Errorless Learning in ABA

The short answer

In applied behaviour analysis, errorless learning is a prompting procedure. The instructor gives the instruction and the prompt at the same time, so the learner gets the response right on the first trial and on nearly every trial after it, and then the prompt is faded on a plan until the learner responds on their own. The name describes the goal, not a guarantee: the target is a very low error rate during acquisition, not zero.

It is used almost entirely for acquisition, meaning skills the learner does not have yet, and mostly inside discrete trial teaching, where the instruction, the response and the consequence are broken into short, repeated trials. It is not a philosophy of the whole program. The same program will use errorless procedures for new targets and give the learner far more room to be wrong on skills they already have.

If you want the general idea and the history first, the main errorless learning page covers that. This page is the practical version for people who are running or watching ABA sessions.

What a trial looks like

A single errorless trial in discrete trial teaching has five parts, and the thing that makes it errorless is that part three comes immediately, not after a pause.

Run that ten times with a full physical prompt and you have a learner who has touched the dog card ten times, been reinforced ten times, and made no errors. Nothing has been learned yet about telling the dog from the cat. That comes from fading, which is why fading is the actual procedure and the prompted trials are just the start of it.

The prompt hierarchy

Prompts are ranked from most intrusive to least. The exact list varies between programs, but the order is stable, because it is the order in which the prompts are faded.

Two things worth knowing. First, a physical prompt is not automatically the "strongest" for every learner: a child who dislikes being touched will resist a hand-over-hand prompt and the trial falls apart, and for that child a positional or stimulus prompt is the reliable one. Second, verbal prompts are the hardest to fade, because the instructor's voice is everywhere in the session already and the learner cannot always tell which part of it was the prompt. Programs that lean on verbal prompts for receptive targets tend to produce learners who wait to be told.

Fading: most-to-least and time delay

There are two standard ways to remove the prompt. Most programs pick one per target and write it into the plan.

Most-to-least prompting

Start at the most intrusive level the learner needs, and step down the hierarchy each time the learner meets a criterion, typically something like 80 or 90 percent correct across two consecutive sessions at the current level. Full physical, then partial physical, then gestural, then independent. The prompt still arrives immediately at every level; it just gets smaller.

This is the more conservative of the two and the more common with learners who have very little of the skill to begin with. Its weakness is that it can crawl: a learner who could have responded on their own at the second level spends a week being pointed at.

Time delay

The prompt stays at the same level but arrives later. In constant time delay, the first sessions use a zero-second delay (prompt with the instruction, exactly as above), then every trial after that uses a fixed wait, often three to five seconds, before the prompt comes. In progressive time delay, the wait grows in steps: zero seconds, then two, then four, then six.

The wait is the point. It opens a window in which the learner can respond before the prompt, and an independent correct response in that window is reinforced more richly than a prompted one. Time delay is the fading method with the best track record against prompt dependency for exactly that reason, and it transfers the response from the prompt to the instruction without the instructor having to judge which prompt level to drop to next.

Examples by program

The same procedure looks a bit different depending on the type of skill. These are the four it is most often used for.

Receptive identification

Instruction "touch dog" with three cards on the table. Errorless version: for the first block, the dog card is alone. Then the dog card plus a card that looks nothing like it, with a gestural prompt to the dog. Then the gesture is faded by time delay. Only when the learner is reliably independent with a very different distractor is a similar one added. This is a visual discrimination being built one step at a time, and the shadow matching game on this site runs the same progression: two obviously different choices first, then three, then four with a similar decoy.

Matching (match to sample)

The learner is handed a card and told "match". Errorless version: the array holds one card, the identical one, so the only possible response is correct. Then a second, very different card is added. A positional prompt (the correct card closest to the learner) is a natural fit here and fades cleanly by moving the card a little further each block.

Labelling (tacting)

The instructor holds up the dog picture and asks "what is it?" Errorless version: "What is it? Dog." with the answer given immediately as a full verbal model, and the learner echoes it. Fading is by partial verbal prompt ("D…") and then time delay. This is the program where prompt dependency is most likely, because the learner is being taught to say a word right after hearing it, and the difference between "echoing the prompt" and "labelling the picture" is invisible in a single trial. The data have to show independent responses at a delay before the target is counted as mastered.

Motor imitation and self-care

"Do this" with the instructor clapping, or a step of hand washing. Full physical prompts make sense here, faded most-to-least: hand over hand, then a touch that starts the movement, then a model only. Chained skills such as hand washing usually combine this with task analysis, prompting only the step the learner cannot do.

Errorless learning vs error correction

These are not competing approaches, and the confusion comes from the names. Errorless learning is what you do before the response so that errors are rare. Error correction is what you do after an error, and errors still happen in an errorless program, particularly during fading.

A common program design: errorless prompting for the first sessions of a new target, then time delay, and a fixed error correction procedure for any error that occurs in the delay window. What the procedure should not include is punishment, "no", or a long pause that reads as disapproval. A neutral "let's try again" with the prompt is the whole correction.

What to record

Errorless procedures live or die on data, because without it nobody can tell a learner who is progressing from one who is being prompted through the same level every day. The minimum per trial is the prompt level used and whether the response was correct, and the useful summary is the percentage of independent correct responses per session, not the percentage correct overall. A session of ten prompted correct trials is 100 percent correct and zero percent independent, and only the second number tells you anything.

For time delay programs, record responses in three columns: correct before the prompt, correct after the prompt, and error. The first column is what mastery is judged on.

Where it goes wrong

• Prompt dependency. The learner waits. Covered on the main page; in ABA the fix is a written fading plan, time delay, and differential reinforcement so that independent responses earn more than prompted ones.
• Prompting the wrong thing. A gestural prompt to the correct card teaches the learner to follow the point, not to look at the card. If independent responding collapses the moment the gesture is removed, the learner was discriminating the prompt, not the stimulus. Stimulus prompts (the correct item bigger or alone) avoid this because there is nothing else to attend to.
• Prompts that drift. A tired instructor at 4pm gives a slightly bigger prompt than the plan says. Across a week the level has crept up and the data no longer mean what they say. Supervisors check this by watching, not by reading the sheet.
• Never moving on. Errorless is an acquisition procedure. A learner who is now responding independently should be moved to maintenance and generalisation, with different materials, people and rooms, and be allowed to make and recover from errors there. Keeping a mastered target in errorless drills is time taken from something new.
• Using it for everything. Some skills are better learned by trying. A learner who can remember what they did and compare it with feedback often retains more from an effortful attempt than from being handed the answer. The comparison with trial and error on the main page sets out when each wins.

For parents watching a session

If you are a parent and you have found this page because a therapist said "we use errorless learning", here is what you should be able to see. In the first sessions on a new skill, your child is helped straight away and praised, and gets almost nothing wrong. Within a couple of weeks the help should visibly be getting smaller or arriving later. If you ask "how is the prompt being faded on this one?" the answer should be specific: a method, a criterion, and a date. If the answer is vague, or the help looks the same as it did a month ago, that is the question to push on.

The one thing you can do at home that helps most is not to prompt faster than the program does. If the plan is at a four-second delay, jumping in after one second with the answer undoes the week.

The games on this site run a softened version of the same idea: low difficulty first, small steps, a help button that prompts on request, and no punishment for a wrong tap. They do not fade prompts on a schedule or keep data, which is the part that belongs to the people who know the child.

What this page is: general information written by the person who builds these games, not by a behaviour analyst. Program decisions for a specific child belong with their BCBA and the team working with them.