Revaluing overselected stimuli: Effects of degree of posttraining extinction on stimulus overselectivity

Gomes‐Ng et al. (2025) asked: if you stop rewarding the one cue a learner always picks, how much extinction is enough to make other cues matter?

They taught 12 college students to click one rich picture that paid 90 % of the time and ignore two lean pictures. Then they ran three extinction doses: 0, 30, or 90 non-reinforced trials on the rich cue while the lean cues stayed available.

The team tracked how soon each learner started clicking the formerly ignored pictures.

More extinction of the overselected picture meant faster and bigger jumps in clicking the other pictures.

After 90 extinction trials, most learners chose the two lean pictures almost equally; after 30 trials the shift was smaller; with zero extinction the rich picture still ruled.

The change looked smooth, not all-or-none, supporting a graded comparator process.

Shvarts et al. (2020) also played with extinction in a single-case lab, but they kept reward signals around to cut resurgence. Their effect was tiny and jumpy; Gomes‐Ng shows a clean dose–response, suggesting removing the signal entirely works better than leaving it on.

van Laarhoven et al. (2003) speed up stimulus control by drilling multiple examples across trials. Gomes‐Ng adds a post-training move: after acquisition, you can rebalance control simply by starving the overselected cue, no extra teaching needed.

Kang et al. (2011) warn that assessment format changes behavior rates; here, the format stays the same—only the payoff schedule changes—so the effect is cleaner and easier to interpret.

If a client with autism always picks the red card, you now have a lever: stop reinforcing red for a while and watch other cards gain value. Start with 30 extinction trials, then probe; if control is still lopsided, push to 90. The lab says the shift will show up fast, usually within one session.