Stimulus discrimination

Differential consequences establish the relation

Suppose a medication label says “refrigerate” or “room temperature.” Sorting into the correct storage location receives feedback. Over time, different label features may control different sorting responses.

The relevant feature must be defined. If every refrigerated item also has a blue border, color can become an accidental shortcut. Later performance may fail when label design changes.

Simple and conditional discriminations differ

In a simple discrimination, one stimulus has a relatively stable function within the arrangement. A “closed” sign might signal waiting while “open” signals entry.

In a conditional discrimination, another stimulus changes which option is correct. A storage icon may require one response for medication and another for food packaging. The relation among sample, comparisons, and consequences determines accuracy.

Errors can reflect access or the procedure

Visual crowding, hearing differences, language, motor effort, fatigue, position bias, prompts, or unfamiliar materials can affect performance. More trials cannot repair a stimulus the person cannot distinguish.

Fisher and colleagues embedded identity matching within a prompting hierarchy for two autistic children who had slow progress in a spoken-word-to-picture task. The study illustrates assessment of observing behavior in a small sample.

Grow and colleagues compared receptive-labeling procedures and reported integrity in detail. The work shows why position, prompts, and observing responses deserve measurement.

A fictional discrimination record

Rosa practices sorting six medication labels: three require refrigeration and three do not. Across 12 balanced independent trials, each label appears twice and left-right location varies.

Rosa responds correctly in 11 of 12 trials: 6 of 6 refrigeration trials and 5 of 6 room-temperature trials. Overall accuracy is 91.7%, but condition-specific counts reveal where the error occurred. These results do not prove the intended text controlled sorting or that the skill will transfer to a pharmacy or home.

Teach the relevant feature

Begin with clearly discriminable examples and prompt the feature that matters. Vary irrelevant dimensions such as position, color, size, and speaker. Fade added prompts and use independent probes.

If errors cluster around one feature, revise the teaching set. If performance is accurate only with one instructor, test other partners. Generalization and discrimination work together: the response should extend across harmless variation while remaining different when a meaningful feature changes.

Avoid harmful overdiscrimination

A narrow teaching history can produce responding only to one exact example. A person may identify one exit sign but miss another approved design. Training should sample the range expected in daily life.

Conversely, broad generalization can be unsafe when differences matter. Testing should include positive examples and close nonexamples so the boundary becomes clear without creating distress.

Preserve communication and choice

Use the person’s reliable speech, sign, gesture, writing, touch, or AAC response. Eye contact is never required. Allow breaks, corrections, and a way to decline participation.

Food, water, bathroom access, mobility, AAC, prescribed care, rest, and emergency help remain available regardless of discrimination performance. Choose consequences that fit the goal and avoid turning every preferred event into payment.

Report trials by stimulus and relation

For each stimulus, report correct independent responses divided by eligible trials. Include array size, position, prompt level, response latency, feedback, correction, invalid trials, and integrity.

Keep teaching, correction, maintenance, and novel probes separate. After feedback on a novel item, label later presentations as teaching or exposed probes.

Lattal reviews contingency beyond simple event sequence. The BACB BCBA Test Content Outline, 6th edition includes discrimination, stimulus control, and prompting in examination content, not a clinical protocol.

Mastery needs balanced stimulus classes

An overall criterion can hide unequal exposure. If one positive example appears ten times and each nonexample appears once, 90% accuracy may say little about the rare items. Set a minimum number of independent trials for every relevant stimulus or relation.

Use confusion matrices when choices matter. Record which stimulus was presented and which response occurred, not only correct or incorrect. That detail shows whether two labels are confused in both directions or whether one response is selected too often.

Novel probes should sample the practical boundary. Include clear new examples and close nonexamples, then report them separately. After a correction, move the item into the exposed set.

Review false positives and false negatives by consequence. In a safety discrimination, missing a warning and pausing unnecessarily may carry different risks, so one combined accuracy percentage can be misleading.