Conditional discrimination

Another stimulus changes what is correct

Consider two comparison pictures, a bus and a train. When the sample is a bus ticket, choosing the bus picture produces feedback. When the sample is a train ticket, the train picture is correct. Neither comparison is always right; its function depends on the sample.

The sample can be visual, spoken, tactile, contextual, or another accessible form. Comparison responses can include selecting, sorting, saying, signing, typing, or using AAC.

Simple and conditional discriminations differ

In a simple discrimination, one stimulus has a relatively stable function across the relevant arrangement. A green sign might signal that entering is reinforced while a red sign signals waiting.

In a conditional discrimination, the function changes with another stimulus. If an arrow’s meaning depends on whether the current task is “find the exit” or “find the restroom,” that task cue provides a conditional context.

Matching-to-sample is a common arrangement

A matching-to-sample trial presents a sample and comparison array. The learner observes the sample, scans the options, and selects a comparison. Identity matching pairs identical stimuli. Arbitrary matching teaches relations that depend on history, such as a written word and a corresponding picture.

Sidman and Tailby distinguished conditional-discrimination performance from the additional tests needed to show stimulus equivalence. Accurate taught relations alone do not establish reflexivity, symmetry, and transitivity.

Attention and scanning can affect learning

Errors may reflect weak observation of the sample, incomplete scanning, inaccessible contrast, visual crowding, motor effort, or prompt history. Repeating more trials without addressing those variables can strengthen position bias or guessing.

A small study by Fisher and colleagues embedded an identity-matching task in a prompting hierarchy for two autistic children with slow progress in spoken-word-to-picture relations. It illustrates the value of assessing observing behavior while remaining too narrow for a universal procedure.

Prompt control can imitate learning

A teacher’s gaze, hand position, item location, voice change, or predictable array can signal the answer. Rotate comparison positions, balance stimulus presentation, and use probe trials that remove added cues.

Grow and colleagues evaluated methods for teaching receptive labeling and reported detailed procedural-integrity measures. Their work shows how position, observing responses, and prompt systems can affect conditional-discrimination teaching.

Trial definitions support valid data

Define the sample, comparison array, response window, correct relation, prompt, consequence, intertrial interval, and invalid-trial rules. State whether a correction begins a new trial and whether repeated instructions count.

Report performance by relation and sample, not only across the combined set. Overall accuracy can hide that one sample is mastered while another stays near chance.

Clinical selection requires meaningful goals

The BACB BCBA Test Content Outline, 6th edition includes simple and conditional discriminations, matching-to-sample, stimulus control, prompting, fading, and evaluation. It is examination content, not a person-specific protocol.

Choose relations that matter in daily life and use the person’s reliable communication and sensory access. Preserve AAC, breaks, mobility, essential care, and a way to pause or stop when applicable. Eye contact is never required to prove stimulus control.

A fictional conditional-discrimination record

Samira practices sorting event cards by a spoken sample: “today” or “later.” Six trials use each sample, with card position balanced. Samira responds correctly in 5 of 6 “today” trials and 5 of 6 “later” trials.

Overall accuracy is 10 of 12, or 83.3%. Sample-specific performance is 5/6 for each. The pattern shows accuracy in the defined session. It cannot prove the controlling variables, generalization, or an equivalence relation. The team schedules probes with new cards and another speaker.

Review errors by possible source

Track correct responses divided by eligible trials for each relation, prompt level, sample observation, comparison scanning, position choices, latency, integrity, and unprompted generalization probes. Keep correction and test trials separate.

When errors cluster by position, sample, speaker, or material, revise the teaching arrangement. A higher total trial count cannot repair inaccessible stimuli or unintended prompt control.

Generalization should test new examples without silently changing the relation. For a spoken-word-to-picture task, vary speakers, picture styles, and array positions while preserving the word-picture mapping. For an arbitrary relation, verify that the person learned the intended pair rather than a shared color, border, or location. Novel performance can support a broader claim only when probe items, training history, and scoring rules are reported clearly.

Document the probe sequence before reviewing its results.