Conditional Discrimination in ABA: How Context Teaches Learners to Choose

Conditional discrimination in ABA is a teaching arrangement in which the correct response depends on a contextual cue, so the same sample stimulus calls for different selections under different conditions. A learner might match a picture of a dog to the word “dog” when the instruction is “match,” but sort it into a category when the cue is “sort.” The behavior is under conditional control: the sample alone is insufficient, and a second stimulus (the conditional or contextual cue) determines which comparison is reinforced.

The three-term contingency inside a conditional discrimination

A simple discrimination has one antecedent, one behavior, and one consequence. A conditional discrimination inserts a second antecedent that changes the reinforcement contingency. In a match-to-sample trial, the sample stimulus (conditional stimulus), the comparison array, the selection response, and the consequence form a four-term contingency. The sample signals which comparison will pay off; without it, every comparison is equally eligible.

Consider a learner with two stimuli, a red card and a blue card, and two comparison options. On red-sample trials, choosing the red comparison is reinforced. On blue-sample trials, choosing blue is reinforced. This is identity matching, the simplest conditional discrimination. Arbitrary matching is harder: the sample and comparison share no physical resemblance, so the relation must be learned. A learner matching a spoken word to a printed word is solving an arbitrary conditional discrimination, which is why the arrangement underpins early reading and receptive language programs.

The mechanism is stimulus control by a conditional relation, not by the sample alone. Behavior analysts describe the sample as setting the occasion for reinforcement of one comparison over another. When the learner responds accurately to novel combinations of samples and comparisons, the relation has become abstract: it generalizes across the specific stimuli used in training. This is the behavioral account of what educators call “understanding the rule.”

What the research base shows about difficulty and mastery

Conditional discrimination tasks vary widely in difficulty, and the literature converges on several patterns. Identity matching is typically acquired faster than arbitrary matching, and arbitrary matching often requires systematic prompting and error correction. Discrimination reversals, in which the reinforced comparison switches, are a standard probe of flexibility. A learner who continues selecting the previously correct comparison after the switch shows perseveration, a clinically meaningful pattern.

Mastery criteria in applied settings are usually expressed as a percentage of correct responses across a session, with common thresholds at 80 to 90 percent over consecutive sessions. These thresholds are conventions, not natural boundaries. A learner at 85 percent across three sessions has met a local criterion; whether the skill is truly established depends on generalization probes with untrained stimuli and untrained instructors.

The counterintuitive detail: high accuracy on trained conditional discriminations does not guarantee the learner is attending to the conditional cue at all. A learner can reach criterion by attending to the comparison array alone if the array composition is fixed. Researchers call this a failure of conditional control, and it is why stimulus control checks matter. Varying the comparison array across trials is the standard safeguard.

Edge cases that change the teaching plan

Several edge cases deserve explicit planning. Learners with limited attending skills may need the conditional cue made more salient before conditional discrimination training begins. Stimulus overselectivity, in which a learner responds to one feature of a complex stimulus and ignores others, can produce apparent conditional discrimination that collapses when the irrelevant feature changes. A learner may also show position bias, selecting the left comparison regardless of the sample, which masks any conditional control.

Error patterns carry diagnostic information. In a conditional discrimination with three comparisons, chance performance is roughly 33 percent. A learner consistently at chance across many sessions is not learning the relation and needs a different teaching arrangement. A learner above chance but below criterion may be learning slowly, which calls for more trials, better prompts, or a simpler stimulus set.

Prompt dependence is a common edge case in applied work. If a prompt is present on every trial, the learner may attend to the prompt rather than the sample. Fading prompts systematically and probing unprompted performance is the standard remedy. A trial that never tests the sample-comparison relation cannot confirm it.

Misconceptions that distort practice

The most persistent misconception is that conditional discrimination is the same as simple discrimination. It is not. In simple discrimination, one stimulus is consistently correct. In conditional discrimination, correctness depends on the sample, so the reinforcement contingency is relational.

A second misconception treats matching tasks as a single skill. Identity matching, arbitrary matching, and symbolic matching are different relations with different teaching requirements. A learner who matches identical pictures may still fail to match a spoken word to a picture, because the latter requires an arbitrary relation.

A third misconception assumes that mastery on trained pairs transfers automatically. Generalization to novel stimuli is a separate behavior that often needs programming. The professional caveat: many published demonstrations of conditional discrimination use tightly controlled stimulus sets, and applied gains can be smaller when teaching occurs in noisy classrooms with competing contingencies.

Practical actions for building conditional discriminations

Start with the simplest arrangement the learner can solve. Use two comparisons before three, and identity matching before arbitrary matching. Keep the sample and comparisons well separated in space and present them simultaneously unless the protocol calls for delayed matching.

Verify stimulus control before advancing: vary the comparison array and confirm accuracy holds. Set a mastery criterion in advance, such as 80 to 90 percent over two or three consecutive sessions. Probe unprompted performance regularly, so prompt dependence does not masquerade as mastery. Run generalization probes with novel stimuli and novel instructors before declaring the skill established. Watch for position bias and stimulus overselectivity; adjust the array if either appears. Keep session length short enough that attention does not degrade, and record error patterns rather than only totals.

Red flags to address immediately: accuracy at chance across many sessions, a sudden drop after a comparison switch, correct responding only when the prompt is present, and correct responding that disappears when the array changes. Each of these signals a stimulus control problem, not a motivation problem.

Limitations of the conditional discrimination framework

The framework describes reinforcement contingencies, not the learner’s internal representation. Two learners can produce identical accuracy with different underlying stimulus control, and the behavior-analytic account cannot distinguish them from accuracy data alone. Trial-by-trial analysis, including error patterns and latency, is needed to make inferences about control.

The evidence base for many applied conditional discrimination procedures is built on single-case designs with small samples. Those designs are rigorous for demonstrating functional relations within a learner, but they do not establish population-level effect sizes. Reviews of matching-to-sample instruction in applied settings report positive outcomes, yet the generality of specific prompt-fading sequences across populations remains an empirical question. This article describes mechanisms and teaching logic; it does not prescribe treatment for any individual, and decisions about a learner’s program belong with a qualified behavior analyst who has assessed that learner directly.

A final limitation is context. Conditional discrimination taught in a clinic may not transfer to home or school unless transfer is programmed. The conditional cue itself may be absent in the natural environment, which is why practitioners often teach the relation across multiple cues and settings rather than one.

The practical takeaway is that conditional discrimination is a relation to be built and verified, not a task to be completed. Accuracy on the trained set is the beginning of the evidence, and stimulus control checks, generalization probes, and maintenance data complete it. A learner who selects correctly only under the training conditions has learned the conditions, not the relation.