An investigation of the matrix training approach to teach social play skills

Wilson and colleagues set up a matrix of play actions. Each cell paired one toy with one action.

They taught only the diagonal cells—toy-action pairs that matched. Kids with autism practiced these trained pairs until they reached mastery.

No one taught the other 86 percent of the matrix. The team watched to see if those untrained pairs appeared on their own.

After mastering the diagonal, each child produced most of the untrained play combinations. The paper reports 86 percent of the off-diagonal cells showed up without direct teaching.

Kids also started asking peers to join them, even though no one had trained that request.

Haring (1985) got similar free learning. They taught play sets with stimulus-equivalence drills and saw kids use new toy sets. Both studies show you can pay for a few lessons and get many more for free.

Watson et al. (2007) used video modeling instead of a matrix. Their children generalized play only when the new toys looked like the trained ones. The matrix method in Wilson et al. (2017) gave broader free combos because it crossed two separate parts—toy and action—rather than relying on looks.

Lambert et al. (2016) chained full basketball moves step by step. They taught every link in the sequence. Wilson skipped most cells and still got the moves, showing a denser return for teaching time.

You can cut direct teaching time by two-thirds and still get most play combinations plus spontaneous peer bids. Next time you run play sessions, lay out a simple matrix—toys on one side, actions on the top—teach the diagonal, then watch the grid fill itself in.