Programming for stimulus and response generalization

A skill that shows up only at one table, with one instructor, in one exact form is not much use to the learner. Since Stokes and Baer (1977), behavior analysts have treated generalization as something you program on purpose, not something you hope for.

Why it matters

Behavior change matters only if it shows up where and when it counts: at home, at school, at work, and in the community. One of the dimensions of applied behavior analysis is generality, which means the change lasts over time, appears in settings beyond training, or spreads to related behaviors.

"Train and hope" is what happens when nobody plans: you teach in one setting and later check whether the skill traveled. Often it hasn't. Practice choosing the best fix for a generalization failure and naming the tactic a plan uses.

Key ideas

• Know the two settings. The instructional setting is where teaching happens. The generalization setting is anywhere else the behavior should occur without direct teaching.
• Know the two kinds of generalization. Stimulus generalization is the same response with new stimuli; response generalization is new, equivalent responses (see Definitions).
• Plan before you teach. List the behaviors that need to generalize and every setting, person, and situation where they should occur. Pick targets the everyday environment is likely to reinforce. Then choose tactics. Cooper, Heron, and Heward sort the tactics into five broad strategies:

1. Teach the full range of what the learner will meet. Use multiple exemplar training: teach several stimulus examples (different shoes for "tie your shoe") and several response examples (different ways to greet). General case analysis goes further: your teaching examples sample the full range of relevant stimulus and response variation the learner will meet. Include negative teaching examples too, so the learner learns when not to respond.
2. Make teaching look like the real world. Program common stimuli so the teaching and generalization settings share important features, usually by bringing materials, people, or furniture from the generalization setting into teaching. Teach loosely by randomly varying noncritical parts of instruction (location, noise, instructor, time of day) so no single irrelevant stimulus gains control.
3. Help the behavior contact reinforcement out in the world. Teach the behavior to a level (rate, accuracy, quality) that the natural contingencies will actually reinforce. Ask people in the generalization setting to reinforce it, teach the learner to recruit reinforcement, and use behavior traps. Use indiscriminable contingencies (intermittent schedules, delayed reinforcement) so the learner can't tell which responses will pay off.
4. Mediate generalization. Contrive a mediating stimulus that travels with the learner, or teach self-management so the learner prompts and reinforces their own behavior.
5. Train to generalize. Reinforce response variability or novel responses, and tell the learner directly to use the skill elsewhere.

• Probe and measure. Probe for generalization before, during, and after teaching. If it doesn't occur, use sequential modification: teach directly in each setting where the behavior failed to appear.

Definitions

• Generalization setting: any place or stimulus situation, different from the teaching setting, where the target behavior should occur.
• Stimulus generalization: the trained response occurs with untrained stimuli, such as new people, places, or materials.
• Response generalization: untrained responses that serve the same function as the trained one occur.
• Multiple exemplar training: teaching with several varied stimulus and response examples to promote generalization.
• General case analysis: choosing teaching examples that represent the full range of stimulus and response variations in the natural environment.
• Programming common stimuli: including typical features of the generalization setting in the teaching setting.
• Teaching loosely: randomly varying noncritical aspects of instruction.
• Behavior trap: a contingency that is easy to enter and, once entered, keeps behavior going with naturally occurring reinforcement.
• Indiscriminable contingencies: reinforcement arrangements the learner cannot predict, so responding continues when reinforcement is not delivered.
• Sequential modification: teaching the behavior directly in each setting where it did not occur.

Examples from a session

Non-example (train and hope): Sam, 9, masters greeting with one therapist at one clinic table. Nobody plans or probes for other people or places, and his parents later report he never greets relatives.

Common mistakes / exam traps

• Waiting until mastery to think about generalization. Planning starts before teaching does.
• Confusing teaching loosely with general case analysis. One varies irrelevant features at random; the other systematically samples relevant variation.
• Calling sequential modification "generalization." If you taught it directly in the new setting, it didn't generalize there.
• Ignoring undesired generalization. Overgeneralization (calling every man "Daddy") and unwanted response forms are problems too. Negative teaching examples help.
• Thinning reinforcement too fast. Indiscriminable contingencies work best after the behavior is well established.

Quick check

Related topics

• Stimulus vs. response generalization (B.14)
• Recognizing response maintenance (B.15)
• Maintaining change after intervention: schedule thinning and natural reinforcers (G.16)
• Teaching simple and conditional discriminations (G.6)
• Fading prompts: prompt delay and stimulus fading (G.8)
• The dimensions of applied behavior analysis (A.5)