A systematic review of the matrix training literature

Kemmerer et al. (2021) read every matrix-training experiment they could find. They located 40 studies published between 1970 and 2020.

The team recorded how each study built its word grids, how many cells were taught, and how trainers checked for new combos. No stats were pooled; they simply mapped the range of methods.

No two studies used the same recipe. Matrix size, teaching order, mastery criterion, and probe format all differed.

Because of this mix, you cannot tell which setup works best or for whom. The authors call for a standard playbook.

Frampton et al. (2016) is inside the review. That single-case study taught only diagonal pairs and still saw kids with autism create new noun-verb phrases. The review shows this lean approach is common, yet no study compares diagonal-only to full-row teaching.

Podlesnik et al. (2023) mapped another corner of behavior science—resurgence procedures—and likewise found wild method variation. Both reviews echo the same warning: when labs tweak too many knobs, we cannot build a clear evidence base.

Nesselrode et al. (2022) surveyed functional-analysis formats in schools and saw a similar drift: most teams still run full analog FAs even though brief and trial-based versions save time. Across these reviews, the field keeps inventing new twists before nailing the basics.

If you plan a matrix next week, pick one recipe and stick to it. Teach the same number of diagonal pairs, use the same mastery rule, and probe the same untrained cells that prior papers used. Document what happens and share the data. Small acts of consistency will speed up the day we know which matrix format truly speeds up generative language.