Generalization, Maintenance, Punishment, Emotional Effects, and Emergent Relations
Key Takeaways
- Generalization (across stimuli, settings, people, responses, time) and maintenance (durability after intervention is thinned) must be PROGRAMMED from the start, using Stokes and Baer strategies.
- Extinction (withholding the maintaining reinforcer) predictably produces an extinction burst, increased variability, induced aggression/emotional behavior, and possible spontaneous recovery and resurgence.
- Function-based reductive plans pair extinction of problem behavior with reinforcement of an alternative (DRA/FCT); extinction alone is rarely appropriate.
- Punishment is defined by a future DECREASE in behavior; positive punishment adds a stimulus, negative punishment (response cost, time-out) removes one - both require justification, safeguards, monitoring, consent, and reinforcement-based alternatives.
- Emergent/derived relations (e.g., A-C and C-A after teaching A-B and B-C) reflect generative responding not directly taught and expand learning beyond trained targets.
Generalization and Maintenance Must Be Programmed
Generalization is when behavior change extends beyond the training conditions — across stimuli, settings, people, responses, and time. Stimulus generalization is responding to new but similar antecedents; response generalization is the emergence of untrained but functionally similar responses. Maintenance is the persistence of behavior after the intervention is thinned or removed.
Both are programmed from the beginning, not bolted on after clinic success. A skill that works only with one therapist at one table has limited social validity. Stokes and Baer's classic framework ("train and hope" is the failure mode) provides the menu of strategies.
Stokes & Baer Generalization Strategies
| Strategy | What you do |
|---|---|
| Train sufficient exemplars | Teach with multiple examples of stimuli/responses until generalization appears |
| Program common stimuli | Embed features of the natural setting into training |
| Train loosely | Vary noncritical aspects of instruction so behavior is not bound to one cue |
| Mediate generalization | Teach a self-prompt/strategy the learner carries across settings |
| Use natural maintaining contingencies | Move the behavior into contact with reinforcers already present in the environment |
| Reinforce generalized responding | Explicitly reinforce novel/untrained instances |
Maintenance specifically requires schedule thinning toward intermittent reinforcement, contact with natural reinforcement, caregiver training, and periodic probes. A behavior that collapses once dense reinforcement ends signals that natural contingencies have not taken over.
Extinction and Its Predictable Side Effects
Extinction is withholding the reinforcer that maintains a behavior, which decreases the behavior over time. Crucially, the extinction procedure must match the function: ignoring an escape-maintained behavior is not extinction (it may even reinforce escape); escape extinction means the demand stays in place. Attention extinction means attention is withheld.
Extinction reliably produces side effects the exam tests:
- Extinction burst: a temporary increase in frequency/intensity/duration right after reinforcement stops.
- Increased response variability and extinction-induced aggression/emotional responding.
- Spontaneous recovery: reappearance of the behavior after time, even without reinforcement.
- Resurgence: return of a previously reinforced behavior when a newer alternative is itself extinguished.
Because extinction alone provokes bursts and emotional responding, it is almost always combined with reinforcement of an alternative — DRA, typically delivered as functional communication training (FCT). FCT teaches a communicative response that produces the same reinforcer as the problem behavior, then places problem behavior on extinction. This is the prototypical function-based reductive package: reinforce the replacement, extinguish the problem behavior, and add antecedent (MO) supports.
Punishment and Its Safeguards
Punishment is defined by a decrease in future behavior following a consequence — never by how harsh the procedure sounds. Positive punishment adds a stimulus (e.g., a reprimand) and behavior decreases. Negative punishment removes a stimulus or access to reinforcement and behavior decreases — response cost (removing earned reinforcers, e.g., tokens) and time-out from positive reinforcement are the canonical examples. Time-out works only if "time-in" is genuinely reinforcing.
Punishment-based procedures demand careful review because they can produce emotional responding, escape/avoidance of the punishing agent, aggression, modeling of coercion, and suppression without teaching a replacement skill. The exam favors reinforcement-based, function-based, least-intrusive procedures unless punishment is clearly justified, consented to, monitored, and paired with reinforcement of alternatives (the least-restrictive/effective treatment principle in the Ethics Code).
When punishment is used, several safeguards are expected: obtain informed consent and review by a human-rights or peer committee where required; deliver it immediately, consistently, and at adequate intensity (gradually escalating intensity builds tolerance and is contraindicated); always run a concurrent reinforcement program for the replacement behavior; and monitor data to discontinue the procedure as soon as the behavior allows. Punishment suppresses behavior but teaches nothing — it never replaces the skill-building half of the plan, which is why reinforcement of alternatives is non-negotiable.
Side Effects of Abrupt Change
Extinction, punishment, and abrupt reinforcement thinning can produce bursts, variability, resurgence, spontaneous recovery, and emotional responding. These are anticipated and monitored with data, with safety, feasibility, and caregiver capacity built into the plan. A procedure that is technically correct but ignores safety or social validity is still a weak selection on the exam.
Emergent Relations and Generative Performance
Emergent (derived) relations appear without direct teaching. Stimulus equivalence is demonstrated when a learner shows three derived properties: reflexivity (A=A, matching identical stimuli without training), symmetry (if A-B is taught, B-A emerges), and transitivity (after A-B and B-C, A-C emerges). The combined A-C plus C-A test is sometimes called the equivalence test. Generative performance is valuable because it multiplies the outcomes of instruction beyond directly trained targets, improving efficiency and supporting language and academic growth — teaching a few relations yields many untaught ones.
For selection items, ask whether the goal is durable cross-context use (generalization/maintenance programming), immediate suppression of dangerous behavior (effective, safeguarded, function-based reduction), or generative responding (equivalence-based instruction). The best answer pairs effectiveness with social validity, data review, and support for alternative behavior.
A final integrating principle: the strongest Domain G plans stack procedures — antecedent/MO supports to prevent evocation, reinforcement of a functionally equivalent replacement (DRA/FCT), extinction of the problem behavior, and explicit generalization and maintenance programming so gains survive outside the clinic. Punishment, if ever used, sits last and only with safeguards. Exam answers that select one isolated technique while ignoring function, safety, or durability are the predictable distractors.
Shortly after a team stops delivering attention for a child's whining (attention extinction), whining briefly becomes louder and more frequent before declining. The team should:
A child's tantrums are maintained by escape from demands. A staff member 'ignores' the tantrum but removes the demand and walks away. Why is this NOT extinction?
In a token system, a learner loses two earned tokens each time he swears, and swearing decreases over time. This procedure is BEST classified as:
After directly teaching a learner that the spoken word 'dog' goes with a picture of a dog (A-B) and that the picture goes with the written word DOG (B-C), the learner - without direct training - matches the spoken word 'dog' to the written word DOG (A-C). This BEST illustrates: