The big idea
A skill that appears only at the therapy table, with one therapist and one set of materials, is of little use. Generalized behavior change means that the behavior also occurs where it was not taught (new settings, people, materials and times), or that useful untaught forms of it appear, without the full training procedures being repeated there. B.14 covers the two kinds: stimulus generalization (the same response under new antecedent conditions) and response generalization (new, untrained forms that serve the same function).
The main lesson from Stokes and Baer (1977) is that generalization rarely takes care of itself. Teaching a skill and hoping it spreads, which they called “train and hope,” is the default, not a strategy. Generalization has to be planned from the start and measured.
Breaking it down
Start with the targets
Before teaching, list the generalization settings: every place, person, material and situation where the behavior should occur. List the response forms that will work there, and the reinforcement the behavior will meet. Then probe for generalization before, during and after teaching.
Teach enough of the range
- Sufficient stimulus exemplars (multiple exemplar training): teach with several examples, such as different dogs, different people giving the instruction or different bathrooms. Generalization to new examples often appears after a few have been taught.
- Sufficient response exemplars: teach several acceptable forms, such as “Hi,” “Hello” and a wave.
- General case analysis: choose teaching examples that sample the full range of stimulus and response variations the learner will meet, including negative examples that should not evoke the response. To teach vending machine use, include machines with different coin slots, buttons and displays, plus a change machine that looks similar but is not for buying.
Make teaching look like the real thing
- Programming common stimuli: put features of the generalization setting into teaching, such as the real materials, the same peers and the same kind of instructions, or teach in the setting itself.
- Teaching loosely: deliberately vary noncritical parts of teaching, such as wording, tone, position, room, background noise and time of day, so no single irrelevant feature gains control.
Get the behavior reinforced out there
- Teach to a level the natural contingencies will reinforce. A request that is too quiet or too slow may never be noticed, so accuracy and fluency both matter.
- Indiscriminable contingencies: make it hard to tell when reinforcement is available, using intermittent or delayed reinforcement, so responding continues at times and places where it is not reinforced right away.
- Behavior traps: use an easy, attractive lure into a natural community of reinforcement that then strengthens and maintains the behavior. Baer and Wolf (1970) described these natural communities of reinforcement.
- Ask people in the generalization setting to reinforce the behavior, or teach the learner to recruit reinforcement (“Can you check my work?”).
Mediate and train to generalize
- Mediation: give the learner something that goes everywhere they go, such as a picture card, a checklist, a self-instruction or a self-monitoring sheet (G.16).
- Train to generalize: reinforce generalization when it happens, tell the learner to use the skill in other places, or reinforce varied responding, for example with a lag schedule that reinforces responses different from the last few.
Stokes and Osnes (1989) later grouped tactics like these under three principles: take advantage of existing functional contingencies, train diversely, and build in functional mediators.
Unwanted generalization
Generalization can go too far. A child taught “dog” may call every animal a dog, or a learner may hug strangers as well as family. The fix is discrimination training that includes negative examples (G.6).
Seeing it in practice
| Learner and problem | Strategy |
|---|---|
| Jamal, 7, greets only his clinic therapist | Several staff, a parent and peers practice greetings with him (multiple exemplars). |
| Leila, 24, shops well in a mock store but freezes at her neighborhood market | Train with the market’s baskets, price tags and checkout layout, then practice at the market (common stimuli). |
| Grace, 10, finishes work only when the aide is watching | Thin and delay checks so she cannot tell which work periods count (indiscriminable contingencies). |
| Hiro, 8, plays alone at recess | Teach him a card game his classmates love, and the peer play that follows keeps his social skills going (behavior trap). |
| Mrs. Alvarez, 68, practices a home exercise routine at the clinic | A picture checklist on her fridge prompts each step at home (mediation). |
Common mix-ups
- Generalization vs. maintenance. Generalization is about new settings, people and forms. Maintenance (G.16) is about time. A skill can last for months at the clinic and never appear at school.
- Common stimuli vs. teaching loosely. Programming common stimuli copies specific features of the target setting. Teaching loosely varies irrelevant features so none of them takes control.
- Multiple exemplars vs. general case analysis. Both use many examples. General case analysis selects them systematically to cover the whole range and adds negative examples.
- Sequential modification is not generalization. Teaching the skill again in every setting where it is needed works, but nothing generalized.
Exam tip: Match the strategy to the problem. Control by one person or object calls for multiple exemplars or general case analysis. Control by one irrelevant feature calls for teaching loosely. Behavior that fails in the real setting, despite the skill being there, calls for common stimuli or for contact with natural reinforcement.
Quick recap
- Generalization must be planned and probed. “Train and hope” is not a strategy.
- Teach sufficient stimulus and response exemplars, using general case analysis to cover the range.
- Program common stimuli and teach loosely so teaching resembles the real setting without narrow control.
- Use indiscriminable contingencies, behavior traps, recruiting and natural reinforcers so the behavior keeps getting reinforced.
- Mediators and explicit training to generalize carry the skill across settings.
Further reading
- Stokes, T. F., & Baer, D. M. (1977). An implicit technology of generalization. Journal of Applied Behavior Analysis, 10(2), 349-367.
- Stokes, T. F., & Osnes, P. G. (1989). An operant pursuit of generalization. Behavior Therapy, 20(3), 337-355.
- Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied behavior analysis (3rd ed.), chapter 30.
Key terms
- Stimulus generalization
- The occurrence of a trained response in the presence of stimuli that were not part of training but share features with the training stimuli. The more a new stimulus resembles the training stimulus, the more likely the response.
- Response generalization
- The occurrence of untrained responses that serve the same function as a trained response. The new responses differ in form but belong to the same response class.
- Multiple exemplar training (Multiple exemplar instruction, MEI, Teaching sufficient exemplars)
- Teaching with several varied examples of the stimuli, settings or response forms involved in a skill, or rotating one stimulus across different response types, so that responding extends to examples that were never taught.
- General case analysis
- A systematic way of choosing teaching examples so they sample the full range of stimulus and response variations the learner will meet, including negative examples that should not evoke the response.
- Programming common stimuli
- A generalization strategy that includes typical features of the generalization setting, such as the same materials, people or kinds of instructions, in the teaching setting, so those features gain control during training.
- Teaching loosely (Train loosely, Teach loosely)
- Deliberately varying noncritical features of instruction, such as wording, tone, location, materials and time of day, so that no single irrelevant feature gains exclusive control over the response.
- Indiscriminable contingency
- A contingency arranged so the learner cannot tell which responses will be reinforced or when, usually through intermittent or delayed reinforcement. It helps behavior continue in settings and at times where reinforcement is not immediately available.
- Behavior trap
- A natural contingency of reinforcement that, once a person enters it with a small, easily reinforced response, keeps reinforcing and expanding the behavior without programmed help.
Check yourself
4 questions on G.15. Choose an answer to see whether you're right and why.
Show the answer
Answer: D. General case analysis picks teaching examples that cover the range of positive examples (dogs that look different) and adds negative examples that should not evoke the response (cats, horses). That narrows control to the features that matter. More trials with one photo keep control too narrow and do nothing about the overgeneralization.
Show the answer
Answer: B. Bringing features of the generalization setting (the trays, the table, the actual lunch partners) into teaching is programming common stimuli. Teaching loosely varies irrelevant features at random rather than copying one target setting. General case analysis samples the full range of examples instead of matching one setting.
Show the answer
Answer: D. Ben's answer is controlled by irrelevant features of the teaching situation. Deliberately varying those features (teaching loosely, along with using several people as exemplars) keeps any one of them from gaining exclusive control. Keeping trials identical is what produced the narrow control in the first place.
Show the answer
Answer: A. Teaching a learner to recruit reinforcement brings the natural contingencies at work into contact with his behavior, and the skill travels with him to any supervisor. Bringing the coach back restores the training contingency instead of transferring control, and retraining at the center does not change what happens at the job.
Adds a step to your study record. Saved in this browser only.