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

Differential Reinforcement: A Practitioner's Guide to DRO, DRA, DRI, DRL, DRH, and NCR

Query target: differential reinforcement · BBC Editorial Team
★ Summary

Differential reinforcement (DR) is the procedural family in which one class of responses is reinforced while reinforcement is withheld, minimized, or shifted away from another class — and it is the field's primary route to behavior change without aversive contingencies. The standardization papers in the corpus argue that DR is best defined not by any single topography but by differential delivery of reinforcement along at least one parameter (rate, magnitude, immediacy, or quality) so that the appropriate response receives a richer consequence than the target behavior, with extinction for the target as the ideal — but not the definitional — case Vollmer et al. (2020). Practically, this means a BCBA's job inside DR is rarely "use DRO" or "use DRA" by reflex; it is to pick the variant that matches the function of the target behavior, calibrate the reinforcement parameters that actually compete with that function, write down an interval or rate criterion that a second clinician could replicate, and plan schedule thinning before treatment effects start to look settled.

01What the Research Says

Differential reinforcement, defined by parameter rather than by topography

Vollmer and colleagues' definitional paper is the operational anchor for the entire DR family: DRA — and by extension every DR variant — is best understood as delivering greater reinforcement along at least one dimension contingent on one form of behavior, while minimizing reinforcement for another Vollmer et al. (2020). Extinction of the target is ideal but not required; the procedure remains DRA when extinction is impossible for safety, ethical, or logistical reasons, as long as the differential is real along quality, magnitude, immediacy, or rate Vollmer et al. (2020). This reframing matters in practice because it tells you what to manipulate when behavior won't change: not whether you have "true DRA," but whether your differential is large enough along the dimensions the client is actually sensitive to.

The systematic review of DR in skill acquisition operationalizes the differential as four nested levels — No DR, DR1, DR2, DR3 — each escalating how much the reinforcer for prompted responses is degraded relative to the reinforcer for independent responses (Cividini-Motta et al., 2024). Across 10 studies of children with ASD in discrete-trial instruction, the more differential the schedule, the faster the acquisition: arrangements that reserved the most potent reinforcer for independent responses and gave only praise (or nothing) for prompted ones consistently produced quicker mastery than non-differential conditions (Cividini-Motta et al., 2024). The review also flags two coverage gaps that should temper certainty: no included study manipulated the immediacy parameter, and maintenance/generalization data after DR-driven mastery are scarce (Cividini-Motta et al., 2024).

The DR family

DRO — differential reinforcement of other behavior. DRO reinforces the absence of the target response across an interval. The corpus is unusually clear that DRO's mechanism is not what most graduate textbooks describe. A systematic replication with three undergraduates showed that DRO contingencies reliably increased non-target alternative responses never explicitly programmed for reinforcement; this adventitious strengthening of "other" responses accounted for much of DRO's suppressive effect on the target Rey et al. (2020). Hangen and colleagues extended the point clinically across five children: the rise in unspecified "other" behavior under DRO disappears when extinction replaces the DRO reinforcement component while target rates stay low — the reinforcement part of DRO, not the extinction part, drives the increase in collateral behavior Hangen et al. (2020). The clinical implication is the inverse of the textbook one: rather than letting DRO "automatically" reinforce whatever the client happens to be doing, define and program a specific desirable alternative — layer DRA onto DRO — to keep the strengthened class clinically relevant Rey et al. (2020) Hangen et al. (2020).

For DRO interval selection, the FM-vs-VM-DRO comparison across four children with autism showed both schedules were roughly equally effective at reducing automatically maintained challenging behavior, but caregivers preferred VM-DRO and noted that FM-DRO can become discriminable and lose effectiveness when the schedule is predictable Wilder et al. (2023). FM-DRO remains a viable, easier-to-implement option when staffing is thin; switch to VM-DRO when you see momentary checks getting "gamed" or when the topography is reliable enough for the client to time the interval Wilder et al. (2023). Even at the case-study layer, momentary DRO has clinical reach: a brief FM-DRO with a small response-cost component produced rapid, large reductions in motor and vocal stereotypy for an adult with ASD across natural work and leisure settings — a reminder that DR can suppress long-standing automatic behavior in adults without restraint or response blocking Butler et al. (2021).

DRA — differential reinforcement of an alternative behavior. DRA reinforces a defined alternative response whose only requirement is that it be different from the target. The Hedquist and Roscoe alternating-treatments comparison across three children with autism is the cleanest direct test in the corpus: when neither procedure used response blocking, DRA produced larger and more consistent reductions in automatically reinforced stereotypy — and bigger gains in task engagement — than DRO did Hedquist & Roscoe (2020). Iannaccone and Jessel's analog DRA-without-extinction model with 32 college students showed that systematically enhancing reinforcer magnitude, immediacy, or quality for the alternative response reliably reduced the target operant, with all three dimensions combined producing the largest, most durable suppression Iannaccone & Jessel (2023). Briggs and colleagues replicated this clinically across four children with escape-maintained destructive behavior: DRA without extinction worked when the reinforcer for compliance was richer in quality and/or magnitude, but maintaining low rates during thinning required keeping both dimensions elevated Briggs et al. (2019). Weinsztok and DeLeon then showed that enhanced magnitude and quality on the alternative response also buffer DRA against the integrity errors that inevitably occur in real settings — front-loading rich reinforcement is itself a fidelity safeguard Weinsztok & DeLeon (2022).

The cleanest extinction-versus-no-extinction comparison comes from Brown and colleagues' resurgence preparation: at equated overall reinforcement rates, DRA with extinction produced lower target responding during treatment and less resurgence during a later extinction challenge than DRA without extinction Brown et al. (2020). The implication is not that extinction is always required — Vollmer's definitional move stands — but that extinction is the strongest single lever for protecting DRA against later resurgence when integrity slips Brown et al. (2020) Vollmer et al. (2020).

DRI — differential reinforcement of incompatible behavior. DRI is DRA with an added topographical constraint: the alternative must be physically incompatible with the target. The corpus does not maintain a sharp procedural boundary between DRA and DRI — the equivalence-based instruction work treats DRO, DRA, and DRI as a coherent set learned together Kelly‐Sisken et al. (2025) — but the conceptual move matters when the function is automatic or when the target is hands-busy work like SIB topographies. Choose DRI when you need the alternative response to mechanically preclude the target (hands occupied with a manipulative for hand-mouthing; sustained vocalization on a sung verse for non-verbal stereotypy) rather than merely competing with it for reinforcement.

DRL — differential reinforcement of low rates. DRL is the variant clinicians underuse, in part because it is the least intuitive: you are reducing — not eliminating — a response by reinforcing only when it occurs at or below a target rate. Becraft and colleagues' alternating-treatments comparison across five preschoolers showed that signal structure is the load-bearing variable: pairing the schedule with an S+ (reinforcement available) and S− (reinforcement unavailable) produced steady, low-rate responding under both spaced-responding and full-session DRL formats, but removing the S+ produced unstable or eliminated responding rather than the targeted low rate Becraft et al. (2018). The procedural implication is concrete — never run DRL with only an S− or no signals at all, because you will end up extinguishing the behavior you wanted to reduce Becraft et al. (2018). Ré and colleagues' classroom application across eight elementary-age children with autism confirmed DRL's reach beyond the lab: an interdependent group contingency tied to a class-wide reinforcer like extra recess produced a functional reduction in vocal disruption with simple frequency counts as the only data burden, making DRL a reasonable first-line move when the target is "too much of an otherwise appropriate response" rather than a behavior that should disappear entirely Ré et al. (2024).

DRH — differential reinforcement of high rates. DRH is the symmetric opposite of DRL: reinforce only when responding meets or exceeds a target rate. The corpus contains no direct DRH outcome studies, but the family-level mechanism is the same as DRL — a rate criterion gates reinforcement, with parameter and signal structure following the DRL logic Becraft et al. (2018) Vollmer et al. (2020). Use DRH for increasing the frequency of an already-acquired response (work-rate building, verbal initiations, exercise), not as a reduction tool.

NCR — noncontingent reinforcement. NCR sits adjacent to the DR family rather than inside it: reinforcers are delivered on a time-based schedule independent of behavior, abolishing the EO rather than reinforcing differential responding. Fritz and colleagues' multiple-baseline study across five children showed how NCR and DR fit together: NCR alone (without extinction) suppressed socially maintained problem behavior and supported thinning to FT-5-min, but two of five cases re-emerged during thinning and were rescued only by adding DRA Fritz et al. (2017). Berth and colleagues' feeding-disorder comparison adds nuance: DR alone eliminated food refusal in one child and liquid refusal in another, and adding DR to escape extinction produced more stable acceptance than NCR did Berth et al. (2019). NCR is good at lowering motivation and good at not requiring extinction, but you usually need DRA layered in to teach something rather than just to suppress something.

Function-based DR: matching the reinforcer to the function

Johnson and colleagues' assessment-based DR work across three children with autism is the corpus's clearest case for individualizing the parameter, not just the topography: a brief alternating-treatments assessment compared quality, schedule, and magnitude variants of DR plus a non-differential control during auditory-visual matching, and a quality-based DR produced the fastest unprompted correct responding for all three learners Johnson et al. (2017). Predictive validity held within the same skill but was inconsistent for novel skills — you can't pick a DR parameter once and assume it will travel Johnson et al. (2017). McCormack and colleagues' differential-outcomes work extended the point: arranging unique reinforcers for each correct response in a conditional discrimination accelerated tact learning for two of three children, but one child showed no added benefit — individual parameter sensitivity is what matters McCormack et al. (2017).

The same logic applies in reverse for behavior reduction. When DRA is delivered without extinction, the function-matched reinforcer for the alternative has to actually compete with the function maintaining the target — Briggs and colleagues' escape-maintained destructive behavior case is the canonical example, where compliance had to be paid in preferred items at sufficient quality and magnitude before destructive behavior dropped, with thinning requiring both dimensions to stay enriched Briggs et al. (2019). When the reinforcer for the alternative is "arbitrary" — chosen for convenience rather than function-match — DR effects are smaller and less durable Iannaccone & Jessel (2023) Weinsztok & DeLeon (2022).

DRO interval selection: whole versus momentary, fixed versus variable

DRO interval selection is one of the few DR parameters with explicit comparative data. Wilder and colleagues' four-child FM-vs-VM-DRO comparison showed equal effectiveness on automatically maintained challenging behavior, with caregiver preference and durability favoring VM-DRO because the variable interval resists discrimination Wilder et al. (2023). The general decision rule: start with the interval below the mean inter-response time observed in baseline; use momentary sampling rather than whole-interval when implementation resources are constrained; prefer variable over fixed intervals when the topography is reliable enough for the client to time a fixed schedule Wilder et al. (2023). Jessel and colleagues' single-case extension to on-task behavior shows momentary DR is also a thinning vehicle: 30-s supervision checks with immediate tokens for engagement progressively raised one adolescent's worksheet engagement while supervision density was thinned to a 5-min schedule Jessel et al. (2017).

Schedule thinning across DR variants

Schedule thinning is the phase where most DR plans quietly fail, and three converging lines of evidence in the corpus point to the same architecture Nevin et al. (2016). Nevin and colleagues' work across pigeons and four children with severe problem behavior showed that rich DRA schedules suppressed problem behavior faster than lean schedules during treatment but produced greater post-treatment resurgence when DRA was discontinued — dense reinforcement builds behavioral momentum that bites you when you stop Nevin et al. (2016). Briggs and colleagues' escape-maintained DRA-without-extinction work showed that thinning was sustainable only when both magnitude and quality remained robust Briggs et al. (2019). Fritz and colleagues' NCR+DRA package thinned successfully from continuous out to FT-5 min by adding DRA the moment NCR thinning produced any return of problem behavior Fritz et al. (2017). The composite rule: thin slowly, never simultaneously thin two reinforcement dimensions, and have a layered DR component ready to insert when thinning the first one produces re-emergence Nevin et al. (2016) Briggs et al. (2019) Fritz et al. (2017).

Signaling is the second thinning consideration. Bland and colleagues' pigeon work showed that signaled DRA produced more persistent treatment effects during the active phase but greater resurgence than unsignaled DRA when the signal was removed Bland et al. (2016). Nevin and colleagues replicated the broader picture, adding that the rate of alternative reinforcement — not signaling — was the primary driver of resurgence magnitude, though signaling shifts the pattern enough to warrant brief extinction probes before full signal fade Nevin et al. (2016). The applied translation: if you build a discriminative signal into FCT or DRA (a card, visual cue, or device prompt), thin both the schedule and the signal, and use intermittent extinction probes to flag resurgence before full fade Bland et al. (2016).

DR and resurgence: the field's most-studied integrity failure

Resurgence — the temporary return of a previously reinforced target response when its reinforcement, or alternative reinforcement, is interrupted — is where DR plans break down in the clinic. The corpus offers four useful tools for guarding against it. (1) DRA plus extinction, when extinction is feasible, produces less resurgence than DRA alone at equated reinforcement rates Brown et al. (2020). (2) Inserting a brief 3-s DRO contingency during early extinction sessions sharply attenuates resurgence of the previously reinforced target, with minimal additional reinforcer cost — and the schedule can be quickly faded once resurgence risk passes King et al. (2025). (3) Delivering the alternative reinforcer in a distinct context from the one in which the target was originally reinforced reduces relapse compared with same-context DRA, suggesting clinicians can deliberately stage replacement reinforcement in a different room, device, or location for stronger renewal protection Craig et al. (2018). (4) Carrying the same DRA contingency back into the original problem context blunts ABA-style renewal of the target response when treatment resumes in baseline settings Kimball et al. (2020). Each of these tools costs little procedurally and stacks well, which is why the typical clinical DR package now layers extinction, brief DRO probes, and context-management on top of the basic DRA contingency.

A complication from the experimental side: King and colleagues' human-operant work, and Jimenez-Gomez and colleagues' choice studies in children and adults, both argue that reinforcers function partly as discriminative stimuli predicting future contingencies — not purely as response strengtheners — which means clients' choices under DR are driven by the contingency that the reinforcer predicts, not just by the reinforcer itself King et al. (2025) Jimenez‐Gomez et al. (2025). Practically: program clear and consistent discriminative signals around when alternative reinforcement is available, monitor choice patterns rather than just rates, and expect more behavior change when the predictive meaning of your DR signal is unambiguous.

DR vs FCT: closely related, not identical

Functional Communication Training (FCT) is a structured DRA in which the alternative response is explicitly a communication response — a mand, sign, picture, or device exchange — and the reinforcer for that mand is the same reinforcer that maintains problem behavior Vollmer et al. (2020). DRA is the broader category; FCT is the specific implementation when the alternative is communicative. The renewal and resurgence work above applies to both, but the FCT signal (card, device, modeled sign) functions as the discriminative stimulus whose presence/absence governs resurgence once treatment is faded Bland et al. (2016) Nevin et al. (2016). Choose FCT when the function is socially mediated and the client has (or can quickly acquire) a matched communication topography; choose generic DRA when the alternative is non-communicative (task engagement, hands-busy activity, worksheet completion) or when the function is automatic and a mand would not match it.

Three-choice DR, onset, and integrity

Kestner and colleagues' systematic review comparing two- versus three-choice DR procedures in three children with ADHD/ASD showed that expanding the number of high-quality alternatives improved persistence of appropriate behavior and reduced relapse of problem behavior, even when all options were equated for reinforcement quality — choice itself functions as a DR parameter Kestner et al. (2023). Three smaller but operationally important findings complete the picture. Campanaro and colleagues compared onset timings across three preschool-age learners and found initiating DR from the very first teaching trial produced the fastest acquisition; "wait-and-see" delays were costly in trial count Campanaro et al. (2020). Chang and colleagues' replication across two preschoolers found item-by-item mastery criteria reduced trials to mastery without harming retention — criterion granularity is itself a DR parameter (Chang et al., 2024). Jones and colleagues' translational work on consequence-based fidelity errors post-mastery showed that brief intermittent omissions accidentally shift FR1 into variable-ratio, increasing persistence of the previously-mastered behavior in ways the team did not program (Jones et al., 2026). DR integrity is a procedural variable, not a soft skill: small changes in onset, criterion granularity, and consequence delivery meaningfully change the schedule the client is actually contacting Campanaro et al. (2020) (Jones et al., 2026).

A definitional caveat from the experimental side

Allen and colleagues' comparison of symmetrical (both topographies reinforced) versus asymmetrical (appropriate behavior on extinction) DR contingencies during functional analyses argues that asymmetrical schedules can confound assessment of EO sensitivity and bias toward challenging behavior, because what looks like low appropriate responding may actually be an extinction effect (Allen et al., 2026). When DR is embedded in an assessment phase, use symmetrical contingencies so you can measure response allocation accurately and set the DR ratio for treatment based on a clean baseline (Allen et al., 2026). The point also reframes Vollmer et al.'s definitional move: if extinction can confound which dimension of reinforcement actually matters, defining DRA by differential delivery rather than by extinction is also methodologically cleaner Vollmer et al. (2020) (Allen et al., 2026).

Beyond clinical DR: rule-governed and operant-economic extensions

Two adjacent lines extend DR's reach. Ruiz Méndez's rule-governed-choice procedure with eight college students showed that when competing rules pointed to different responses, the rule paired with the richer schedule reliably won — DR governs which instructions a learner follows under conflict (Ruiz Méndez, 2024). Magoon and colleagues' free-operant work, equating positive and negative reinforcement on symmetrical schedules, produced steeper matching slopes for negative reinforcement — escape and presentation contingencies are not behaviorally equivalent even when scheduled identically Magoon et al. (2017). Gomes-Ng and colleagues' delayed-matching work in pigeons quantifies the general lesson: the ratio of reinforcers for correct versus error responses directly redistributes stimulus control, so inadvertent reinforcement of errors weakens control by the dimension you intended to teach Gomes‐Ng et al. (2023).

Teaching DR concepts to staff

Kelly-Sisken and colleagues' between-subjects comparison of equivalence-based instruction (EBI) versus PowerPoint lecture for teaching DRO, DRA, and DRI to college students found EBI produced stronger acquisition of names, definitions, and vignette-based application, with better generalization to novel exemplars Kelly‐Sisken et al. (2025). The training implication: build EBI modules with at least two exemplar vignettes per procedure and probe a third for generalization, rather than relying on passive lecture Kelly‐Sisken et al. (2025). Halbur and colleagues' enhanced-data-sheet study extends the logic to ongoing supervision: a visible "DR" column on the program sheet cues contingent reinforcement delivery without extra observation time (Halbur et al., 2024). Waite and colleagues' caretaker-implemented ear-cleaning protocol for companion dogs is a sanity check that DR generalizes outside human clinical settings: a two-choice DR procedure delivering food plus marker for cooperative handling, with protocol termination for non-cooperation, produced acceptance of intrusive ear care without restraint (Waite et al., 2025).

02Evidence Tier Breakdown

The DR literature lives mostly in single-subject experimental design (SCED), with a smaller layer of systematic reviews, a thin layer of group/quasi-experimental work, and a healthy strip of translational human-operant and animal studies that directly inform clinical thinking (Cividini-Motta et al., 2024) Iannaccone & Jessel (2023).

Systematic reviews. Cividini-Motta and colleagues synthesize 10 SCED studies across DR1-DR3 levels and anchor the differential-parameter reasoning on this page (Cividini-Motta et al., 2024). Kestner and colleagues establish the empirical basis for treating number of alternatives as a DR parameter Kestner et al. (2023).

Group / quasi-experimental work. Kelly-Sisken and colleagues' between-subjects comparison of EBI versus PowerPoint lecture for teaching DR concepts is the closest the page has to a group design, and it sits in the staff-training rather than the client-outcome layer Kelly‐Sisken et al. (2025). Halbur and colleagues' enhanced-data-sheet study evaluates fidelity rather than client outcomes at the same tier (Halbur et al., 2024).

Clinical SCED. The applied DR evidence base is overwhelmingly single-subject Hedquist & Roscoe (2020): Hedquist and Roscoe (DRA vs DRO) Hedquist & Roscoe (2020); Briggs et al. (DRA without extinction for escape) Briggs et al. (2019); Johnson et al. (assessment-based parameter selection) Johnson et al. (2017); McCormack et al. (differential outcomes) McCormack et al. (2017); Campanaro et al. (DR onset) Campanaro et al. (2020); Becraft et al. (DRL signals) Becraft et al. (2018); Wilder et al. (FM vs VM DRO) Wilder et al. (2023); Hangen et al. (DRO "other" mechanism) Hangen et al. (2020); Fritz et al. (NCR+DRA) Fritz et al. (2017); Berth et al. (feeding) Berth et al. (2019); Ré et al. (classroom DRL group contingency) Ré et al. (2024); Butler et al. (adult stereotypy DRO) Butler et al. (2021); Fulton et al. (distributed vs accumulated reinforcement) Fulton et al. (2020); Jessel et al. (on-task momentary DR) Jessel et al. (2017); Chang et al. (acquisition criteria) (Chang et al., 2024).

Translational and basic experimental work. Iannaccone & Jessel (parameter manipulation) Iannaccone & Jessel (2023); Brown et al. (DRA with vs without extinction; resurgence) Brown et al. (2020); Rey et al. (DRO adventitious reinforcement) Rey et al. (2020); Kimball et al. (ABA renewal under DR) Kimball et al. (2020); King et al. (DRO during extinction) King et al. (2025); Bland et al. (signaled DRA, pigeons) Bland et al. (2016); Nevin et al. (rich vs lean DRA) Nevin et al. (2016); Craig et al. (distinct-context delivery) Craig et al. (2018); Gomes-Ng et al. (divided stimulus control) Gomes‐Ng et al. (2023); Weinsztok & DeLeon (DRA integrity) Weinsztok & DeLeon (2022); Magoon et al. (positive vs negative reinforcement) Magoon et al. (2017); Jimenez-Gomez et al. (discriminative control of choice) Jimenez‐Gomez et al. (2025); Ruiz Méndez (rule-governed choice) (Ruiz Méndez, 2024); Jones et al. (post-mastery fidelity decay) (Jones et al., 2026).

Theoretical / conceptual. Vollmer et al.'s definitional paper on DRA without extinction is the operational anchor for parameter-based DR thinking Vollmer et al. (2020). Allen et al.'s methodological argument for symmetrical DR contingencies during FA reframes how DR should be embedded in assessment phases (Allen et al., 2026).

Bottom line. The evidence is strong for this page's operational claims — that DR variants differ procedurally but share a parameter-based logic, that DR plans need extinction or extinction-like protections to resist resurgence, that schedule thinning is the highest-failure phase, and that integrity is a procedural variable Vollmer et al. (2020) Brown et al. (2020) Nevin et al. (2016) (Jones et al., 2026). It is weaker for any claim that one DR variant produces durably better outcomes than another at the group level: the head-to-head comparisons we have (DRA vs DRO, FM- vs VM-DRO, signaled vs unsignaled DRA) are SCED with small Ns and would benefit from larger replication Hedquist & Roscoe (2020) Wilder et al. (2023) Nevin et al. (2016).

03Decision Logic

The DR decisions a senior practitioner makes are not "DRO or DRA" so much as "which variant fits the function, the rate, the setting, and the resurgence risk." A defensible logic, drawn from the corpus:

  1. Severe, low-rate target with no clear functional alternative yet. Run DRO with momentary sampling set below the baseline mean inter-response time; default to VM-DRO unless training capacity forces FM-DRO Wilder et al. (2023). Always layer a defined desirable alternative onto DRO to direct the adventitious reinforcement DRO will produce anyway Rey et al. (2020) Hangen et al. (2020).
  2. Skill replacement is the goal. Use DRA, ideally with extinction if the function and severity allow it. Front-load enhanced quality and magnitude on the alternative — both as the primary mechanism and as an integrity buffer Iannaccone & Jessel (2023) Weinsztok & DeLeon (2022).
  3. Hands-busy or topographically incompatible target. Use DRI: choose an alternative that mechanically precludes the target (object manipulation for hand-mouthing, sustained motor pattern competing with stereotypy).
  4. Reduce, but not eliminate, an otherwise appropriate response. Use DRL with explicit S+ and S− signals — never DRL with only an S− or no signals Becraft et al. (2018). For classroom-level rate reduction, an interdependent group contingency tied to a class-wide reinforcer is a reasonable first-line move Ré et al. (2024).
  5. Increase the rate of an already-acquired response. Use DRH with a clear rate criterion and signal structure; this is a skill-building tool, not a reduction tool.
  6. Suspected EO/motivation problem before DR works. Run NCR (ideally without extinction) to abolish the EO and stabilize behavior, then layer DRA for the replacement; thin no thinner than FT-5-min before reassessing, and add DRA the moment thinning produces re-emergence Fritz et al. (2017).
  7. Socially mediated function with a matched communication topography. Use FCT and plan signal/schedule thinning carefully, because both rate and signal removal modulate resurgence Bland et al. (2016) Nevin et al. (2016).
  8. Extinction is impossible for safety, ethics, or logistics. DRA without extinction is still DRA if the differential along quality and magnitude (and ideally immediacy) is large enough to compete with the function Vollmer et al. (2020) Iannaccone & Jessel (2023). Keep both dimensions enriched throughout thinning Briggs et al. (2019).
  9. High resurgence risk. Stack four safeguards: DRA + extinction when feasible Brown et al. (2020); brief DRO during early extinction King et al. (2025); alternative reinforcement delivered in a context distinct from the original problem context Craig et al. (2018); the DRA contingency carried back into the original context to blunt renewal Kimball et al. (2020).
  10. Choice problem. Default to three-choice DR for active treatment phases — choice itself functions as a DR parameter, and three high-quality alternatives outperform two at equated reinforcement Kestner et al. (2023).
  11. Onset of DR for new skill acquisition. Begin DR at the very first teaching trial; do not wait for errors to emerge Campanaro et al. (2020). Use item-level mastery criteria rather than set-level when efficiency matters (Chang et al., 2024).
  12. Embedding DR in FA or assessment phases. Use symmetrical contingencies (both topographies reinforced) so what looks like low appropriate behavior cannot be a hidden extinction effect (Allen et al., 2026).

04Across Settings

Clinic (outpatient and intensive)

Outpatient and intensive-clinic settings host most of the SCED work that anchors the DR family — Hedquist's DRA-vs-DRO comparison, Briggs's DRA-without-extinction protocol, Johnson's assessment-based parameter selection, and Berth's feeding-disorder DR studies all run in clinics with structured sessions and embedded preference assessments Hedquist & Roscoe (2020) Briggs et al. (2019) Johnson et al. (2017) Berth et al. (2019). The operational pattern is consistent: assess parameters individually before locking them in Johnson et al. (2017), front-load magnitude and quality on the alternative Iannaccone & Jessel (2023) Weinsztok & DeLeon (2022), pair DRA with extinction when feasible Brown et al. (2020), and thin slowly with both magnitude and quality elevated until rates are stable Briggs et al. (2019). Feeding disorders are a particularly clean clinic case: DR alone can occasionally replace escape extinction outright, and DR + escape extinction produces faster acceptance and fewer disruptive vocalizations than NCR + escape extinction Berth et al. (2019).

School (K-12)

Schools impose two structural constraints DR has to bend around: classroom-level group dynamics and limited 1:1 staff time. Ré and colleagues' DRL interdependent group contingency across eight elementary-age children with ASD shows that class-wide DRL tied to a shared reinforcer like extra recess can functionally reduce vocal disruption with frequency counts as the only data burden — the right first-line move for over-talkers, calling out, or transition disruption Ré et al. (2024). Becraft and colleagues' signal study tells you exactly how to set DRL up so it produces low rates rather than zero rates: pair the schedule with both an S+ and S− signal — students need both Becraft et al. (2018). For DRA in classrooms — hand-raising, on-task engagement, transition compliance — Jessel's momentary DR case shows you can run staff "spot checks" every 30 s with immediate token delivery and thin to 5-min checks while preserving rates Jessel et al. (2017). The data-sheet integrity move from Halbur and colleagues — a "DR" column on the program sheet — costs nothing and improves contingent delivery during instructional trials (Halbur et al., 2024).

Home

Home-based DR runs into two problems clinics rarely face: caregivers are the implementers, and the reinforcer competing with the function is often something the caregiver cannot fully withhold. Vollmer and colleagues' definitional move — DRA without extinction is still DRA when the differential is real along quality, magnitude, or immediacy — is the load-bearing claim for home practice Vollmer et al. (2020). Fritz and colleagues' NCR-without-extinction-plus-DRA package is built precisely for this constraint: parents deliver a rich time-based schedule of attention/items, layer in mand reinforcement when problem behavior re-emerges during thinning, and avoid extinction entirely Fritz et al. (2017). Halbur and colleagues' enhanced-data-sheet finding suggests parents implementing DR benefit from a tracking artifact (printed sheet, app, or whiteboard) that prompts contingent reinforcement delivery without requiring continuous BCBA observation (Halbur et al., 2024).

Residential and adult disability services

Residential settings concentrate three problems: severe topographies, dispersed staff, and high integrity-decay risk. Butler and colleagues' adult-stereotypy DRO case in a 20-year-old with ASD shows momentary DRO with a brief response-cost component working across natural work and leisure settings, with effects maintained without specialized environments — a useful counter-example to the assumption that adult automatic-reinforcement topographies require restrictive procedures Butler et al. (2021). The Briggs DRA-without-extinction protocol applies when destructive behavior is escape-maintained: keep both magnitude and quality of the reinforcer for compliance enriched throughout thinning, and expect re-emergence if either dimension drops Briggs et al. (2019). Jones and colleagues' fidelity-decay finding is particularly important here: post-mastery integrity erosion is fast and largely invisible to antecedent-only fidelity checks, so consequence-fidelity probes need to live in the supervision plan (Jones et al., 2026).

OBM and adult workplace applications

DR's logic generalizes to organizational behavior management when the goal is shifting employee response rates rather than reducing problem behavior. DRH is the OBM-natural variant; DRL applies when the goal is reducing an over-occurring response (interruptions, off-topic comments, over-emailing). Magoon and colleagues' free-operant work showing steeper matching slopes for negative reinforcement is directly relevant: monetary loss-avoidance and presentation of positive consequence are not interchangeable even at equated schedules, so OBM DR plans should pick the reinforcement modality the workforce is most sensitive to rather than defaulting Magoon et al. (2017). The integrity layer applies the same way — visible tracking artifacts, three-choice options where possible, and consequence-fidelity probes (Halbur et al., 2024) Kestner et al. (2023) (Jones et al., 2026).

05Common Pitfalls

  • Running DRO with intervals too long for the baseline rate. If your interval is longer than the client's typical inter-response time, you are reinforcing the existing behavior rate rather than its absence. Set the initial DRO interval below the mean baseline IRT and thin from there Wilder et al. (2023).
  • Treating DRO as a behavior-reduction-only procedure. DRO inevitably reinforces something during the absence interval; the question is whether that something is clinically useful. Layer a defined desirable alternative onto DRO so the strengthened class is something you want to see more of Rey et al. (2020) Hangen et al. (2020).
  • Choosing arbitrary reinforcers for the alternative. If the alternative's reinforcer doesn't actually compete with the function maintaining the target — or is degraded relative to the natural reinforcer — DR effects are smaller and less durable, especially without extinction Iannaccone & Jessel (2023) Weinsztok & DeLeon (2022) Briggs et al. (2019).
  • Letting integrity drift after mastery. Brief intermittent consequence omissions shift FR1 into a variable-ratio schedule, making the target response (when it returns) more persistent — the opposite of what the team intended (Jones et al., 2026). Build consequence-fidelity probes into ongoing supervision, not just antecedent step checks.
  • Omitting schedule thinning from the plan. Most DR plans never write a thinning schedule, and behavior that "looks stable" gets treated as maintained. Thin slowly, never simultaneously thin two reinforcement dimensions, and have a layered DR component ready to insert if thinning produces re-emergence Nevin et al. (2016) Fritz et al. (2017) Briggs et al. (2019).
  • Abandoning DRA when target rate doesn't drop fast enough. Stubborn target rates usually mean the differential is too small, not that the variant is wrong. Increase magnitude and/or quality on the alternative before switching to DRO or layering on extinction Iannaccone & Jessel (2023) Briggs et al. (2019). If you must switch variants, run a brief parameter-sensitivity assessment first Johnson et al. (2017).
  • Running DRL without an S+. DRL with only an S− or no signals destabilizes responding rather than producing the targeted low rate; in some cases it eliminates the response entirely Becraft et al. (2018). Always pair DRL with explicit S+ and S− signals.
  • Ignoring resurgence risk during fade. Rich DRA schedules build behavioral momentum; when they end, the target response can resurge harder than baseline Nevin et al. (2016). Plan layered fade with brief DRO probes and context management King et al. (2025) Craig et al. (2018) Kimball et al. (2020).
  • Embedding asymmetric DR in assessment phases. When DR is part of an FA or comparison condition, asymmetric contingencies (alternative on extinction) confound the read of EO sensitivity and bias. Use symmetrical contingencies for assessment, then move to asymmetric DR for treatment (Allen et al., 2026).

06When to Refer Out

  • Suspected medical or biological substrate. Behavior with possible pain, GI, sleep, or seizure involvement, especially when the topography is severe or sudden in onset; pica with non-food items that could cause injury; or any topography producing tissue damage. Refer for medical evaluation before running any DR plan that requires evoking the behavior, and document the consult.
  • Persistent high-rate destructive behavior that is unresponsive to enriched DRA. When DRA with enhanced magnitude and quality (and, where feasible, extinction) does not produce clinically meaningful reductions across two thinning attempts, refer to a specialized BCBA team or inpatient behavior unit rather than continuing to escalate parameters in-place Briggs et al. (2019) Brown et al. (2020).
  • Resurgence pattern that won't resolve with stacked safeguards. When DRA + extinction + brief DRO probes + context management still produces clinically significant resurgence at every fade attempt, the case likely needs a specialist review of the underlying contingencies and possibly an experimental analysis of the resurgence parameters Brown et al. (2020) King et al. (2025) Craig et al. (2018).
  • Automatic reinforcement that does not respond to DRO + DRA + sensory enrichment. Refer to a setting with capacity for extended observation, latency-based metrics, and matched-stimulation programming Butler et al. (2021).
  • Resource ceiling. When neither in-person nor telehealth training brings staff to ≥80% procedural integrity on the DR variant in question after two cycles, refer the case to a regional consultation team rather than running an underpowered plan in-house (Halbur et al., 2024).
  • Active psychiatric crisis or imminent safety concern. DR is not the right first move; refer to licensed mental-health crisis services and resume the DR plan after stabilization.

07Future Research Directions

The corpus's operational claims sit on solid SCED and translational evidence, but several gaps have direct clinical consequences (Cividini-Motta et al., 2024).

The systematic review of DR in skill acquisition notes that no included study manipulated the immediacy parameter and that maintenance and generalization data after DR-driven mastery are sparse (Cividini-Motta et al., 2024). Prospective small-n studies that orthogonally manipulate immediacy against quality and magnitude, then track maintenance over months rather than weeks, would clarify which parameter combinations produce durable mastery in DTT (Cividini-Motta et al., 2024).

Head-to-head comparisons of DR variants beyond the DRA-vs-DRO and FM-vs-VM-DRO contrasts already in the literature would also help Hedquist & Roscoe (2020) Wilder et al. (2023). There is no clean DRA-vs-DRI head-to-head, no DRO-vs-DRA-vs-DRI three-arm for matched topographies, and only sparse DRL-vs-DRH evidence for rate-shaping rather than reduction Becraft et al. (2018). Each comparison has clear clinical relevance and requires no new methodology.

The resurgence and renewal literature — DRA + extinction, brief DRO during extinction, distinct-context delivery, ABA renewal protection — is dense in basic and translational work but thin in clinical implementation Brown et al. (2020) King et al. (2025) Craig et al. (2018) Kimball et al. (2020). A multi-site implementation study adopting all four safeguards as a packaged protocol and tracking resurgence rates across 12 months would convert a methodologically promising literature into a defensible clinical standard Brown et al. (2020).

The integrity layer also needs work. Halbur's enhanced-data-sheet finding is promising but small; a multi-site replication with standardized DR data sheets (one per variant) and consequence-fidelity probe schedules would let agencies adopt a coherent integrity protocol (Halbur et al., 2024). Jones and colleagues' post-mastery fidelity-decay finding is convincing inside a translational analogue but needs longitudinal field replication that tracks consequence-delivery accuracy across months in actual programs (Jones et al., 2026).

Finally, the choice and reinforcer-as-discriminative-stimulus work of King, Jimenez-Gomez, and Bland raises a question the field has not yet answered: how should the predictive properties of reinforcers be programmed and faded inside applied DR plans, beyond the binary signaled-versus-unsignaled comparison King et al. (2025) Jimenez‐Gomez et al. (2025) Bland et al. (2016)? Applied work that systematically varied the discriminative properties of DR signals — and traced the resurgence consequences of each — would move the field from "schedule selection" to "stimulus-control engineering" Bland et al. (2016).

08Practitioner Takeaways

  1. Define DR by parameter, not by procedural label. A DRA in which the alternative gets richer reinforcement along quality, magnitude, or immediacy is doing the actual work; extinction of the target is ideal but not required for the procedure to count Vollmer et al. (2020). When a plan stalls, ask which parameter the differential is failing on, not whether you have "real DRA."
  2. Begin DR at the very first teaching trial. Do not wait for errors to emerge or run several non-differential sessions first; the cost of "wait-and-see" is measured in trials Campanaro et al. (2020).
  3. For skill acquisition, escalate the differential. Reserve the most potent reinforcer for independent responses and give only praise (or nothing) for prompted responses; the more differential, the faster the acquisition (Cividini-Motta et al., 2024).
  4. Front-load magnitude and quality on the alternative response. Both as the primary mechanism of DRA and as a buffer against the integrity errors that will inevitably occur during real implementation Iannaccone & Jessel (2023) Weinsztok & DeLeon (2022).
  5. Pair DRA with extinction when feasible, and stack four resurgence safeguards when not. Extinction reduces both ongoing target responding and later resurgence at equated reinforcement rates Brown et al. (2020). When extinction isn't feasible, layer in brief DRO probes during fade King et al. (2025), deliver alternative reinforcement in a distinct context Craig et al. (2018), and carry the DRA contingency back into the original problem context Kimball et al. (2020).
  6. Always layer DRA onto DRO. DRO inevitably reinforces something; defining and programming a specific desirable alternative ensures that the strengthened class is clinically useful rather than incidental Rey et al. (2020) Hangen et al. (2020).
  7. For DRO interval selection, default to variable-momentary unless training resources are limited. VM-DRO and FM-DRO are roughly equally effective, but VM resists discrimination and is preferred by caregivers; FM is the easier-to-implement fallback Wilder et al. (2023).
  8. Never run DRL without explicit S+ and S− signals. Without an S+, DRL destabilizes or eliminates the response rather than producing the targeted low rate Becraft et al. (2018). For class-wide rate reduction, an interdependent group contingency tied to a shared reinforcer is a defensible first-line tool Ré et al. (2024).
  9. Plan schedule thinning before treatment effects look settled. Thin slowly, never simultaneously thin two reinforcement dimensions, and have a layered DR component ready to insert the moment thinning produces re-emergence Nevin et al. (2016) Fritz et al. (2017) Briggs et al. (2019).
  10. Build three high-quality alternatives into DR active phases when possible. Choice itself functions as a DR parameter; three alternatives outperform two even at equated reinforcement Kestner et al. (2023).
  11. Run a brief DR-parameter assessment for new skills. Quality, magnitude, and schedule variants of DR can produce different acquisition rates; an alternating-treatments assessment within the same skill is a small upfront cost that pays off quickly Johnson et al. (2017). Reassess for novel skills — predictive validity does not always travel.
  12. Use NCR (without extinction) to abolish the EO when the case stalls before any DR contingency works. Layer DRA on top once NCR has stabilized behavior, and add DRA the moment thinning produces re-emergence Fritz et al. (2017).
  13. Use distributed escape rather than accumulated escape for escape-maintained problem behavior. Brief, frequent breaks (FR1) outperform accumulated longer breaks (FR15 → 7.5-min); accumulation does not produce clinically acceptable reductions for many learners Fulton et al. (2020).
  14. Build consequence-fidelity probes into ongoing supervision. Antecedent-step checks miss post-mastery integrity decay; intermittent consequence omissions accidentally make the target response more persistent (Jones et al., 2026).
  15. Add a "DR" column to your program data sheets. A visible tracking artifact acts as a discriminative stimulus for staff, increasing contingent reinforcement delivery without extra observation time (Halbur et al., 2024).
  16. For staff training on DRO/DRA/DRI, use equivalence-based instruction with at least two exemplar vignettes per procedure. EBI outperforms passive lecture and generalizes better to novel cases Kelly‐Sisken et al. (2025).

09Frequently Asked Questions

What's the difference between DRO, DRA, DRI, DRL, and DRH?

All five are differential reinforcement procedures — reinforcement is delivered contingent on one class of responses while withheld or minimized for another. DRO reinforces the absence of a target response across an interval; the "other" behavior is anything except the target Rey et al. (2020) Hangen et al. (2020). DRA reinforces a defined alternative response that simply has to be different from the target Vollmer et al. (2020). DRI is DRA with an added topographical constraint — the alternative must be physically incompatible with the target. DRL reinforces the target response only when it occurs at or below a target rate (used to reduce, not eliminate, an otherwise appropriate response) Becraft et al. (2018). DRH is the symmetric opposite of DRL — reinforce only when responding meets or exceeds a target rate (used to build response rate). NCR sits adjacent to the family — reinforcers are delivered on a time-based schedule independent of behavior, so there's no differential, just an EO manipulation Fritz et al. (2017).

Do I have to use extinction with DRA?

No. The current definitional position is that DRA is DRA when reinforcement is delivered differentially along at least one parameter (quality, magnitude, immediacy, or rate) — extinction is ideal but not required Vollmer et al. (2020). Translational and clinical evidence shows DRA without extinction works when the differential along magnitude, immediacy, and quality is large enough to compete with the function maintaining the target Iannaccone & Jessel (2023) Briggs et al. (2019). The trade-off: DRA with extinction produces less resurgence at equated reinforcement rates, so when extinction is feasible it's the strongest single safeguard Brown et al. (2020).

When should I use DRA vs DRO?

The cleanest direct test in the corpus is Hedquist and Roscoe's three-child alternating-treatments comparison: when neither procedure used response blocking, DRA produced larger and more consistent reductions in automatically reinforced stereotypy and bigger gains in task engagement than DRO Hedquist & Roscoe (2020). The general rule: prefer DRA when you need a specific alternative response taught (skill replacement, task engagement, communication); use DRO when no suitable alternative is yet available or when the target rate is severe/low and you need to suppress before you can teach. Always layer DRA onto DRO when you do choose DRO, so the adventitious reinforcement DRO produces is directed somewhere clinically useful Rey et al. (2020) Hangen et al. (2020).

How do I pick the DRO interval?

Set the initial interval below the mean baseline inter-response time so the first interval is achievable, then thin from there Wilder et al. (2023). Use momentary sampling (response absent at the moment the interval ends) rather than whole-interval (response absent for the entire interval) when implementation resources are constrained. Default to variable over fixed intervals when the topography is reliable enough for the client to time a fixed schedule — VM-DRO and FM-DRO are roughly equally effective at suppression but VM resists discrimination Wilder et al. (2023).

Is FCT just DRA?

FCT is a specific implementation of DRA in which the alternative response is explicitly a communication response — a mand, sign, picture, or device exchange — and the reinforcer for the mand is the same reinforcer that maintains problem behavior. The resurgence and renewal logic that applies to DRA applies to FCT, and the FCT signal itself (the card, device, or modeled sign) functions as the discriminative stimulus whose presence/absence governs resurgence patterns once treatment is faded Bland et al. (2016) Nevin et al. (2016). Choose FCT when the function is socially mediated and the client has (or can quickly acquire) a matched communication topography; choose generic DRA when the alternative is non-communicative.

How do I prevent resurgence when I fade DR?

Stack four safeguards: (1) pair DRA with extinction during the active treatment phase when feasible Brown et al. (2020); (2) insert a brief 3-s DRO contingency during early extinction sessions and fade it once resurgence risk passes King et al. (2025); (3) deliver alternative reinforcement in a context distinct from the original problem context Craig et al. (2018); (4) carry the DRA contingency back into the original problem context to blunt ABA renewal Kimball et al. (2020). Each is procedurally cheap and stacks with the others.

Why does my DRO seem to "work" but odd new behaviors keep appearing?

Because DRO is reinforcing something during the absence interval — and unless you've defined a specific desirable alternative, that something is whatever the client happens to be doing when the interval ends Rey et al. (2020). Adventitious reinforcement of unspecified "other" responses is the mechanism, not a side effect Hangen et al. (2020). Define and program a specific desirable alternative — i.e., layer DRA onto DRO — to direct the strengthening toward responses you actually want.

What should I do if DRA without extinction isn't reducing the target enough?

The differential is usually not large enough — not the variant. Increase magnitude and/or quality of the reinforcer for the alternative response before switching to DRO or layering on extinction Iannaccone & Jessel (2023) Briggs et al. (2019). Combining magnitude, immediacy, and quality enhancements produces the largest and most consistent suppression in the translational data; pick the dimensions the client is most sensitive to and run a brief assessment if you're unsure Iannaccone & Jessel (2023) Johnson et al. (2017).

How do I run a DRL so that I reduce the rate without eliminating the response?

Always pair the schedule with explicit S+ (reinforcement available) and S− (reinforcement unavailable) signals — never run DRL with only an S− or no signals Becraft et al. (2018). For class-wide rate reduction (vocal disruption, transition lateness, calling out), an interdependent group contingency tied to a shared reinforcer like extra recess is a low-burden first-line tool Ré et al. (2024).

When does NCR fit in?

Use NCR when the target appears motivation-driven and you need to abolish the EO before any DR contingency will work — or when extinction is impossible and you want to keep delivering the maintaining reinforcer non-contingently while teaching a replacement Fritz et al. (2017). Layer DRA on top of NCR before thinning, because NCR alone often re-evokes problem behavior during thinning unless a contingent reinforcement contingency is also operating Fritz et al. (2017).

10References

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