimulus prompt. Printing the drawer labels larger and brighter is a within-stimulus prompt, because it changes the very feature he should attend to.
Common confusion
An instruction and an added prompt have different roles; merely repeating the instruction does not establish effective prompting, although a planned repetition can function as an additional verbal cue depending on the procedure and learning history. Physical guidance is the most intrusive response prompt, a gesture is far less so. A prompt with no fading plan, or data where prompted responses never become independent, points to prompt dependence.
Key points (4)
- Response prompts are instructor behaviors directed at the learner's response (verbal, gestural, model, physical); stimulus prompts change or add to the stimuli themselves (within-stimulus or extra-stimulus).
- Most-to-least prompting starts with the most assistance and reduces it across trials; least-to-most starts with the least and adds assistance only when needed. The choice depends on the learner's skills and how costly errors are.
- Errorless teaching arranges prompts and stimulus changes so that errors are unlikely from the start, rather than correcting errors after they happen.
- The least intrusive effective prompt is preferred, and every prompt is planned with a way to fade it; data on prompted and independent responses reveal prompt dependence.
Fading prompts
Time delay changes when a prompt arrives. Constant time delay starts with the prompt given immediately and then waits a fixed number of seconds; progressive time delay lengthens the wait in small steps. Graduated guidance adjusts physical help moment to moment during a single response, down to shadowing the learner's hand. Stimulus fading gradually changes one dimension of a stimulus prompt, such as size or color, while stimulus shaping gradually changes the form of the stimulus itself. Let performance data set the pace: fading too fast can increase errors, while unnecessary delay in fading can contribute to prompt dependence.
Example
In the first sessions, the Instructor shows a sight word and says it immediately, so the learner repeats it correctly. From then on she waits four seconds before saying it. Most trials soon end with the learner reading the word before the prompt comes.
Common confusion
Changing the delay before a prompt is not the same as changing how much help the prompt gives. Graduated guidance is adjusted within a response; a most-to-least hierarchy is changed across trials. And a prompt is not faded on a calendar but when the data show the learner is ready.
Key points (4)
- Time delay inserts a pause between the SD and the prompt: constant time delay moves from a 0-second delay to a fixed delay, whereas progressive time delay lengthens the delay gradually.
- Graduated guidance adjusts physical help moment to moment within a response, including shadowing; a most-to-least hierarchy changes the prompt level across trials.
- Stimulus fading gradually changes a dimension of a prompt (intensity, size, color, position); stimulus shaping gradually changes the form of the stimulus itself.
- Fading decisions follow performance data so that stimulus control transfers to the natural SD: fading too fast produces errors, too slowly produces prompt dependence.
Models, instructions, teaching formats and groups
How teaching is delivered matters as much as what is taught: by showing, by telling, in planned trials or in everyday moments, and sometimes for a whole group at once.
Modeling and imitation procedures
Assess whether the learner attends to the relevant visual or auditory model and has the component responses needed for the task; teach or prompt missing prerequisites as needed, including imitation skills when these are the teaching target. Imitation means the model is what controls the response: the response follows shortly after and resembles the model in form. Models can be live, recorded on video or recorded of the learners themselves; a model similar to the learner, or one seen getting reinforced, tends to be copied more. Behavioral skills training combines instructions, a model, rehearsal and feedback until a performance criterion is met. Generalized imitation is checked with new models that were never trained.
Example
New staff learn to run a preference assessment. They read the steps, watch a short video of a colleague doing it, practice with a supervisor playing the learner, and receive feedback after each run until they complete every step correctly twice in a row.
Common confusion
Two people doing the same thing is not imitation unless one person's action controls the other's: neighbors opening umbrellas when rain starts are responding to the rain. Instructions plus a demonstration, without rehearsal and feedback, are not behavioral skills training. Copying trained models accurately does not yet show generalized imitation.
Key points (4)
- Modeling requires that the learner attends to the model and has an imitative repertoire; missing prerequisites are taught first.
- Imitation means the model is the controlling stimulus, the response follows soon after and resembles it in form; similar behavior that occurs for other reasons is not imitation.
- Models can be live or recorded (video modeling, video self-modeling); similarity to the learner and seeing the model's behavior reinforced can make imitation more likely.
- Behavioral skills training combines instructions, modeling, rehearsal and feedback until the trainee meets a performance criterion; generalized imitation is checked with novel, untrained models.
Instructions and rules
A rule states a contingency in words, such as "if you save every week, you will afford the trip." Behavior controlled by rules differs from behavior shaped directly by contact with its consequences. Rules help most when consequences are delayed, small but cumulative, or improbable on any single occasion, though following them still needs supporting contingencies. An instruction should specify the action clearly in language the learner understands; it need not state a consequence, whereas a rule describes a contingency. Check attending and provide the planned opportunity to respond and consequence. Rule-following can persist after the real contingencies change.
Example
A student has missed several library returns. Before the next loan, she reads that returning a book by Friday avoids the stated late fee. She sets a reminder and returns it on Thursday, before she has ever paid such a fee. The stated contingency can guide her action without direct experience of the fee; the reminder also supports performance.
Common confusion
Stating a rule does not guarantee that it will be followed; that depends on the person's history with rules and on the present contingencies. Repeating an instruction several times without follow-through can teach that the first few can be ignored. Watch the data when conditions change, because rule-governed behavior can lag behind.
Key points (4)
- Rules and instructions are verbal stimuli that describe contingencies; behavior controlled by them (rule-governed) differs from behavior shaped directly by contact with consequences (contingency-shaped).
- Rules are especially useful when consequences are delayed, small but cumulative, or improbable; following them still needs supporting contingencies.
- Effective instructions specify the behavior, the conditions and the consequence, use the learner's vocabulary, are given once when the learner is attending and are followed through.
- Rule-governed behavior can be insensitive to changes in the actual contingencies, so data are checked when conditions change.
Trial-based and free-operant teaching
Trial-based arrangements divide teaching into distinct opportunities, each with a beginning, a response opportunity and an outcome. In a free-operant arrangement, the response can recur without a new trial being initiated for each occurrence. This distinction concerns response opportunity, not whether teaching occurs at a table or during play, or whether the learner initiates an interaction. Naturalistic and incidental teaching can include discrete opportunities. Choose measures that answer the question: accuracy per opportunity, rate, latency, duration or independence may be relevant in either setting.
Example
In one activity, an Instructor presents a picture, waits for a selection, records the outcome and removes the materials before the next trial. In another, an accessible sorting tray remains available for ten minutes, and the learner can place another item whenever ready, without waiting for a new adult-presented trial. The team records correct selections per trial in the first activity and independent placements per minute in the second.
Common confusion
Learner initiation and natural materials do not by themselves establish a free-operant arrangement. Inspect whether another response opportunity requires the next trial to begin. Select the measure for the behavior and decision, rather than assuming that all trial teaching uses only percent correct or all naturalistic teaching uses rate.
Key points (4)
- Trial-based teaching (such as discrete trials) presents instructor-controlled learning opportunities, each with an SD, a response, a consequence and an intertrial interval; it is often summarized as percent correct.
- In free-operant arrangements, the learner can respond at any time and repeatedly, without a new trial for each response; behavior is often summarized as rate. Incidental and naturalistic teaching, often started by the learner, can be organized either way.
- Formats are chosen by goal: trial-based teaching gives many controlled opportunities for acquisition, while naturalistic teaching supports motivation, spontaneity and generalization; they are often combined.
- Each format is evaluated for its typical risks: rote or prompt-bound responding in trials, and too few learning opportunities when the environment is not arranged in free-operant teaching.
Group contingencies
A group contingency applies one consequence arrangement to a group. In an independent contingency, everyone faces the same rule and each person earns based on his or her own behavior. In a dependent contingency, the whole group's reinforcement depends on the performance of one member or a few. In an interdependent contingency, reinforcement for group members depends on a collective criterion, such as every member meeting a target or the class meeting a specified average. Every member must be able to meet the criterion.
Example
A teacher offers ten extra minutes of recess. If each student who scores at least 80% earns them, the contingency is independent. If the class earns them when one selected student reaches 80%, it is dependent. If the class earns them only when the class average reaches 80%, it is interdependent.
Common confusion
Dependent and interdependent are easy to swap: dependent hangs on one member or a subset, interdependent on everyone or on the group as a whole. Group contingencies carry risks, such as peer pressure, sabotage, and a member losing reinforcement because of someone else's behavior, so they are monitored, and a criterion some members cannot reach is unfair by design.
Key points (4)
- In an independent group contingency, the same contingency applies to every member, and each earns reinforcement based on his or her own behavior.
- In a dependent group contingency, the whole group's reinforcement depends on the performance of one member or a small subset.
- In an interdependent group contingency, every member must meet the criterion, or the group must meet it as a whole (group average, randomly selected member, team games), for anyone to earn reinforcement.
- Group contingencies are designed so that every member can meet the criterion and are monitored for peer pressure, sabotage and members being penalized for others' behavior.
Building new behavior: shaping and chaining
Both procedures create behavior the learner does not yet show, but in different ways. Shaping changes one response a little at a time; chaining links several different responses into a sequence.
Figure 4
Shaping and chaining build behavior differently
Shaping
One response; the criterion moves.
- 5 s
- 15 s
- 1 min
- 10 min
- All morning
- No longer reinforced
- Reinforced now
- Still ahead
Chaining
Several different responses, linked in order.
- Pick up the brushIts result reinforces this step and signals the next
- Add toothpasteIts result reinforces this step and signals the next
- BrushIts result reinforces this step and signals the next
- RinseIts result reinforces this step and signals the next
- Put the brush awayRoutine finished: the final reinforcer
Figure 5
Three ways to teach the same routine
- 1 Pick up brush
- 2 Toothpaste
- 3 Brush
- 4 Rinse
- 5 Put away
Forward. Step 1 is taught; the Instructor helps with the rest.
Backward. Step 5 is taught; the Instructor helps with the steps before it.
Total task. Every step, every trial, with prompts where needed.
- Taught in this phase
- Done with the Instructor's help
- Done by the learner, prompted as needed
Shaping
Shaping reinforces successive approximations to a terminal behavior. It starts from something the learner already does, reinforces it, and then moves the criterion closer to the goal; earlier approximations stop being reinforced once closer ones occur. It can produce new forms of behavior or change one form along a dimension such as duration, latency, rate or force. The data set the pace: a step that is too large leaves the learner without reinforcement and responding fades, while staying too long at one step makes it hard to move past.
Example
A child pulls out a new hearing aid within seconds. The Instructor reinforces wearing it for five seconds, then for fifteen, then for a minute, raising the requirement only when the current one is met reliably. The response does not change form; its duration is shaped.
Common confusion
Shaping changes the response itself; stimulus fading changes an antecedent stimulus; chaining links different responses. Continuing to reinforce an old approximation after a closer one appears stalls progress. And shaping is not only for new forms of behavior: it also shapes how long, how fast or how forcefully a behavior happens.
Key points (4)
- Shaping differentially reinforces successive approximations of a terminal behavior, starting from a response already in the repertoire and extinguishing earlier approximations as closer ones occur.
- Shaping can work across topographies (new forms) or within one topography along a dimension such as duration, latency, rate or magnitude.
- Step size and the time spent at each step are set by the data: steps that are too large produce extinction of responding, and lingering too long at one step makes it hard to move on.
- Shaping changes the response; stimulus fading changes an antecedent stimulus; chaining links distinct responses into a sequence.
Chaining
A task analysis breaks a complex skill into teachable steps, and it is checked and adjusted by watching which steps this learner already performs. In a chain, the stimulus change each response produces does two jobs: it signals the next step, and it reinforces the step just finished as a conditioned reinforcer. Forward chaining teaches the first step first. Backward chaining teaches the last step first, so the learner finishes the chain and meets the final reinforcer on every trial. Total-task chaining teaches every step on every trial. Choose from the learner's repertoire, the chain's length and difficulty, and the data.
Example
Brushing teeth is analyzed into five steps. In backward chaining, the Instructor helps with the first four and the learner puts the brush away independently, finishing the routine each time; once that step is mastered, the learner takes over rinsing too.
Common confusion
Backward chaining starts at the last step and adds earlier ones; forward chaining starts at the first. Total-task chaining is not forward chaining: every step is practiced on every trial, with prompts where needed. A generic task analysis copied from elsewhere should be validated before use.
Key points (4)
- A task analysis breaks a complex skill into teachable steps and is validated and adjusted to the learner by observing which steps the learner already performs.
- In a behavior chain, the stimulus change produced by each response serves as the SD for the next response and as a conditioned reinforcer for the previous one.
- Forward chaining teaches the first step first; backward chaining teaches the last step first, so the learner completes the chain and contacts the terminal reinforcer on every trial; total-task chaining teaches all steps on every trial.
- The chaining method is selected from the learner's repertoire, the length and difficulty of the chain and the data; problem chains can be interrupted or broken at early links.
Generalization and maintenance
A skill that appears only with one person, in one room, for one month has not been fully taught. Spread across conditions, and persistence over time, have to be planned from the start and measured.
Figure 6
Generalization goes sideways, maintenance goes up
Later
Still greets the same staff, three months after teaching ended
↑ Time after teaching stops
Taught
Greets three staff members, at school, while teaching is running
-
New person
Greets a visiting teacher he has never practiced with
-
New place
Greets his grandmother in the hallway at home
Untrained conditions →
Response generalization
Taught to say "hi," he also starts to wave or say "hello," which do the same job
Ways to program generalization
- Teach enough examples
- Program common stimuli
- Train loosely
- Indiscriminable contingencies
- Self-mediation
- Natural contingencies
Programming for generalization
Stimulus generalization is responding under untrained conditions: new stimuli, places or people. Response generalization is the appearance of untrained responses that do the same job as the trained one. Strategies include teaching enough examples to sample the range the learner will meet, bringing common stimuli from the target setting into teaching, training loosely, making it unclear which responses will be reinforced, teaching self-mediating cues, and teaching behavior that natural contingencies will keep going. Plan generalization from the beginning and probe it in untrained conditions.
Example
A boy learns to greet people with three different staff members, in the hallway and in his classroom. After mastery, his grandmother waits in the hallway at home and the team records whether he greets her: a probe of a person and a place never used in teaching.
Common confusion
Generalization is spread across conditions; maintenance is persistence over time. Programming common stimuli brings features of the target setting into teaching, while teaching multiple examples varies the trained stimuli. Generalization can also be unwanted, for example calling every adult man "daddy," and that is handled by teaching the relevant discrimination.
Key points (4)
- Stimulus generalization is responding in the presence of untrained stimuli, settings or people; response generalization is the emergence of untrained responses that are functionally equivalent to the trained one.
- Strategies include teaching enough exemplars (general case analysis), programming common stimuli, training loosely, using indiscriminable contingencies, mediating generalization and teaching behavior that will contact natural contingencies.
- Generalization is planned from the start and probed in untrained conditions; it is not assumed after training ends.
- Undesired generalization (overgeneralization) is also possible and is addressed by teaching the relevant discriminations.
Programming for maintenance
Maintenance is continued behavior over time after specified intervention supports are reduced or withdrawn. Plan follow-up probes and state which supports have changed. Gradual schedule thinning and a shift toward everyday reinforcement may help; a history of intermittent reinforcement can increase resistance to extinction under some conditions, but neither intermittency nor thinning guarantees maintenance. Thin according to performance and safety data, and watch for deterioration or ratio strain. Self-management, recruiting reinforcement and booster teaching may also help. Retain supports that the person continues to need.
Example
A young adult learned to greet coworkers, with praise from a job coach after every greeting. The coach praises less and less often while coworkers' replies take over. Probes at one and three months show the greetings still happening; if they had dropped, a short booster session would follow.
Common confusion
Mastery during teaching does not establish maintenance after support changes. Continued performance under an unchanged schedule answers a different question from performance after that schedule is withdrawn. Continuous or contrived reinforcement can remain appropriate; the goal is an effective, acceptable arrangement, with evidence about what happens when particular supports change.
Key points (4)
- Maintenance is the continued occurrence of a behavior after the intervention, or part of it, has been withdrawn; it is planned and checked with follow-up probes.
- Schedule thinning gradually makes reinforcement less frequent, for example from continuous to intermittent or to a leaner intermittent schedule, guided by data; intermittent schedules make behavior more resistant to extinction, and thinning too fast risks ratio strain.
- Control is transferred to naturally occurring reinforcers by pairing contrived reinforcers with natural ones, using behavior traps and teaching the learner to recruit reinforcement.
- Other maintenance supports include delaying reinforcement gradually, self-management and booster sessions when probes show decline.
Punishment procedures and side effects
Punishment procedures are identified by their effect on future behavior. Considering them requires an individual assessment of risks and alternatives, appropriate authorization and review, teaching useful alternatives, and ongoing checks for benefit and harm.
Punishment procedures
Punishment is a process defined by its effect: a consequence that makes the behavior it follows less likely. Positive punishment adds something, such as a reprimand, contingent exercise or overcorrection; restitutional overcorrection restores the environment beyond its earlier state, and positive practice repeats a correct form. Negative punishment removes access to reinforcement: time-out, exclusionary or not, and response cost, which removes a set amount of a reinforcer the person holds. Before recommending a restrictive or punishment procedure, document evidence that less intrusive approaches have not achieved the needed outcome. Any risk-based exception requires the existing intervention team to document a comparison of the danger to the client and the risks of the proposed procedure. Obtain informed consent and any required organizational or external review, seek assent as applicable, verify implementer competence, and reinforce useful alternatives. Set explicit outcome, adverse-effect and stopping criteria; promptly revise or end an ineffective or harmful procedure.
Example
A student who marks his desk with a pen cleans his desk and two neighboring desks before recess (restitutional overcorrection). Over two weeks the marking becomes rare, and he earns praise and tokens for keeping his workspace clean. If the marking had not decreased, the procedure would not be functioning as punishment.
Common confusion
A procedure meant to be unpleasant is not punishment until the data show a decrease. Time-out depends on a reinforcing time-in, and for behavior maintained by escape it can deliver the very break the behavior was getting. Time-out removes access for a period; response cost removes a specific amount. When a reduction procedure is not working, review function, reinforcement and implementation before making it longer or stronger.
Key points (4)
- Positive punishment adds a stimulus or activity contingent on the behavior (for example reprimands, contingent exercise, overcorrection); negative punishment removes access to reinforcement (time-out, response cost). Either is confirmed only by a decrease in the behavior.
- Time-out removes access to reinforcement for a period (nonexclusionary or exclusionary) and depends on a reinforcing time-in; it can reinforce behavior maintained by escape.
- Response cost removes a specific amount of a reinforcer the learner has, such as tokens, and requires a reserve so that the learner does not go into debt; overcorrection can be restitutional (restoring the environment beyond its prior state) or positive practice (repeating a correct form).
- Punishment procedures are considered after less restrictive options, combined with reinforcement of alternatives, applied immediately and consistently, and monitored with data and for side effects.
Emotional and elicited side effects
Reduction procedures can have effects beyond the target behavior. During extinction, the behavior may briefly increase, become more varied or intense, come with emotional responding or aggression, and return briefly after a break; none of these is certain. Punishment and other aversive procedures can elicit emotional reactions and aggression and can lead the person to escape or avoid the people and places involved, which may come to function as conditioned aversive stimuli. Behavioral contrast can occur when changing the reinforcement arrangement in one component is accompanied by an opposite change in responding in another component whose reinforcement arrangement remains unchanged; behavior worsening elsewhere should be assessed before being labeled contrast. Good plans anticipate all of this.
Example
At a day program, staff stop giving snacks when a man cries for them and start reinforcing his requests instead. For two days the crying gets louder, and once he pushes a chair. Staff had warned his family, had a safety plan ready and kept reinforcing requests; by the second week the crying is rare.
Common confusion
A burst during extinction does not mean the procedure has failed, and a brief return after a break is consistent with spontaneous recovery rather than proof of failure. Aggression appearing during extinction may be elicited, belong to the same operant response class, or have other controlling variables; its timing alone does not establish its function, so assess it and apply the safety plan. Behavior in other settings is worth watching too.
Key points (4)
- Extinction can produce a temporary increase in the behavior (extinction burst), emotional responding and aggression, and later spontaneous recovery.
- Punishment and other aversive procedures can elicit emotional reactions and aggression, and can evoke escape and avoidance of the people and settings involved.
- People and settings paired with aversive procedures or heavy demands can become conditioned aversive stimuli; reducing behavior in one setting can increase it in another setting whose contingencies did not change (behavioral contrast).
- Plans anticipate these effects: inform caregivers, prepare safety measures, combine reduction procedures with reinforcement and track collateral behavior.
Emergent relations and generative learning
Some procedures are valued for what they produce without direct teaching. The measure of success is how many untrained responses appear for each relation that was trained.
Figure 7
Matrix training: teach three, test six
| Color / object | Car | Ball | Cup |
|---|---|---|---|
| Red | red car Taught | red ball Probe, no feedback | red cup Probe, no feedback |
| Blue | blue car Probe, no feedback | blue ball Taught | blue cup Probe, no feedback |
| Green | green car Probe, no feedback | green ball Probe, no feedback | green cup Taught |
3 combinations taught, 6 probed
Emergent relations and generative learning
Stimulus equivalence is shown by untrained relations: reflexivity (matching a stimulus to itself), symmetry (after A→B, choosing A given B) and transitivity (after A→B and B→C, choosing C given A), and combined tests such as C→A. A training structure, whether one-to-many, many-to-one or linear, decides which relations are trained; emergent ones are tested in probes without reinforcement or feedback. Matrix training teaches a selected set of combinations, often the diagonal, so that the rest can appear through recombinative generalization. A generative procedure is judged by how many untrained relations emerge per trained relation.
Example
A child is taught "red car," "blue ball" and "green cup" by labeling pictures. Without further teaching, she correctly names "red ball," "green car" and the other combinations in probe trials. Three trained combinations have produced six untrained ones.
Common confusion
Emergent relations are derived from trained relations, not from physical resemblance: a photo matched to a drawing of the same dog may be generalization. If a probe gives feedback, later correct responses may have been learned during the test. And not every cell of a matrix needs direct teaching.
Key points (4)
- Stimulus equivalence is shown by untrained relations: reflexivity (A = A), symmetry (after A→B, B→A) and transitivity (after A→B and B→C, A→C), with combined tests such as C→A.
- Training structures (one-to-many, many-to-one, linear) determine which relations are trained; emergent relations are tested in probes without reinforcement or feedback.
- Matrix training teaches a selected subset of combinations (often a diagonal) so that untrained combinations emerge through recombinative generalization.
- Emergent relations are derived from trained relations, not from physical similarity; generative procedures are evaluated by how many untrained responses emerge per trained relation.
Keep going
Practice the whole chapter in the course
Original questions on every task of area G, with explanations and a record of which tasks you have mastered.
Try a sample free, without an account