The big idea
Teaching a discrimination means arranging reinforcement so a response occurs in the presence of one stimulus and not another. In a simple discrimination, one stimulus is the SD and another is the S-delta: responding to the SD pays off, and responding to the S-delta does not. In a conditional discrimination, which stimulus is correct changes from trial to trial, depending on another stimulus called the sample.
Much of what early learners are taught is conditional: pointing to the cup when you hear “cup,” but to the shoe when you hear “shoe.” If teaching is accidentally set up as a simple discrimination, the learner can look successful while learning the wrong thing.
Breaking it down
Simple discriminations
A simple discrimination is a three-term contingency: SD, response, reinforcer (see B.12). It comes in two arrangements.
- Simultaneous: SD and S-delta are present together, and the learner chooses. A learner picks up the red block whenever red and blue blocks are on the table.
- Successive: one stimulus at a time; respond to one, withhold to the other. A child answers when their own name is called and not when other names are called.
Conditional discriminations
A conditional discrimination adds a fourth term. The sample (the conditional stimulus) decides which of several comparisons is the SD on that trial. The same picture can be correct on one trial and wrong on the next.
Matching-to-sample (MTS) is the standard procedure. Present a sample, then an array of comparisons, and reinforce selecting the comparison that goes with the sample.
- Identity matching: the sample and the correct comparison are the same (match a picture of a cat to an identical picture).
- Arbitrary matching: they are related only by teaching (hear “cat” and select the cat picture; see the printed word and select the picture).
Listener responding, reading comprehension and many academic skills are conditional discriminations. They are also the building blocks for the emergent relations in G.19.
Discrimination training
Discrimination training is differential reinforcement across stimuli: reinforce responses in the presence of the SD (or to the correct comparison) and withhold reinforcement otherwise. Errors are often followed by a correction procedure that re-presents the trial with a prompt. Design choices that matter:
- Use arrays of three or more comparisons. With two, a learner who guesses is right half the time.
- Rotate targets from the start. Teaching “cup” mixed with other targets makes the spoken sample, not repetition, the cue for which item is correct.
- Make every stimulus correct on some trials and incorrect on others. If the cup is always correct, the learner may learn “touch the cup” without listening.
- Add a differential observing response when samples look or sound alike. Have the learner name or touch the sample before the comparisons appear, so you know they attended to it.
Errorless learning
Errorless learning arranges teaching so the learner rarely errs. You might introduce the S-delta faint, small or absent and bring it in gradually, or prompt the correct response immediately and then fade the prompt (see G.7 and G.8). Terrace’s work with pigeons showed that discriminations learned this way produced far fewer errors and less emotional behavior than trial and error. The cost is planning: prompts and fading steps must be designed carefully, or learning stalls on the prompt.
Counterbalancing and unwanted control
Learners will follow any reliable cue, including the wrong one. Common sources of unwanted control:
- Position bias: always choosing the same side, or the nearest item.
- Stimulus bias: favoring one comparison regardless of the sample.
- Instructor cues: glancing at the correct card, adjusting it or placing it last.
- Trial-order patterns: for example, repeating whatever was reinforced on the last trial.
Counterbalance by placing the correct comparison in each position equally often, in an unpredictable order. Keep a neutral face and place all comparisons at once. When data look odd, sort errors by position and by comparison; the pattern usually shows what is really controlling responding.
Evaluating the procedure
Judge accuracy against chance: with three comparisons, about 33 percent correct means no discrimination at all. Look for accuracy across all positions and all targets, then probe with new arrangements, new materials and a different instructor before calling a skill mastered.
Seeing it in practice
| Skill | Type | Sample | Watch for |
|---|---|---|---|
| Responding to own name | Simple, successive | None | Responding to any voice |
| Finding red among blocks | Simple, simultaneous | None | Choosing by size or position |
| “Touch cup” with cup, shoe, ball | Conditional, arbitrary | Spoken word | Position bias, favorite picture |
| Matching coins | Conditional, identity | A coin | Always choosing the largest coin |
| Matching printed words to pictures | Conditional, arbitrary | Printed word | Guessing from the first letter |
Dalia, age 5, scores 90 percent on “touch cup” for three sessions. Her supervisor notices the cup was always on the right. When positions are rotated and other targets are mixed in, accuracy drops to 30 percent. The earlier data showed position control, not listener responding.
Common mix-ups
- Receptive identification is not a simple discrimination. If the same picture can be right or wrong depending on what was said, it is conditional.
- High accuracy is not proof of the right control. Check positions and rotate samples.
- An S-delta is not a punisher. It signals that reinforcement is unavailable; nothing is added.
- Errorless learning still needs fading. Without it, the learner learns to follow the prompt.
Exam tip: Ask whether the correct choice changes depending on another stimulus. If it does, it is a conditional discrimination, and the design must rotate samples and counterbalance positions so the sample controls responding.
Quick recap
- Simple discrimination: one SD, one S-delta. Conditional: the sample decides which comparison is correct.
- Matching-to-sample can be identity or arbitrary.
- Use three or more comparisons, rotate targets, counterbalance positions and avoid instructor cues.
- Errorless methods prevent errors with prompts or by bringing in the S-delta gradually.
- Judge accuracy against chance and probe for unwanted sources of control.
Further reading
- Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied behavior analysis (3rd ed.), chapter 17.
- Grow, L., & LeBlanc, L. (2013). Teaching receptive language skills: Recommendations for instructors. Behavior Analysis in Practice, 6(1), 56-75.
- Terrace, H. S. (1963). Discrimination learning with and without “errors.” Journal of the Experimental Analysis of Behavior, 6(1), 1-27.
Key terms
- Simple discrimination
- A discrimination in which a response is reinforced in the presence of one stimulus (the SD) and not in the presence of another (the S-delta), so responding comes to occur mainly when the SD is present. It is a three-term contingency.
- Conditional discrimination
- A discrimination in which the correct choice depends on another stimulus, the sample (or conditional stimulus). The same comparison stimulus is correct on some trials and incorrect on others, so the relation is a four-term contingency.
- Matching-to-sample (MTS)
- A procedure for teaching and testing conditional discriminations. A sample stimulus is presented, then two or more comparison stimuli, and choosing the comparison that goes with the sample is reinforced. In identity matching the correct comparison is the same as the sample; in arbitrary matching the relation is set by teaching.
- Discrimination training
- Reinforcing a response in the presence of one stimulus, or one sample-comparison relation, and withholding reinforcement in the presence of others, until responding occurs reliably only under the intended stimulus conditions.
- S-delta (SΔ, Stimulus delta)
- A stimulus in whose presence a particular behavior has not been reinforced. Because of that history, the behavior is less likely when the S-delta is present. It signals that reinforcement is unavailable; it does not signal punishment.
- Position bias (Side bias)
- Faulty stimulus control in which a learner's choices are controlled by where a stimulus is placed, such as always the left side, rather than by the stimulus itself or the sample. Counterbalancing the position of the correct choice helps prevent it.
- Errorless learning (Errorless teaching)
- An approach to teaching that arranges prompts or stimulus changes so the learner makes few or no errors, then gradually removes that help. Examples include most-to-least prompting, time delay that starts at 0 seconds and introducing the S-delta faintly before bringing it to full strength.
Check yourself
4 questions on G.6. Choose an answer to see whether you're right and why.
Show the answer
Answer: C. Because the dog is always correct, the learner can succeed by choosing the dog picture without listening, which is a simple discrimination. The spoken sample has never been needed to pick the right card. Rotating which picture is requested makes each picture correct on some trials and wrong on others. Positions were already rotated, so position bias does not explain the data.
Show the answer
Answer: C. Which block is correct depends on the spoken sample, so the same block is right on one trial and wrong on the next: a conditional discrimination. Always choosing red is a simple simultaneous discrimination. Clapping to 'clap' and crossing on the walk signal are simple successive discriminations, with one SD and one S-delta.
Show the answer
Answer: A. Omari's responding is controlled by position, and his accuracy is at chance. Prompting the correct card wherever it falls, with reinforcement only for correct selections, builds control by the sample instead. A two-card array raises chance to 50 percent without teaching anything, and keeping the correct card on the left reinforces the bias.
Show the answer
Answer: A. Errorless learning arranges the early steps so the learner rarely errs, for example by introducing the S-delta faintly and strengthening it gradually. Full-strength trial and error allows many errors, which is what triggers Ines's tantrums. A firm 'no' after errors adds a consequence but does nothing to prevent them.
Adds a step to your study record. Saved in this browser only.