Extinction Burst: A Practitioner's Guide to Predicting, Preventing, and Riding Out the Spike

The burst itself

An extinction burst is a temporary increase in the rate, magnitude, or duration of the target response immediately after reinforcement stops, before the behavior declines. It's the form most practitioners are taught to expect by default, but the evidence does not support treating it as universal. Reported prevalence across the literature clusters in the roughly 24 to 39 percent range, meaning most learners placed on extinction do not show a classic escalation before improvement. When a burst does occur, it tends to be more pronounced after a rich, continuous, or recently dense reinforcement history (Lambert et al., 2024), and less pronounced when the schedule has already been thinned toward intermittent (MacDonald et al., 2013), or when the maintaining function was confirmed through a functional analysis rather than assumed from descriptive data.

Extinction-induced response variability

Extinction can produce new, sometimes more elaborate response topographies that weren't part of the learner's typical repertoire for that function, not just more of the original behavior. This form isn't purely a hazard. Several demonstrations have used extinction-induced variability deliberately, capturing and reinforcing a novel, more appropriate topography, an emerging word approximation, an autoclitic frame, a communicative gesture, as it appears during the variability window, effectively using extinction as a controlled route into shaping a better form (Cengher et al., 2020). Anticipate this form as an opportunity to differentially reinforce a better response, not only a side effect to wait out.

Extinction-induced aggression and emotional responding

Extinction can also produce aggression, emotional responding such as crying or agitation, or other behavior with real safety implications, independent of whether a classic rate burst occurs. This form is distinct from extinction-induced response variability above: it doesn't yield a more elaborate or communicative topography, just heightened distress or aggression that a plan needs to anticipate on its own terms, not treat as a milder version of the same thing.

Spontaneous recovery

Spontaneous recovery is the reappearance of the extinguished behavior after a period with no exposure to the original discriminative context, typically after a delay such as a weekend or a break between sessions, even though extinction appeared complete at the end of the prior session. It's an expected pattern, not a sign that extinction failed or that reinforcement is being delivered inadvertently, and a brief re-extinction exposure at the start of the next session is generally sufficient to bring the rate back down, usually faster than the original extinction curve took.

Resurgence and renewal

Resurgence is the return of the target behavior when a previously reinforced alternative response is placed on extinction or its reinforcement is thinned, a relapse of the original problem behavior driven by the alternative response's own reinforcement history rather than by a return of the original contingency. Renewal is the related but distinct phenomenon where the target behavior reappears when the learner moves to a context different from the one where extinction and differential-reinforcement training occurred, such as a return from a clinic room to a classroom, even with no change in the contingency itself. Both are documented risks for any plan that pairs extinction with a differential-reinforcement alternative, and each traces to a different point of failure: resurgence traces to the alternative response's own reinforcement schedule, renewal traces to the training context itself.

Extinction is appropriate as a component when a functional analysis has confirmed the specific reinforcer maintaining the behavior and that reinforcer can actually be withheld with fidelity across every relevant person and setting. It is contraindicated, or at minimum needs heavy modification, when the function is automatic, meaning the reinforcement is produced by the response itself rather than delivered by another person, since there is no external contingency to remove. Extinction of automatically reinforced behavior generally requires sensory extinction (Maag et al., 1986), response blocking, or protective equipment layered in, not withholding attention or escape that was never the maintaining variable to begin with. It is also contraindicated on its own, without a safety plan, for any behavior with an aggression or self-injury history severe enough that a burst or extinction-induced aggression would create real risk before the behavior comes down, and the literature offers no reliable way to predict in advance which learners will show response variability, which will show aggression, which will show both, and which will show neither, which is exactly why that contraindication holds regardless of what you expect for a given case.

Do not run extinction alone as the plan. The evidence for extinction paired with a reinforcement component, DRA, functional communication training, or noncontingent reinforcement, is stronger and more consistent than for extinction run in isolation, and combining them is also the more humane path: a rich, competing source of reinforcement for the alternative response gives the learner something to do instead while the original contingency is removed, rather than simply removing access and waiting. Reach for extinction alone only when the case genuinely cannot support building an alternative response in the available time and immediate risk demands rapid suppression, and treat that as the exception that still needs its own safety contingency, not as the default plan.

To blunt rather than ride out a burst, three tactics have direct support, and the choice between them follows from the case in front of you. Thin a continuous or recently dense reinforcement schedule toward intermittent before starting extinction when the history shows dense, continuous delivery, since continuously reinforced behavior is harder to extinguish and more prone to a larger burst (MacDonald et al., 2013; Lambert et al., 2024). Choose noncontingent reinforcement on a fixed or variable time schedule when the confirmed function is attention or another deliverable stimulus, since it has outperformed plain extinction directly for attention-maintained behavior in head-to-head comparisons (O'Reilly et al., 1999), and a brief competing-stimulus assessment can identify a tangible item to use the same way when noncontingent attention isn't the deliverable in play (Fisher et al., 2004). And choose a signaled, multiple-schedule structure when the plan pairs extinction with an alternative response that needs to stay clearly discriminable from the extinguished one, rather than landing as an unsignaled, blanket change in contingency.

Whatever combination you choose, a safety contingency for aggression and emotional responding belongs in the plan before the first session, independent of whether you expect a classic burst. That genuine uncertainty is the reason an extinction-only plan, without a layered reinforcement component and a named safety contingency, is the wrong call more often than the field's shorthand teaching suggests.

Extinction burst is the concept where honesty with a caregiver matters most, and the data genuinely disagree on how often the textbook escalation shows up. Treat a possible burst as a contingency you plan for, not a guarantee you promise, and put your planning effort into execution rather than prediction.

Before the first extinction session, write the safety contingency into the plan itself, not into a mental note for if things go sideways. Name the specific behaviors that trigger a pause or a shift to a protective procedure, name who on the team is authorized to make that call mid-session, and confirm every implementer, not just the supervising BCBA, has read it. Do this whether or not you expect a classic burst, since extinction-induced response variability and extinction-induced aggression can each show up without any preceding escalation in rate.

Once you've decided to pair extinction with a reinforcement component, the first-session mechanics matter as much as the choice itself. If you're layering in noncontingent reinforcement, start the fixed or variable time schedule dense enough that the learner contacts it before the extinguished response would typically occur, then thin it gradually across sessions rather than starting lean and adjusting up after a burst appears (O'Reilly et al., 1999). If you're building a signaled multiple schedule, make the discriminative stimulus obvious and consistent from session one, a specific card, light, or phrase, not a subtle cue only the clinician notices, so the learner can tell from day one when the alternative response will and won't pay off.

Check the reinforcement history before you flip the switch. A client who has just had a dense, recent run of the reinforcer maintaining the behavior is a worse candidate for starting extinction today than one whose recent reinforcement has been leaner (Lambert et al., 2024). If the schedule has been continuous, run a brief thinning phase toward intermittent first rather than starting extinction cold and bracing for a bigger burst (MacDonald et al., 2013).

Don't assume extinction only touches the target behavior. One classroom demonstration found that ignoring a single target response had genuine side effects elsewhere in the day: some desirable behaviors dropped along with it, and some other unwanted behaviors increased, changes a plan tracking only the target response would miss entirely (Sajwaj et al., 1972). Add a short list of the learner's other notable behaviors, both ones you want to hold steady and ones you'd rather not see rise, to your data sheet before the first extinction session, not after you notice something has shifted.

Build the two relapse mitigations into the schedule from day one rather than reacting after the fact. Start the context fade in week one or two of treatment, not at discharge: swap one element of the treatment setting at a time, room color, then lighting, then which staff member runs the session, holding each swap until responding is stable before moving to the next. And run at least some of the alternative-response training in the room, with the people, and on the schedule the learner will actually face day to day, rather than perfecting it in a single clinic room with one trained clinician and hoping it holds up elsewhere.