Join Free →
About Weekly CEU Resource Hubs ↳ CEU Library ↳ Research Library Blog Contact Join Free →
Topic Guide · Practitioner

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

Query target: extinction burst · BBC Editorial Team
★ Summary

An extinction burst is a temporary increase in the frequency, duration, intensity, or topographical variability of a previously reinforced response when that response no longer contacts its maintaining reinforcer Fisher et al. (2023). It is documented in basic and applied research, but it is not universal — modern clinical and laboratory data place burst prevalence between roughly 24% and 39% of cases under common conditions, and burst probability and magnitude vary with reinforcement schedule, reinforcer value, baseline density, and the specific measurement window the team uses Muething et al. (2024) Nist & Shahan (2021) Woods & Borrero (2019). The practical job for a BCBA, RBT, or school behavior team is therefore not to "expect a burst every time" or to abandon extinction at the first spike — it is to predict whether this case is high-risk for a burst, pre-position competing reinforcement so the burst stays small if it occurs, document continuously enough to distinguish a burst from treatment failure, and pre-decide the safety threshold at which extinction stops being appropriate.

01What the Research Says

What an extinction burst actually is in 2026

The clinical shorthand "things get worse before they get better" maps onto four documented topographies of change when a previously reinforced response is placed on extinction: (1) a transient increase in rate, (2) an increase in magnitude or force of individual responses, (3) an increase in response duration, and (4) extinction-induced variability — novel response forms appearing inside the same operant class Fisher et al. (2023) Alessandri & Lattal (2021). Fisher and colleagues' integrative review of basic and applied work treats the burst as a brief, early spike in target responding that rapidly declines as extinction proceeds, and proposes the temporally weighted matching law (TWML) as the formal mechanism — historical reinforcement value of the target response temporarily outweighs the now-absent reinforcer, then declines with continued extinction exposure Fisher et al. (2023) Shahan (2022). Alessandri and Lattal's force-tracking work in 22 adults extends the topography point: when the operative dimension is response force rather than rate, the early-extinction signature still appears — humans push harder, not (or not only) more often, in the first minutes after reinforcement is withdrawn Alessandri & Lattal (2021). Practically, this means a "no burst" call based on rate alone can still miss a real escalation if the team isn't also tracking intensity and duration.

How often bursts actually happen — be evidence-honest

The honest read of the modern literature is that bursts are common but not universal, and the reported prevalence is heavily a function of how the team measures responding Muething et al. (2024) Nist & Shahan (2021) Katz & Lattal (2021). Muething and colleagues' retrospective record review of 108 outpatient ABA cases treated with extinction observed bursts in 24% of cases, with magnitude that declined over sessions but probability that did not meaningfully diminish across repeated baseline-to-treatment transitions; no reliable links emerged between burst occurrence and topography, function, or specific intervention package elements Muething et al. (2024). Woods and Borrero's review of pediatric food-refusal cases treated with escape extinction found bursts in 30% of children (3/10), with bursts that were briefer and less severe than typically reported for general problem behavior and that did not compromise treatment efficacy Woods & Borrero (2019). Older clinical work reanalyzed by Lerman and Iwata had reported burst rates of 35.7–62% when extinction was used alone, but only 12–15% when extinction was combined with another treatment such as differential reinforcement — a pattern that argues for never running extinction alone in clinical settings if a competing reinforcement source is feasible Nist & Shahan (2021) Saini et al. (2016).

The basic-research picture is just as nuanced. Lattal, Kuroda, and Cook ran three experiments with pigeons and rats placed on extinction after variable-ratio or variable-interval baselines and found little evidence of bursts in the first 20 minutes of extinction following intermittent reinforcement — questioning the assumption that bursts are ubiquitous after intermittent schedules Lattal et al. (2020). Katz and Lattal's pigeon work across 5–7 repeated reinforcement/extinction cycles documented bursts on the first extinction exposure but found that magnitude and probability shifted unpredictably across successive cycles and animals — the prototypical "fixed" burst after every reinforcement interruption is not what the data show Katz & Lattal (2020). Nist and Shahan formally demonstrated that prevalence is measurement-dependent: when the analysis window was restricted to the first minute of extinction, burst prevalence rose above 75% in rats; when reinforcer-consumption time was included in baseline rate calculations, prevalence inflated further Nist & Shahan (2021). Shahan's TWML paper — and Katz and Lattal's conceptual analysis of transitional behavior — both argue that a single uniform definition of "burst" is inadequate, and that whether one is observed depends on baseline reinforcement parameters, how extinction is introduced, and the temporal resolution of the data Shahan (2022) Katz & Lattal (2021).

The takeaway is not that bursts aren't real — they reliably occur for a subset of cases, often early, sometimes large, and Cox's 2025 wild-bird study demonstrated bursts in genuinely free-ranging organisms outside any laboratory or clinic Cox (2025). The takeaway is that quoting "things always get worse before they get better" to a parent or teacher is not what 2026 evidence supports Muething et al. (2024). A defensible practitioner statement is closer to: "in roughly 1 in 4 cases under extinction-with-other-treatment, and more often when we use extinction alone, problem behavior briefly increases in the first minutes or sessions before declining" Muething et al. (2024) Nist & Shahan (2021).

What makes a burst more likely or larger

The corpus is consistent on the parametric variables that intensify bursts when they occur. Lambert and colleagues' randomized 2 × 2 factorial trial across 69 adults with developmental disabilities showed that higher reinforcer value relative to demand intensity magnifies the burst during subsequent extinction — pre-extinction reinforcement density is a tunable lever Lambert et al. (2024). Shahan and Avellaneda's 12-rat crossover comparing extinction after 1-pellet versus 6-pellet baseline reinforcement found that burst magnitude scales with the previously earned reinforcement magnitude — roughly proportional doubling at the peak Shahan & Avellaneda (2025). Fisher and colleagues' three-subject reversal study of destructive behavior in children showed that richer baseline reinforcement rates intensified resurgence-style bursts during extinction even when an alternative communication response was concurrently reinforced at matched density Fisher et al. (2019). Shahan, Browning, and Nall's six-rat parametric study added another lever: longer exposure to alternative reinforcement before the alternative is also placed on extinction produces larger post-treatment bursts, because the added reinforcement increases the behavioral mass that has to be disrupted Shahan et al. (2020). The pattern across these studies is operationally simple: dense baseline schedules, large reinforcer magnitudes, and long histories of richly reinforced alternatives all loaded the burst when the contingency eventually broke.

Schedule type matters less cleanly than schedule density. The classical clinical lore — "intermittent reinforcement schedules cause bigger bursts" — is partially supported (Lerman and colleagues' early work and Saini and colleagues' multiple-schedule review note bursts as a recurring side effect of multiple-schedule thinning into extinction components Saini et al. (2016)) but partially complicated. Lattal and colleagues found "little evidence of bursts" in the first 20 minutes of extinction following VR/VI baselines, suggesting that intermittent reinforcement does not automatically guarantee a visible early burst — at least not in the lab under those parameters Lattal et al. (2020). The practitioner-relevant version: a long history of dense reinforcement loads the burst more reliably than schedule type per se, and an abrupt cutover from a rich schedule to extinction is the configuration most likely to produce a visible spike Shahan & Avellaneda (2025) Lambert et al. (2024).

Mitigation: the burst is a tunable variable, not a fixed cost

The literature offers four documented and clinically translatable levers that each meaningfully attenuate burst magnitude. First, alternative reinforcement during extinction is the largest lever the basic literature has tested: Shahan and Avellaneda's six-rat within-session reversal showed that increasing the magnitude of concurrent alternative reinforcement progressively reduced burst responding, with high-magnitude (6-pellet) alternative reinforcement virtually eliminating the burst Shahan & Avellaneda (2025). The applied analogue is the well-documented advantage of pairing extinction with differential reinforcement of alternative behavior (DRA) or functional communication training (FCT) rather than running extinction alone — which is why Lerman and colleagues' 12–15% burst rate under extinction-plus-DR is so much lower than the 35.7–62% under extinction-alone Nist & Shahan (2021) Saini et al. (2016). Brown and colleagues' two-child outpatient comparison of DRA-with-extinction and DRA-without-extinction further shows that, when the alternative is reinforced richly, the post-treatment "burst-like" resurgence stays at a low, similar level whether or not extinction is layered on top — meaning the alternative-reinforcement source does most of the burst-mitigation work Brown et al. (2020). Saini, Fisher, and Pisman's three-child reversal study makes the same point from the other direction: noncontingent reinforcement (NCR) without extinction produced a larger later burst when reinforcement was eventually withheld than NCR-with-extinction, indicating that combining alternative reinforcement with the contingency change is more durable than relying on alternative reinforcement alone Saini et al. (2017).

Second, discriminative-stimulus signaling of the extinction context can virtually eliminate burst responding in some configurations. Fisher and colleagues' four-child reversal study showed that presenting an explicit discriminative stimulus (Context C) during extinction cut resurgent destructive behavior by 52–100% without any added reinforcement — visual or auditory cues that predict "no reinforcement is available right now" reliably reduced the post-extinction spike Fisher et al. (2020). Operationally this is the multiple-schedule logic most BCBAs already use during schedule thinning: a colored placemat, a visual timer, a session card — anything that lets the learner discriminate "currently SΔ" from "currently SD."

Third, schedule thinning and reinforcement-magnitude fading before extinction lowers the behavioral mass that has to be disrupted. The reinforcer-magnitude work above implies that fading magnitude across baseline before suspending the contingency reduces both the slope and peak of any subsequent burst Shahan & Avellaneda (2025) Lambert et al. (2024). Saini, Miller, and Fisher's review of 49 multiple-schedule applied studies notes that gradual schedule thinning paired with dense reinforcement of an alternative response is the field's standard recipe for keeping burst magnitude manageable across the SΔ component Saini et al. (2016).

Fourth, antecedent manipulation of the establishing operation — abolishing or attenuating the EO that drives the target response in the first place — is implied by every paper documenting that burst magnitude is moderated by reinforcer value relative to demand intensity Lambert et al. (2024). In practice this is the standard FCT precondition: ensure the learner is not in a state of high deprivation or aversive demand pressure when the extinction phase begins, and the burst that occurs (if it occurs) is smaller and shorter.

Two procedural variations on traditional extinction are now in the corpus and deserve brief mention. Kind extinction (Tarbox and colleagues, four autistic children) delivers genuine warmth and validation while functionally withholding the maintaining reinforcer (escape, tangible); the design produced 94–100% reductions in tantrums without an observable burst, suggesting that providing functionally arbitrary affective comfort does not undermine the extinction contingency when attention is not the maintaining reinforcer Tarbox et al. (2023). The caveat is procedural: confirm via FA or IISCA that attention is not a reinforcer before adding kind verbal contact during extinction, otherwise the "kindness" is reinforcement. Resurgence-mitigation procedures that include occasional brief extinction probes of the alternative response during DRA — Trask, Keim, and Bouton's 32-rat work — reduce the size of the eventual burst when the alternative is finally placed on extinction during schedule thinning Trask et al. (2018). The applied translation is to intermix small, planned non-reinforcement opportunities for the alternative response during DRA so the eventual leaning of that schedule generates less rebound.

Burst versus resurgence versus spontaneous recovery

The clinical ABA community sometimes uses "extinction burst" to label phenomena the basic literature treats as distinct, and that imprecision causes preventable arguments inside teams. Podlesnik and Baum's allocation/induction/covariance reanalysis is explicit: when reinforcement is withdrawn from a target behavior (B2), resources previously allocated there redistribute and briefly increase the formerly reinforced now-extinguished behavior (B1) — what the field often calls an "extinction burst" of B1 is, mechanistically, resurgence of B1, a lawful reallocation rather than a burst of strength Podlesnik & Baum (2024). Fujimaki, Goto, and Sakagami's discrete-trial rat work confirmed resurgence even in single-response-per-trial preparations, ruling out free-operant response density as a necessary condition Fujimaki et al. (2025). Randall and colleagues documented resurgence (operationally defined as any extinction-session response rate exceeding the immediately preceding phase) in three children with autism during extinction of an FCT response, again under clinically relevant conditions Randall et al. (2021). Spontaneous recovery — a separate phenomenon where an extinguished response reappears after a delay, even without additional reinforcement — is also distinct from a burst and from resurgence, though it is regularly conflated with both inside agency progress notes Fisher et al. (2023).

Practically, the distinction matters because the mitigation differs. A true burst of the target response in the first minutes/sessions of extinction calls for holding the line plus alternative reinforcement and discriminative-stimulus signaling Shahan & Avellaneda (2025) Fisher et al. (2020). Resurgence of a previously extinguished problem behavior when the alternative is later thinned calls for occasional alternative-response extinction probes during DRA and gradual schedule thinning Trask et al. (2018) Shahan et al. (2020). Spontaneous recovery calls for booster sessions and ongoing maintenance probes rather than a procedural change. Keep the labels separate in the BIP.

Detecting a burst versus calling treatment failure

A team that abandons extinction at the first sign of escalation is often inadvertently moving the contingency to intermittent reinforcement of the worst version of the target response — a far more pernicious learning history than the original baseline. Two things help. First, Barnard-Brak and colleagues' Bayesian N-of-1 method offers an objective rule for flagging burst data points in a single-case design: the first three treatment points that exceed the highest of the last five baseline points are operationally tagged as candidate burst data, distinguishable from sustained treatment failure by what happens in points 4 through N Barnard-Brak et al. (2020). Second, Nist and Shahan's measurement work argues for a short-window analysis — count or graph the first 5 minutes of extinction separately from session-wide rates, or you risk masking transient spikes inside session averages and either missing a real burst or inflating one that isn't there Nist & Shahan (2021). Pinkston and Foss's force-sensitive lever work in rats adds the corollary that within-class response intensity (gentle vs. hard hitting) can shift even when count metrics do not, so tracking force or magnitude alongside rate is recommended whenever the topography matters for safety Pinkston & Foss (2018).

Maintenance: the burst-after-the-burst

Even after an initial burst resolves, schedule thinning and contingency renewal can produce later spikes that look like new bursts. Fisher and colleagues' FCT work — and Saini and colleagues' multiple-schedule review — both note that introducing or re-introducing extinction during schedule thinning produces transient surges that can be mitigated with dense initial reinforcement of the alternative response and gradual movement toward leaner schedules Fisher et al. (2018) Saini et al. (2016). Muething and colleagues' retrospective is the single most relevant clinical data point here: burst probability did not diminish across repeated baseline-to-treatment transitions in their 108-case sample, so the team that "got through it once" should not be casual about the next contingency change Muething et al. (2024). Plan for vigilance at every extinction implementation, not just the first one.

When extinction is the wrong call

Because extinction can produce transient escalation in a subset of cases, the procedure is contraindicated — or requires modification — when that escalation has plausible safety consequences the team cannot manage. Severe self-injury, aggression that could injure caregivers or peers, elopement with imminent danger, and other "high-stakes topographies" all warrant an explicit pre-decision: either deliver extinction in a setting and staffing configuration that can absorb the burst safely (e.g., protected room, two-person staffing, padding), or don't run extinction-only and instead pair from the start with rich alternative reinforcement, antecedent manipulation, and discriminative-stimulus signaling that the corpus shows attenuates burst magnitude Shahan & Avellaneda (2025) Fisher et al. (2020) Saini et al. (2017). For severe SIB specifically, the field's more cautious posture is also driven by the fact that within-session topography (force, duration, novel forms) can intensify even when rate does not, so a "no rate change" reading is insufficient evidence that the procedure is safe Alessandri & Lattal (2021) Pinkston & Foss (2018).

02Evidence Tier Breakdown

A foundation page should be honest about where its evidence comes from Fisher et al. (2023). The extinction-burst literature lives mostly in the basic single-subject experimental design (SCED) layer with non-human subjects, plus a small but growing applied SCED and case-series layer with humans, plus one randomized factorial trial and a band of theoretical and review papers around the edges Fisher et al. (2023) Katz & Lattal (2021).

Randomized / quasi-experimental group studies. Lambert, Osina, and Copeland's 2 × 2 factorial randomized trial across 69 adults with developmental disabilities is the closest the corpus comes to a true between-subjects test, and it confirms that reinforcer value relative to demand intensity moderates burst magnitude and persistence Lambert et al. (2024). There is no head-to-head RCT comparing extinction-only versus extinction-plus-DR for burst outcomes in humans Fisher et al. (2023).

Reviews and theoretical analyses. Fisher and colleagues' 2023 JABA integrative review of basic and applied extinction-burst research is the most current synthesis and the source most BCBAs should anchor on for the temporal pattern and TWML mechanism Fisher et al. (2023). Shahan's Perspectives paper is the formal theoretical model Shahan (2022). Katz and Lattal's transitional-behavior analysis argues that "a single uniform definition of burst is inadequate" and reframes the construct around measurement and pre-extinction parameters Katz & Lattal (2021). Podlesnik and Baum's allocation/induction/covariance derivation reframes much of what clinicians call burst as resurgence — a lawful reallocation effect rather than a burst of strength Podlesnik & Baum (2024). Saini, Miller, and Fisher's narrative synthesis of 49 multiple-schedule applied studies anchors the burst-as-side-effect picture in clinical practice Saini et al. (2016).

Case series and applied retrospectives. Muething and colleagues' 108-case outpatient ABA record review is the largest clinical prevalence number in the corpus (24%, magnitude declining over sessions, no diminishment across repeated transitions) Muething et al. (2024). Woods and Borrero's pediatric food-refusal review (n = 10, 30% prevalence) is the only setting-specific clinical prevalence number Woods & Borrero (2019).

Single-subject experimental designs — basic / non-human. Most of the extinction-burst evidence base is here Katz & Lattal (2021). Nist and Shahan's measurement-dependence analysis (rats, FR schedules) Nist & Shahan (2021); Lattal, Kuroda, and Cook's three-experiment "little evidence of bursts" series in pigeons and rats following VR/VI baselines Lattal et al. (2020); Katz and Lattal's nine-pigeon multi-cycle study of burst variability Katz & Lattal (2020); Shahan and Avellaneda's two reinforcement-magnitude studies (12-rat baseline magnitude crossover Shahan & Avellaneda (2025); six-rat alternative-magnitude reversal Shahan & Avellaneda (2025)); Shahan, Browning, and Nall's choice-model treatment-duration study (six rats) Shahan et al. (2020); Fujimaki and colleagues' discrete-trial rat preparation Fujimaki et al. (2025); Pinkston and Foss's force-requirement study Pinkston & Foss (2018); and Trask, Keim, and Bouton's 32-rat alternating-extinction work on resurgence mitigation Trask et al. (2018). Cox's 2025 free-ranging-bird study is the only ecological-validity check in the corpus and confirms bursts occur outside controlled lab and clinic settings Cox (2025).

Single-subject experimental designs — applied / human. Fewer in number, but procedurally central. Alessandri and Lattal's force-tracking work in 22 adults (early-extinction force increases as a topographical burst) Alessandri & Lattal (2021); Tarbox and colleagues' four-child kind-extinction multiple-baseline (94–100% reduction without observable burst) Tarbox et al. (2023); Randall and colleagues' three-boy FCT-extinction reversal documenting clinically relevant resurgence Randall et al. (2021); Fisher and colleagues' four-child discriminative-stimulus mitigation study (52–100% burst reduction) Fisher et al. (2020); Fisher and colleagues' three-subject baseline-rate-and-resurgence study Fisher et al. (2019); Brown and colleagues' two-child DRA-with-vs-without-extinction comparison Brown et al. (2020); Saini, Fisher, and Pisman's three-child NCR-with-vs-without-extinction comparison Saini et al. (2017); and Fisher and colleagues' three-boy FCT differential-EO study (extinction side-effects during multiple-schedule thinning) Fisher et al. (2018).

Methodology / decision tools. Barnard-Brak, Richman, and Watkins' Bayesian N-of-1 rule for objectively flagging burst data points in single-case designs is the only formalized analytic procedure in the corpus and is useful for distinguishing burst from treatment failure Barnard-Brak et al. (2020).

Bottom line. The convergent picture is strong for the operational claims this page makes — that bursts are common but not universal, that prevalence is roughly 24–39% under combined treatment regimens (and substantially higher under extinction-alone or finer measurement windows), that burst magnitude scales with pre-extinction reinforcement value and density, and that alternative reinforcement, discriminative-stimulus signaling, and gradual thinning each meaningfully attenuate burst magnitude Muething et al. (2024) Shahan & Avellaneda (2025) Shahan & Avellaneda (2025) Fisher et al. (2020). It is weaker for any claim that one specific clinical mitigation package produces durably better outcomes than another in head-to-head human trials — that comparative-effectiveness layer barely exists Lambert et al. (2024). Most translational specifics are inferred from rat and pigeon parametric work and modest applied SCED, and the team should treat them accordingly when documenting clinical decisions.

03Decision Logic

The decisions a senior practitioner makes around an extinction burst are not "expect a burst or don't" so much as "is this case high-risk for a burst, and if it is, what mitigations are pre-positioned before extinction starts." A defensible logic, drawn directly from the corpus:

  1. New behavior plan with extinction component. Estimate burst risk by reviewing pre-extinction reinforcement parameters: reinforcer magnitude, baseline reinforcement density, duration of reinforcement history, and whether the alternative response has a long history of dense reinforcement Lambert et al. (2024) Shahan & Avellaneda (2025) Fisher et al. (2019) Shahan et al. (2020). High values on these load the burst.
  2. High burst-risk profile. Default to extinction plus dense reinforcement of an alternative response (DRA or FCT) rather than extinction-alone. Lerman-era data plus Fisher-era resurgence work converge: combined treatment cuts burst rate from ~36–62% to ~12–15% Nist & Shahan (2021) Saini et al. (2016).
  3. Severe topography (SIB, aggression, elopement-with-danger). Either run in a setting and staffing configuration that can absorb the burst safely, or do not run extinction-only. Pre-position alternative reinforcement, antecedent manipulation, and a discriminative-stimulus context Fisher et al. (2020) Shahan & Avellaneda (2025). If those mitigations are not feasible in the setting, escalate to a specialized program rather than running an underpowered extinction protocol.
  4. Function is escape and the maintaining reinforcer is escape from demands. Escape extinction will likely produce a brief burst in a minority of cases (≈30% in pediatric food refusal); pre-warn caregivers, plan continuous within-session data collection, and do not abort the demand sequence on the first spike Woods & Borrero (2019).
  5. Function is attention. Do not layer "kind extinction" on top — affective contact in this configuration is itself reinforcement. When attention is not the function, kind verbal validation during extinction can blunt distress without compromising suppression Tarbox et al. (2023).
  6. Multiple-schedule thinning into an extinction component. Use an explicit discriminative stimulus (visual/auditory cue) for the SΔ component; this can cut resurgent destructive behavior by 52–100% without added reinforcement Fisher et al. (2020). Thin gradually rather than abruptly, with dense initial reinforcement of the alternative Saini et al. (2016).
  7. Alternative response has been densely reinforced for an extended period. Expect a larger eventual burst when that alternative is later thinned. Mitigate by intermixing brief extinction probes of the alternative during DRA (Trask-style) and by gradual rather than abrupt schedule changes Trask et al. (2018) Shahan et al. (2020).
  8. Behavior worsens when extinction starts — burst or treatment failure? Apply a two-part test. (a) Use the Barnard-Brak rule: if early-treatment data points exceed the highest of the last five baseline points only for the first three points and then drop below baseline, that is operationally a burst, not failure Barnard-Brak et al. (2020). (b) Pull the first 5 minutes of each session out as its own data series — within-session patterns can mask either a real burst or a non-existent one inside session-wide averages Nist & Shahan (2021). If escalation persists past the burst window, revisit the FA/IISCA hypothesis, the EO, and the alternative-response density before declaring procedural failure.
  9. Plan to discontinue alternative reinforcement (e.g., FCT response on extinction). Expect resurgence of the original target behavior — clinically relevant in published children's data even when the original target had been suppressed for weeks Randall et al. (2021) Fujimaki et al. (2025). Do not abandon the protocol when this happens; thin gradually and pre-warn the family.
  10. Schedule-thinning step ahead of vacation, illness, or staff change. Treat caregiver capacity and consequence-fidelity risk as primary variables. Bursts during expected disruptions are predictable and worth pre-positioning extra coaching for, rather than treating as supervision failures after the fact.

04Across Settings

Clinic and outpatient

Clinic settings carry most of the published applied burst literature. Muething and colleagues' 108-case retrospective in an outpatient ABA clinic puts the prevalence number that most BCBAs should default to when communicating with families — bursts in 24% of cases, brief, with magnitude declining across sessions Muething et al. (2024). Brown and colleagues' DRA-with-vs-without-extinction comparison (two children, outpatient) and Saini, Fisher, and Pisman's NCR-with-vs-without-extinction comparison (three children, social-reinforcement maintenance) both anchor the practical recommendation that alternative reinforcement plus extinction is more durable than alternative reinforcement alone when the contingency is later disturbed Brown et al. (2020) Saini et al. (2017). Fisher and colleagues' four-child discriminative-stimulus mitigation study lives in the clinic literature as well and gives clinic teams the "colored-cue / session-card" lever — visible signaling of extinction context cuts burst magnitude by 52–100% Fisher et al. (2020). Risk-screening logic borrowed from the broader behavior-analytic literature applies here too: the higher the pre-extinction reinforcement density and reinforcer magnitude, the larger the expected burst, and the more important pre-positioned alternative reinforcement and discriminative signaling become before the first extinction session Shahan & Avellaneda (2025) Lambert et al. (2024).

School (K–12)

School-based extinction has two structural realities the clinic does not. First, the ethics of letting a burst happen in a classroom involve other students who did not consent to be present for it. Second, classroom staff implementing extinction are often paraprofessionals or general-education teachers without consequence-fidelity training. Both push toward the same operational answer: never run extinction-alone in a general-education classroom, always pair with rich differential reinforcement of an alternative response and an explicit discriminative stimulus that the team can discriminate at a distance (a card on the desk, a colored placemat, a teacher-held cue) Fisher et al. (2020) Saini et al. (2016). Saini, Miller, and Fisher's review of 49 multiple-schedule applied studies notes that gradual schedule thinning paired with dense reinforcement of the alternative is the standard recipe for keeping burst magnitude inside what classroom staff can absorb Saini et al. (2016). For escape-maintained classroom behavior in particular, the data point most relevant to classroom teams is that bursts in escape extinction are typically brief and resolve quickly when the demand sequence is held — but only when caregivers/staff are pre-trained to ride out the first minutes rather than abort the demand Woods & Borrero (2019).

Home and parent training

Home-based extinction is where the burst is most often misidentified as treatment failure, because parents who were not pre-warned read a brief escalation as "this isn't working" and quietly start reinforcing again — moving the contingency from extinction to intermittent reinforcement of the worst form of the target response. Two practical implications follow. First, pre-warn explicitly, in writing, with a number — "in roughly 1 in 4 cases under this combined plan, problem behavior briefly increases for a few minutes to a few sessions before decreasing; we will know it's a burst and not failure if X" Muething et al. (2024). Second, give caregivers a continuous-data structure rather than asking them to retrospectively judge severity; Nist and Shahan's measurement work argues for short within-session windows specifically because session-level summary judgments mask exactly the temporal patterns the team needs to see to distinguish burst from failure Nist & Shahan (2021). The kind-extinction procedure (when attention is not the function) is particularly appropriate for home implementation because it gives caregivers something to do during the extinction window — deliver warm verbal validation while functionally withholding the maintaining reinforcer — rather than asking them to ignore in silence Tarbox et al. (2023).

Residential and adult disability services

Residential settings concentrate three burst-relevant problems: severe topographies, dispersed staff, and inconsistent procedural fidelity across shifts. The Brown and colleagues / Saini, Fisher, and Pisman pairing applies here with particular force — extinction-alone configurations are higher-risk than combined extinction-plus-alternative-reinforcement configurations, and the difference is largest exactly when consequence fidelity drifts, which is where residential settings live Brown et al. (2020) Saini et al. (2017). Fisher and colleagues' discriminative-stimulus mitigation work is directly applicable: a session card or colored-token system that lets shift staff discriminate the SΔ (extinction) component from the SD (reinforcement) component cuts dependence on staff vigilance and reduces burst magnitude across the implementing team Fisher et al. (2020). For severe SIB or aggression, the field's more conservative posture applies: consider whether extinction-only is the right procedure at all, or whether the case sits in the band where a specialized inpatient or partial-hospital behavioral unit is the appropriate referral.

05Common Pitfalls

  • Quoting "things always get worse before they get better" as if it were universal. It isn't. Modern clinical data place burst prevalence around 24–39% under combined treatment regimens; quoting universality undermines clinical credibility and primes families to abandon plans at the first spike Muething et al. (2024) Woods & Borrero (2019).
  • Running extinction-alone when an alternative-reinforcement layer is feasible. Older clinical data place extinction-alone burst rates at 35.7–62%, dropping to 12–15% when extinction is combined with differential reinforcement; the burst is largely a tunable variable, and there is rarely a clinical reason to leave the lever unpulled Nist & Shahan (2021) Saini et al. (2016) Shahan & Avellaneda (2025).
  • Abandoning extinction during the burst. This converts the contingency into intermittent reinforcement of the worst version of the target response and is the single most pernicious clinical error the corpus identifies. Use the Barnard-Brak first-three-points rule and short within-session windows to distinguish burst from failure before changing the plan Barnard-Brak et al. (2020) Nist & Shahan (2021).
  • Failing to pre-warn caregivers and staff. Pre-warning is a procedural variable, not a soft skill. The teams that most often misread bursts as failure are the teams whose families and paraprofessionals were never given a number, a duration estimate, or a "we will know it's a burst if X" criterion Muething et al. (2024) Woods & Borrero (2019).
  • Tracking rate only. Bursts can manifest as increases in force, duration, or topographical variability without any change in count. Force-tracking work in 22 adults and force-requirement work in rats both show this directly; whenever topography matters for safety, track magnitude and duration alongside rate Alessandri & Lattal (2021) Pinkston & Foss (2018).
  • Including reinforcement-consumption time in baseline rate calculations. This inflates apparent burst prevalence by depressing the baseline-rate denominator. Exclude reinforcer-access time when the goal is to estimate ongoing response rate cleanly Nist & Shahan (2021).
  • Conflating burst, resurgence, and spontaneous recovery in the BIP. They are mechanistically distinct and have different mitigations. A burst calls for holding the line plus alternative reinforcement; resurgence during alternative-response thinning calls for graded thinning and intermixed extinction probes; spontaneous recovery calls for booster sessions and maintenance probes Podlesnik & Baum (2024) Trask et al. (2018).
  • Skipping the discriminative-stimulus layer during multiple-schedule thinning. A 52–100% reduction in resurgent destructive behavior at zero additional reinforcement cost is one of the largest free levers in the literature, and it is regularly omitted Fisher et al. (2020).
  • Using "kind extinction" without confirming function first. Verbal validation only attenuates the burst when attention is not the maintaining reinforcer. If attention is the function, kind verbal contact during extinction is reinforcement and the plan is no longer extinction Tarbox et al. (2023).
  • Assuming "we got through one burst" generalizes to future schedule changes. Burst probability did not diminish across repeated baseline-to-treatment transitions in 108 cases. Vigilance at every contingency change, not just the first one Muething et al. (2024).

06When to Refer Out

  • Severe SIB or aggression with realistic risk of significant injury during a burst. Refer to a specialized behavioral inpatient or partial-hospital program with the staffing and protected-setting capacity to implement extinction safely, rather than running an underpowered protocol in a setting that cannot absorb the burst Alessandri & Lattal (2021) Pinkston & Foss (2018).
  • Pica or other ingestion-related extinction protocols where the burst could produce medical harm. Move to a setting with the medical and behavioral integration to use baited-environment-with-safe-items strategies under supervision; do not run unsupervised in a home or general-education classroom.
  • Caseloads where consequence fidelity cannot be sustained across staff. When residential or school staffing cannot reliably hold the extinction contingency across shifts, the larger eventual resurgence (when the contingency inevitably drifts) is a treatment-design problem, not a staffing-management problem; refer to a more concentrated service intensity or restructure the plan to require less consequence-fidelity precision Saini et al. (2017) Brown et al. (2020).
  • Behavior that escalates well past the typical burst window with no within-session decline pattern. When the first-three-points rule and short-window analysis still produce sustained elevation rather than a brief spike followed by decline, the team is likely looking at a function or EO miss rather than a true burst; refer for FA/IISCA review by a specialized team rather than continuing to ride out an apparent burst that isn't one Barnard-Brak et al. (2020) Nist & Shahan (2021).
  • Active psychiatric crisis or imminent safety concern that overwhelms behavioral assessment timelines. Behavioral planning is not the right first move; refer to licensed mental-health crisis services and resume the extinction-based plan after stabilization.

07Future Research Directions

The honest read of the corpus is that the parametric story — what loads a burst, what attenuates it — sits on solid SCED evidence in non-humans and modest applied SCED evidence in humans, while the comparative-effectiveness layer for clinical mitigation packages is thin to nonexistent Fisher et al. (2023) Lambert et al. (2024). Lambert and colleagues' randomized factorial trial across 69 adults is the closest existing between-subjects test, and a head-to-head clinical trial of extinction-only versus extinction-plus-DRA versus extinction-plus-DRA-plus-discriminative-stimulus signaling would change how teams write behavior plans Lambert et al. (2024) Fisher et al. (2020). Prospective burst-prevalence studies in school and home settings would also clarify whether the 24–39% range from outpatient and feeding samples generalizes Muething et al. (2024) Woods & Borrero (2019).

The measurement question is unresolved. Nist and Shahan showed that burst prevalence inflates dramatically when the analysis window narrows and reinforcer-consumption time is included in baseline calculations; the field does not yet have a standardized within-session analysis protocol for clinical extinction sessions, and a published consensus method (e.g., "5-minute sliding window, reinforcer time excluded") would reduce noise in agency-to-agency burst-rate reports Nist & Shahan (2021). Barnard-Brak and colleagues' Bayesian N-of-1 rule offers an analytic anchor but has been tested on simulated rather than clinical data Barnard-Brak et al. (2020).

Theoretical work has accelerated. Shahan's TWML model and Podlesnik and Baum's allocation/induction/covariance reanalysis both reframe parts of the clinical "burst" picture as resurgence-style reallocation, and a clinical study explicitly testing TWML predictions against applied human data would clarify which mitigations the model endorses Shahan (2022) Podlesnik & Baum (2024). Kind extinction has been validated only in four autistic children; replications across larger and more diverse samples, and direct comparison to traditional extinction with observable burst metrics, are obvious next steps Tarbox et al. (2023). The Trask alternating-extinction-during-DRA finding, validated in 32 rats, has not been translated to applied clinical work — a clinical SCED testing whether intermixed extinction probes of the FCR during DRA reduce later resurgence in children would close one of the more directly translatable gaps Trask et al. (2018).

08Practitioner Takeaways

  1. Do not quote universality. The defensible practitioner statement is "in roughly 1 in 4 cases under combined treatment, and more often under extinction-alone, problem behavior briefly increases for a few minutes to a few sessions before declining." Communicate the number, not the metaphor Muething et al. (2024) Woods & Borrero (2019).
  2. Don't run extinction-alone when an alternative-reinforcement layer is feasible. Combined treatment cuts burst rates from ~36–62% to ~12–15% — pair with DRA, FCT, or NCR by default Nist & Shahan (2021) Saini et al. (2016) Shahan & Avellaneda (2025).
  3. Pre-warn caregivers and staff in writing, with a number, before extinction starts. Pre-warning is a procedural variable. Include the prevalence estimate, expected duration, and a written "we will know it's a burst if X" criterion for the first three sessions Muething et al. (2024).
  4. Track rate, magnitude, and duration — not rate alone. Force-tracking and force-requirement studies show topography can intensify even when count metrics do not. Whenever safety depends on topography, log all three Alessandri & Lattal (2021) Pinkston & Foss (2018).
  5. Use a short-window analysis. Pull the first 5 minutes of each early-extinction session as their own data series; session-wide averages mask exactly the temporal patterns you need to distinguish burst from failure Nist & Shahan (2021).
  6. Apply the Barnard-Brak rule before declaring treatment failure. First three treatment points exceeding the highest of the last five baseline points is operationally a burst. Sustained elevation past that window is something else and warrants FA/IISCA review Barnard-Brak et al. (2020).
  7. Exclude reinforcement-consumption time from baseline rate calculations. Including it inflates apparent burst prevalence by depressing the denominator. This is a clean, free measurement improvement Nist & Shahan (2021).
  8. Fade reinforcement magnitude before the first extinction session when the burst-risk profile is high. Burst magnitude scales with prior reinforcer magnitude; thinning richness in the lead-up reduces both peak and slope of any subsequent burst Shahan & Avellaneda (2025) Lambert et al. (2024).
  9. Use a discriminative stimulus to signal the extinction context. A colored placemat, session card, or visual timer can cut resurgent destructive behavior by 52–100% with no added reinforcement — one of the largest free levers in the literature Fisher et al. (2020).
  10. Confirm function before adding "kind extinction." Verbal validation only blunts the burst when attention is not the maintaining reinforcer. If attention is the function, the kindness is reinforcement and the plan is no longer extinction Tarbox et al. (2023).
  11. Plan for the burst-after-the-burst. Burst probability did not diminish across repeated baseline-to-treatment transitions in 108 outpatient cases; treat every contingency change as a fresh extinction event with full pre-positioning Muething et al. (2024).
  12. For long-history densely reinforced alternatives, intermix brief extinction probes during DRA. Trask-style alternating reinforcement/extinction of the alternative reduces the eventual resurgence size when that alternative is later thinned Trask et al. (2018) Shahan et al. (2020).
  13. Distinguish burst, resurgence, and spontaneous recovery in the BIP. They have different mechanisms and different mitigations; conflating them in supervision and progress notes produces preventable arguments and the wrong procedural change Podlesnik & Baum (2024) Trask et al. (2018) Fisher et al. (2023).
  14. For severe topographies, pre-decide the safety threshold at which extinction stops. Write down — before extinction starts — the topography, force, duration, or self-injury count at which the team pauses, calls supervision, and reverts to a richer alternative-reinforcement-only configuration. The pre-decision is the safety net Alessandri & Lattal (2021) Pinkston & Foss (2018).
  15. Document everything. Burst data is the field's primary defense against family or payer claims of "the plan made it worse." Continuous within-session data, an explicit pre-warning record, the Barnard-Brak determination, and the discriminative-stimulus configuration in the room together constitute the audit trail.

09Frequently Asked Questions

How often does an extinction burst actually happen?

The honest answer is "it depends on how you measure and what else is in the plan." In a 108-case outpatient ABA retrospective, bursts occurred in 24% of cases Muething et al. (2024). In a 10-child pediatric food-refusal review using escape extinction, bursts occurred in 30% of cases — brief, less severe than typically reported, and not compromising treatment Woods & Borrero (2019). Older clinical reanalyses by Lerman and colleagues placed bursts at 35.7–62% under extinction-alone and 12–15% when extinction was combined with another treatment Nist & Shahan (2021). Quote a range, not a metaphor.

Is the "things always get worse before they get better" line accurate?

Not really. Bursts are common but not universal, and the modern data make it clear that prevalence varies sharply with measurement window, schedule history, and whether extinction is paired with alternative reinforcement Muething et al. (2024) Nist & Shahan (2021) Lattal et al. (2020). Tell families and teachers what the literature actually shows, with a number.

What forms can a burst take?

Four, documented across the basic and applied corpus: (1) increase in rate (the textbook version), (2) increase in response magnitude or force, (3) increase in response duration, and (4) extinction-induced variability — novel response forms appearing inside the same operant class Fisher et al. (2023) Alessandri & Lattal (2021). A "no rate change" reading is not by itself evidence the procedure is safe; whenever topography matters, track magnitude and duration too Alessandri & Lattal (2021) Pinkston & Foss (2018).

How do I tell a burst from treatment failure?

Use a two-part rule. (a) The Barnard-Brak first-three-points rule: if the first three treatment points exceed the highest of the last five baseline points but then drop below baseline, that is operationally a burst Barnard-Brak et al. (2020). (b) Pull the first 5 minutes of each session as a separate data series — session-wide averages can mask exactly the temporal patterns that distinguish burst from failure Nist & Shahan (2021). If escalation persists with no within-session decline pattern past the burst window, revisit the FA/IISCA, the EO, and the alternative-response density before declaring failure.

What predicts a bigger burst?

Higher reinforcer value relative to demand intensity, larger baseline reinforcement magnitude, denser baseline reinforcement schedules, and a longer history of densely reinforced alternative responses each load the burst when the contingency eventually breaks Lambert et al. (2024) Shahan & Avellaneda (2025) Fisher et al. (2019) Shahan et al. (2020). Schedule type (intermittent vs. continuous) matters less reliably than density and magnitude.

What's the single most effective way to make the burst smaller?

Pair extinction with rich alternative reinforcement. The basic literature shows that high-magnitude alternative reinforcement during extinction can virtually eliminate the burst Shahan & Avellaneda (2025). The applied analogue — DRA or FCT layered on top of extinction — is the difference between Lerman-era 35.7–62% extinction-alone burst rates and 12–15% combined-treatment rates Nist & Shahan (2021) Saini et al. (2016).

When is extinction the wrong procedure?

When the topography is severe enough that the burst could realistically injure the learner or others and the setting cannot absorb the burst safely; when consequence fidelity cannot be sustained across staff or shifts; when caregivers cannot be reliably pre-warned and trained to ride out the first minutes; or when the case sits in a band where a specialized inpatient or partial-hospital behavior unit is the appropriate referral Alessandri & Lattal (2021) Fisher et al. (2020) Saini et al. (2017). In any of these, pair from the start with rich alternative reinforcement, antecedent manipulation, and discriminative-stimulus signaling — or escalate the case rather than running an underpowered extinction protocol.

Is "extinction burst" the same thing as "resurgence" or "spontaneous recovery"?

No, and the distinction matters because the mitigation differs. A burst is a transient increase in the target response in the first minutes/sessions of extinction — hold the line plus alternative reinforcement and discriminative-stimulus signaling Fisher et al. (2023) Shahan & Avellaneda (2025). Resurgence is the reappearance of a previously extinguished response when the alternative response is later placed on extinction — mitigate with intermixed extinction probes during DRA and gradual rather than abrupt thinning Podlesnik & Baum (2024) Trask et al. (2018). Spontaneous recovery is the reappearance of an extinguished response after a delay without additional reinforcement — handle with booster sessions and maintenance probes. Keep the labels separate in the BIP.

Does the burst get smaller each time we go through one?

Magnitude often does decline within a single course of extinction, but probability across repeated baseline-to-treatment transitions does not reliably diminish — Muething and colleagues' 108-case retrospective found no meaningful drop in burst probability across repeated transitions Muething et al. (2024). Plan with full pre-positioning at every contingency change, not just the first one.

Can I use "kind extinction" — being warm during the procedure?

Sometimes. When the maintaining reinforcer is escape or tangible, providing genuine affective warmth and validation while functionally withholding the maintaining reinforcer produced 94–100% reductions in tantrums without an observable burst across four autistic children Tarbox et al. (2023). When the maintaining reinforcer is attention, however, kind verbal contact during extinction is itself reinforcement and the plan is no longer extinction. Confirm function via FA or IISCA before adding the kind component.

10References

Primary research synthesized in this guide. DOIs link to the original source.