Generalization

Generalization extends beyond the teaching example

Learning can become tied to the exact person, wording, object, or room used during instruction. A generalization probe samples a predeclared, untrained stimulus, situation, or response form without teaching it during that probe. Once the team directly teaches the response under that condition, later performance shows learning there; it is not an untrained generalization probe.

Common forms include:

• Stimulus generalization across examples: After learning to identify one picture of a bus, a person identifies unfamiliar bus pictures and real buses without direct teaching of those examples.
• Stimulus generalization across people or settings: A person uses an accessible help message with a parent at home after learning it with a clinician at the center, if that person and setting were not directly taught for the response.
• Response generalization: After learning one help request, a person uses another untrained speech, sign, gesture, writing, or AAC form that serves the same function.

The variation should be relevant to the goal. Using a fork at two different tables may matter for daily participation. Performing an arbitrary clinic task in many rooms may add little value.

Generalization, maintenance, and independence answer different questions

These terms can describe the same skill from different angles:

A response can generalize while still using ordinary supports. Glasses, visual schedules, AAC, mobility devices, written instructions, and environmental accommodations can be part of successful participation. Removing an access tool simply to label performance “independent” can make the measure less meaningful.

How teams plan for generalization

The 1977 paper “An Implicit Technology of Generalization” organized earlier research under strategies such as teaching sufficient examples, varying nonessential features, programming shared cues, and connecting performance with naturally occurring consequences. It argued that teams should actively assess and program generalization. This historical review predates modern autism practice guidance and uses outdated disability language, so its framework requires current clinical and ethical judgment.

A practical plan identifies:

1. the response in observable, flexible terms
2. the untrained people, places, materials, examples, or instructions that matter
3. which access supports remain available
4. which conditions will be taught and which will be reserved for probes
5. how assent, dissent, safety, privacy, and family burden will be protected
6. what the team will change if transfer does not occur

The BACB BCBA Test Content Outline (6th ed.) separately lists stimulus and response generalization, response maintenance, procedures to promote both kinds of generalization, and procedures to maintain desired behavior change following intervention. It identifies entry-level professional knowledge and skills; it does not decide which variation matters to a particular person.

A fictional measurement example

Leila is a fictional nine-year-old who uses speech and a tablet-based AAC system. She chooses a goal of requesting help with lunch containers in everyday settings. With her technician, she independently communicates during 9 of 10 teaching trials using one container type.

Before probing, the team defines six planned opportunities distributed across two unfamiliar container types, two familiar partners who did not conduct teaching, and two everyday settings not used for teaching. An opportunity is eligible only when the planned materials are present, Leila has her usual communication access, the setting is safe and accessible, and Leila assents. Each probe has a brief agreed response interval; Leila may communicate in any established mode or decline, and the adult provides needed help after the interval regardless of the score.

All six opportunities are eligible. Leila independently communicates in 4 of 6, receives a prompt in one, and does not use the defined response in one. AAC remains available in all six and is used in three of the four independent responses. The report states 4 independent responses in 6 eligible generalization probes, or 67% when rounded, rather than “67% mastery.” It lists each sampled variation, access support, prompt rule, date range, and response form.

The result supports a narrow conclusion about those conditions, not every container, person, or setting. In a later six-probe set, if adults fail to provide predeclared communication access in one planned probe, report adult access implementation as 5 of 6, or 83%, and calculate learner responding only across the five eligible probes. Record declined probes and safety or access barriers separately under predeclared rules, never as skill errors.

A useful generalization metric needs a defined denominator

“The skill generalized 80%” leaves major questions unanswered. A quality measure should state:

• Numerator: eligible probe opportunities with the defined response at the stated support level
• Denominator: all eligible probe opportunities sampled during the reporting window
• Window: exact dates or treatment period
• Variations: people, settings, materials, instructions, and response forms included
• Exclusions: predeclared reasons an opportunity was invalid, reported separately
• Purpose: clinical decision, family discussion, quality review, or another defined use

Keep teaching trials separate from probes. Report prompted and independent responses separately. Include partner behavior when the outcome depends on it, such as the percentage of recognizable communication attempts that adults honor. A high aggregate percentage can hide a weak setting, missing AAC access, or repeated partner failure, so teams should review the underlying conditions.

Generalization should serve the person

Useful transfer expands access, communication, safety, choice, relationships, or participation. It should not become a demand to perform identically with every person or in every environment. A change in context can reveal sensory barriers, unclear instructions, different expectations, inaccessible materials, or a reasonable preference.

The current BACB Ethics Code for Behavior Analysts applies to BCBA and BCaBA certificants and applicants. It addresses appropriate client and stakeholder involvement, required informed consent and applicable assent, assessment and intervention design, client and stakeholder preferences, risk minimization, correct data collection and use, and continual evaluation. The public CASP Version 3.0 summary describes guidance for planning, implementing, and evaluating ABA assessment and treatment for autism. Full guideline access requires a licensing agreement, and this page does not attribute detailed procedures to the licensed text. CASP's public resources page provides additional organizational and practice resources.

Questions to ask about generalization

• Which variations matter to the person and family?
• Were those conditions taught, or were they reserved as probes?
• Did the person have the same communication and accessibility supports?
• Are prompted, independent, and partner-supported responses separated?
• What exactly is the numerator, denominator, reporting window, and exclusion rule?
• Could weak performance reflect an environment or access problem?
• How will the team respond to assent, dissent, or a changed preference?
• When will maintenance be checked after generalization is observed?