Modeling Subtypes of Automatically Reinforced Self-Injurious Behavior with the Evolutionary Theory of Behavior Dynamics

The authors built computer models of self-injury that runs on its own reward.

They used a system called ETBD. It acts like digital creatures that learn by doing.

The goal was to see why some kids hurt themselves even when nothing outside pays off.

The models spit out three clear subtypes of automatic self-injury.

Each subtype matched a different pattern seen in real functional analyses.

Knowing the subtype lets you guess which treatment will flop before you even try it.

Cox et al. (2025) also let computers learn from behavior. They used Q-learning and hit a large share accuracy guessing the next move. Both papers show that feeding operant data into code beats old guess-work.

Rispoli et al. (2018) took the opposite road. They ran extra trial-based FA sessions until the automatic payoff for vocal scripting popped out. Their hands-on method works, but it takes sessions. Morris gives you a short-cut: run the model first, then test only the likely pattern.

Becraft et al. (2020) teach you to pool tiny single-case studies with math. Morris adds a new tool to that kit—simulation—so you can preview results before you ever touch a client.

You can open the free ETBD software, plug in your FA data, and get a likely subtype in minutes. Pick the treatment the model flags as strong, start there, and save weeks of trial and error.