ABA Multiple-Exemplar Training Worksheet

An ABA multiple exemplar training worksheet should define the meaningful target, the dimensions that may vary, the features that must stay relevant, and which examples are reserved for later probes. It should keep teaching trials, novel-exemplar probes, nonexample checks, accessibility, implementation fidelity, and invalid events separate so varied practice is not mistaken for demonstrated generalization.

Use this ABA multiple exemplar training worksheet for one individualized teaching plan and review. It does not determine how many examples are sufficient or make multiple-exemplar teaching appropriate for every person or skill. A qualified team must select the target, examples, nonexamples, teaching arrangement, supports, safeguards, probes, and decision rules from current assessment and direct client input.

Clinicians & ABA Professionals / Assessment and Treatment Planning.

Define the purpose and response before choosing examples

Begin with what the person wants the skill to make possible. Operationally define the response and accepted accessible forms. Identify the stimulus features or contextual relations that should control responding, along with irrelevant features that can vary without changing the correct response.

Plan identityComplete before implementationClient-selected purpose and observable targetRelevant dimensions or relationsIrrelevant dimensions planned to varyAccepted spoken, signed, gestural, AAC, written, or motor responsesOrdinary communication, sensory, motor, medical, and environmental supportsConsent, assent, dissent, pause, withdrawal, and stop processCurrent assessment and interdisciplinary recommendationsPlan author, qualified approver, effective date, and review date

The BACB Sixth Edition BCBA Test Content Outline identifies stimulus and response generalization, discrimination, multiple-exemplar training, natural contingencies, contextual fit, procedural integrity, and data-based modification as distinct competencies. It is examination content, not a recipe or endorsement of this worksheet. The BACB Test Content Outlines hub identifies the current outline.

Build an exemplar map instead of a miscellaneous list

Each exemplar should have a reason for inclusion. Mark which relevant feature it represents and which irrelevant features vary. Avoid a set in which one accidental feature predicts every answer. List nonexamples that help test the intended boundary, not merely items that look obviously different.

Exemplar IDTeaching, withheld probe, or nonexampleRelevant feature or relationIrrelevant features variedExpected responseAccessibility or cultural noteWhy included

The review of multiple-exemplar instruction strengths and limitations describes variation across examples as a strategy for programming generalization while noting that the relevant literature varies in procedures and reporting. That supports an explicit matrix and cautious interpretation; it does not supply a universal number, sequence, or diversity rule.

Protect withheld probes and exposure history

Name probe exemplars before teaching starts. Record any prior exposure, accidental teaching, modeling, correction, or feedback. Once a withheld item has been taught or its correct response disclosed, it no longer answers the same novel-exemplar question.

Probe exemplarWithheld from teaching sinceKnown earlier exposureNovel dimensionProbe cue and response windowFeedback or teaching allowed?Exposure breach and actionNo / defineNo / defineNo / define

An applied study of multiple-exemplar training and sharing responses evaluated responding across trained and untrained materials under its own participant, task, and teaching conditions. A more recent study of multiple-exemplar instruction and emergent responses likewise used explicitly arranged teaching and probe conditions. These studies illustrate why trained and untrained stimuli must be labeled; they do not show that any one exemplar set will produce generalization for another person or target.

Specify the teaching contract and access protections

Define the natural or programmed cue, response window, prompt and correction rules, consequence, intertrial boundary, randomization method, and exposure limit. A change in any of these may require a new version. Preserve ordinary access and do not treat an AAC device, glasses, hearing support, mobility aid, sensory accommodation, interpreter, or medically necessary support as a removable prompt.

Teaching elementCurrent approved definitionCue, context, and motivating conditionResponse window and observable initial responseApproved prompt or correction and timingConsequence and feedback rulesExemplar selection, rotation, and repetition limitsTrial separation and stop conditionsValid, interrupted, unavailable, and invalid rulesGeneralization and maintenance checks kept outside acquisition totals

ASHA's Augmentative and Alternative Communication Practice Portal says AAC users should retain access to their tools or devices at all times. Teaching and probe arrangements must preserve that access. A plan can name a particular instructional prompt, but removing the person's communication system cannot create a valid independence measure.

Run a readiness check before each block

Confirm the active version and whether the person is willing and ready to participate.
Confirm access to communication, sensory, mobility, medical, and environmental supports.
Check that the selected exemplar has the correct status: teaching, withheld probe, or nonexample.
Check that accidental exposure or material differences have not changed the trial question.
Verify who may teach, probe, observe, pause, and revise the plan.
Hold when distress, illness, withdrawal, missing access, safety, privacy, or scope conditions make the event unsuitable.

An invalid setup is implementation evidence, not an incorrect learner response. Preserve the reason and determine whether the event can be rescheduled under the current version.

Record acquisition and novel probes row by row

Capture the initial response before any prompt or correction. Keep the final teaching outcome in a separate field. For a no-feedback probe, do not add teaching merely to complete the row; end it under the approved probe rule and return to teaching only after the probe block is closed.

Date/timeVersionExemplar and statusRelevant/varied dimensionsInitial responsePrompt, correction, or feedbackFinal responseValid?Access, assent/dissent, or exposure noteTeaching / Novel probeIndependent correct / incorrect / no response / unavailableCorrect / incorrect / no response / not applicableYes / No, reason

Do not pool prompted teaching outcomes into independent probe accuracy. Teaching data describe performance with trained examples under the active arrangement. Novel-probe data ask a narrower question about the particular withheld examples and conditions sampled.

Reconcile teaching and novel-exemplar results separately

The fictional example uses a labeled-storage-location routine. Twenty-four acquisition or probe trials are scheduled. Two are invalid: one exemplar was inadvertently exposed during teaching and one trial used the wrong label set. The 22 valid events comprise 14 teaching trials and eight novel-exemplar probes.

Trial typeScheduled or reconciledInvalidValidIndependently correctOther valid resultTeaching exemplars14 valid after reconciliationIncluded in combined invalid count14113 prompted, corrected, incorrect, or no responseWithheld novel-exemplar probes8 valid after reconciliationIncluded in combined invalid count853 incorrect or no responseCombined acquisition/probe schedule24222Do not poolDo not pool

Valid acquisition/probe coverage is 22 / 24 × 100 = 91.7%.
Independent teaching performance is 11 / 14 × 100 = 78.6%.
Independent novel-exemplar probe performance is 5 / 8 × 100 = 62.5%.

The 91.7% figure is schedule coverage, not skill performance. Trained performance is not generalization: the 78.6% figure describes only the examples used in instruction. The 62.5% figure describes eight valid withheld probes under their documented conditions. None proves causation, mastery, broad generalization, maintenance, benefit, assent, or safety.

Stokes and Baer described generalization as a technology requiring active programming and measurement. That framework supports documenting conditions and testing outcomes beyond teaching. It does not convert variation during acquisition into evidence that behavior generalized.

Keep nonexample discrimination checks in their own denominator

Nonexamples ask whether responding remains appropriately bounded when a relevant feature or relation is absent. They should be selected with care so the task is meaningful, accessible, and not a guessing trap.

The fictional plan schedules six nonexample checks. One is invalid because an unintended visual cue reveals the answer. Five valid checks remain, and four contain the defined correct rejection.

The reconciled record is short but explicit:

Scheduled nonexample checks: 6.
Invalid nonexample checks: 1.
Valid nonexample checks: 5.
Correct rejections: 4.
Incorrect acceptance or no response: 1.

Valid nonexample coverage is 5 / 6 × 100 = 83.3%.
Correct rejection is 4 / 5 × 100 = 80.0%.

Do not average 80.0% correct rejection with teaching or novel-probe performance. A correct rejection and an independently correct positive example answer different discrimination questions.

Score implementation fidelity apart from learner performance

Select components that define the approved procedure. The fictional fidelity review includes 21 observable events across five components: correct exemplar/status, planned cue and response window, approved prompting or no-feedback rule, planned consequence, and honored access or stop conditions. This creates 21 × 5 = 105 component opportunities. Ninety-six are implemented as written and nine are mismatches.

Implementation componentEvents observedImplemented as writtenMismatchesPattern or next actionCorrect exemplar and teaching/probe status21192Cue and full response window21201Prompt, correction, or no-feedback rule21183Planned consequence or feedback21192Access, pause, and stop conditions21201Reconciled total105969Review component rows

Component implementation is 96 / 105 × 100 = 91.4%. It describes implementer behavior, not learner skill. Twenty-one scorable fidelity events also differ from 22 valid acquisition/probe trials and five valid nonexample checks. Do not force these denominators to match or use a global score to hide repeated errors in one component.

Review balance, burden, and unwanted stimulus control

Review questionEvidence and datesWhat the evidence supportsMissing or conflicting informationOwner and next actionDoes the target remain meaningful and client-informed?Do exemplars represent relevant variation without stereotyping?Has an irrelevant feature become a shortcut or bias?Were probes truly withheld and accessible?What assent, dissent, fatigue, distress, or withdrawal occurred?Are repetition, error exposure, and session burden acceptable?Is additional assessment, another teaching strategy, or referral indicated?

Variation can add burden or reduce clarity when examples change too quickly, contain inaccessible materials, rely on culturally narrow assumptions, or combine several new dimensions at once. Record those effects. Do not increase variety merely to meet a count.

Separate generalization, maintenance, and acquisition decisions

Create one dated line for each review type that applies:

Acquisition review: condition or exemplar ; previously taught yes; teaching available ; valid opportunities ; independent responses ; result and limitation ___.
Novel-exemplar probe: condition ; prior exposure ; teaching available ; valid opportunities ; independent responses ; result and limitation .
Setting, person, or activity probe: changed dimension ; prior exposure ; teaching available ; valid opportunities ; independent responses ; result and limitation .
Maintenance check: later date ; prior teaching ; teaching available ; valid opportunities ; independent responses ; result and limitation .

A later maintenance check needs a stated time boundary. A setting, person, or activity probe should identify exactly what differs from training. Results support only the sampled conditions, response definitions, dates, and plan version.

Document revisions, exceptions, and qualified decisions

For every decision trigger that applies, note the supporting evidence, alternatives, pause or stop boundary, responsible reviewer, and date:

Continue the current exemplar set: ___.
Hold and collect missing exposure, fidelity, or access evidence: ___.
Revise examples, nonexamples, prompts, or consequences: ___.
Add a separately defined probe or maintenance check: ___.
Stop or refer: ___.

For covered certificants, the live BACB Ethics Codes hub and Ethics Code for Behavior Analysts address areas such as competence, dignity, client participation, applicable consent and assent, risk, written programs, accurate records, and continuing evaluation. Neither source approves this artifact or the exemplar choices in a particular case. CASP's ABA Practice Guidelines access page identifies Version 3.0 and its licensing conditions. No licensed guideline language is reproduced here, and compliance is not claimed.

Use an exception log for exposure breaches, mislabeled examples, missing AAC or other supports, interruptions, health changes, distress, withdrawal, unexpected responding, unwanted stimulus control, context imbalance, and procedure mismatches. Preserve the active version on each row and document the evidence, decision, rationale, owner, and effective date of any change.

Use this worksheet only as a companion to the current assessment and approved treatment plan. The controlling record still includes the applicable consent and assent process, safety plan, authorization, law, payer requirements, and interdisciplinary recommendations. This planning aid is neither a prescribed curriculum nor a validated generalization test, automated mastery engine, consent record, or replacement for qualified oversight.

ABA Multiple-Exemplar Teaching Matrix and Generalization Review Worksheet