Empirically derived consequences: a data-based method for prescribing treatments for destructive behavior.

The team worked with two clients who had hard-to-treat destructive behavior. Their first functional assessments gave muddy results, so the authors built a quick test to find what each person actually wanted and what they really disliked.

In just a few short sessions they let the clients earn or lose items and activities while they watched the behavior numbers. The items that cut problem behavior the most became the reinforcers. The items that suppressed it fastest became the punishers.

When the chosen reinforcers and punishers were used together, destructive behavior dropped sharply for both clients. The change was fast and stayed low while the team watched.

No extra extinction or fancy package was needed. The simple mix of the right reward plus the right mild punisher did the heavy lifting.

Allan et al. (1994) ran the same method on pica and got the same big drop, showing the trick works across topographies. Schmidt et al. (2021) updated the idea 27 years later by using a five-condition brief experimental analysis to pick NCR or DRO instead of reinforcer-plus-punisher, proving the data-first mindset still holds.

Owen et al. (2020) looks like a contradiction at first—they got 97% drops without any punisher at all. The key difference is they ran a mand analysis first and found the behavior was really pre-current to getting adults to comply. Once they reinforced mands on a lean schedule, the punisher piece became unnecessary.

Matson et al. (1989) cataloged hundreds of punishment studies and showed the field was already moving toward milder, data-driven choices. The current paper is a direct child of that shift: it keeps the consequence idea but lets brief data pick the least intrusive mix.

When your FA is unclear you don’t have to guess. Run a five-minute reinforcement assessment, find what works and what suppresses, then pair them. You get fast drops in destruction without jumping to heavy extinction or trial-and-error DRA. Try it next time the standard FA gives you static.