A parametric analysis of errors of commission during discrete-trial training.

The team watched therapists run discrete-trial lessons with kids with autism. They asked: What happens if the adult makes the wrong correction half the time?

They set up three conditions. In one, every correction was right. In the second, half were wrong. In the third, every correction was wrong. Then they counted how fast each child learned new skills.

Perfect corrections won. Kids learned fastest when adults never made a mistake.

Half-wrong and all-wrong corrections tied for last. Most kids learned just as poorly with 50% errors as with 100% errors. A few lucky kids did slightly better at 50%, but not much.

Zhi et al. (2024) extends this idea. They showed that simply giving the right correction—without extra praise—was enough for listener responses. Together the papers say: get the correction right, then stop; frills don’t help.

van Abswoude et al. (2015) seems to disagree. They let kids with cerebral palsy make lots of motor errors and saw no harm. The clash disappears when you look at the skill type. Discrete-trial language tasks break under error, but gross-motor swimming strokes don’t.

Geiger et al. (2018) backs the target from the staff side. Live or computer-based BST both pushed therapists toward the zero-error sweet spot the target says we need.

Your error rate is a dial you can turn today. Run a quick self-check: count the next ten corrections you give. If even one is off, fix it before the next trial. Use a data sheet with a simple “+” for right correction and “–” for wrong. Aim for ten “+” in a row. Your learner’s speed of new skills depends on it.