Factors that encourage generalization from extinction to test reduce resurgence of an extinguished operant response
Briefly withholding or freely giving the new response’s reinforcer during later tests cuts resurgence of the old problem behavior in half.
01Research in Context
What this study did
Trask et al. (2018) worked with lab rats. The rats first learned to press one lever for food. The team then taught the animals to press a second lever instead. Next they stopped all rewards to see if the first lever would come back.
During the final test the researchers either (a) turned off rewards for the new lever too, or (b) gave that food for free. They counted how often the old, problem lever returned.
What they found
Both tricks cut the return of the first lever by about half. Letting the new lever go unrewarded some of the time worked. Giving its food for free also worked.
The rats kept the new, good lever press and the old, bad one stayed low.
How this fits with other research
Shahan et al. (2026) now go further. They show you can wipe out resurgence almost to zero by using short off periods (one minute off, one minute on). Trask’s 50 % drop is good, but the newer paper gives a stronger, faster fix.
Nist et al. (2023) seem to disagree. They gave rats five new levers instead of one and saw no benefit. The lesson: it is not how many new responses you offer, but how you schedule their extinction that counts.
Perez et al. (2015) extend the idea to people. Adults with developmental disabilities had less problem behavior return when staff rotated two good responses instead of sticking with one. The same logic—mess with the new response schedule—works across species.
Why it matters
If you run DRA or FCT and later thin reinforcement, plan for the old behavior to pop back. You can buffer this by (1) briefly stopping rewards for the new skill during some checks, or (2) giving that reinforcer free a few times. Both are easy adds to your next session. Start small: insert two 30-s extinction probes into the new response and watch if the old one stays quiet.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →During your next DRA session, run two short 30-second probes where the new communicative response earns nothing and note if problem behavior stays low.
02At a glance
03Original abstract
Two experiments investigated methods that reduce the resurgence of an extinguished behavior (R1) that occurs when reinforcement for an alternative behavior (R2) is discontinued. In Experiment 1, R1 was first trained and then extinguished while R2 was reinforced during a 5-or 25-session treatment phase. For half the rats, sessions in which R2 was reinforced alternated with sessions in which R2 was extinguished. Controls received the same number of treatment sessions, but R2 was never extinguished. When reinforcement for R2 was discontinued, R1 resurged in the controls. However, the alternating groups showed reduced resurgence, and the magnitude of the resurgences observed during their R2 extinction sessions decreased systematically over Phase 2. In Experiment 2, R1 was first reinforced with one outcome (O1). The rats then had two types of double-alternating treatment sessions. In one type, R1 was extinguished and R2 produced O2. In the other, R1 was unavailable and R2 produced O3. R1 resurgence was weakened when O2, but not O3, was delivered freely during testing. Together, the results suggest that methods that encourage generalization between R1 extinction and resurgence testing weaken the resurgence effect. They are not consistent with an account of resurgence proposed by Shahan and Craig (2017).
Journal of the Experimental Analysis of Behavior, 2018 · doi:10.1002/jeab.446