Stimulus generalization

The stimulus changes while the response remains defined

A learner may sort a taught recycling bin correctly and later use the same sorting response with a new approved bin design. The bin changes; the response class remains the same.

If the learner invents a new way to communicate the sorting decision, that variation may also involve response generalization. Report both when they occur together.

Novel stimuli must be identified before testing

A material is not novel after prompting, correction, rehearsal, or feedback. List training examples and probe examples in advance. Record exposure outside the formal program when known.

The first clean probe offers the clearest evidence of untrained transfer. Later trials can show maintenance or learning after exposure, but they should carry a different label.

Relevant dimensions define useful transfer

Generalization should follow the feature that matters. Recycling symbols and accepted material types may be relevant; bin color or location may vary. Training only green bins risks narrow control by color.

Osnes and Lieblein analyze generalization through stimulus control, emphasizing the need to identify what should govern responding.

Discrimination sets the boundary

Responding to every similar stimulus can be unsafe. A medicine label, pedestrian signal, or allergy warning requires sensitivity to important differences.

Include close nonexamples where errors matter. The goal is a useful range: broad enough for ordinary variation and narrow enough to preserve meaningful distinctions.

A fictional stimulus-generalization probe

Ben learns to place recyclable containers into two training bins with different approved symbols. Six predeclared novel bins vary in color, size, setting, and symbol layout while retaining the local recycling standard.

Without prompts or feedback, Ben sorts correctly on 5 of 6 first-exposure probes. The 83.3% result describes that novel set. It does not establish transfer to every bin, city, material, or disposal rule. The error remains visible and informs the next training example.

Generalization can be programmed

Stokes and Baer organized tactics such as training sufficient exemplars, programming common stimuli, contacting natural contingencies, training loosely, and teaching people to generalize. The framework replaces passive “train and hope” with explicit planning.

Select tactics from the setting. Multiple exemplars may address variation in materials; a shared checklist may bridge environments; partner training may make consequences more consistent. Packages can combine tactics, so outcome changes rarely isolate one component.

Published examples show uneven transfer

Marzullo-Kerth and colleagues taught sharing with several material categories to four autistic children using video modeling, prompting, reinforcement, and multiple exemplars. All showed within-category generalization, while one showed across-category generalization.

The small package study shows why clinicians should measure the exact type of transfer instead of assuming that success with some novel examples will extend everywhere.

Preserve supports that belong in the setting

Generalization does not require removing AAC, interpreters, visual schedules, mobility tools, sensory supports, or ordinary assistance. If a support will be available in daily life, include it in the generalization plan.

Ask whether the new settings and goals matter to the person. Performance across more contexts is not automatically better when it increases burden or overrides a preference.

Measure by dimension and context

Report correct independent responses divided by eligible novel opportunities. Identify each new person, material, setting, or wording change. Record prompts, feedback, prior exposure, errors, latency, integrity, and the person’s view.

Keep training, maintenance, stimulus-generalization, and response-generalization data separate. The BACB BCBA Test Content Outline, 6th edition includes programming for generalization and maintenance in examination content, not a universal clinical procedure.

Map generality across several dimensions

A skill may generalize across materials while remaining tied to one person or setting. Build a matrix with the dimensions that matter: people, locations, objects, instructions, time, modality, and ordinary support. Mark which combinations have been trained, probed, or remain unknown.

Avoid testing every possible combination when a smaller, representative sample answers the practical question. General case analysis can identify likely extremes and meaningful variations. Document the sampling logic so a reader knows what the claim covers.

Report breadth alongside accuracy. Five correct responses across five novel settings show a different pattern from five correct responses with one new material in one room. Neither is inherently better; the goal determines which dimension matters.

Recheck after environmental changes. A new device interface, staff routine, sign standard, or accessibility barrier can alter performance without erasing the underlying skill.