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

Differential reinforcement is a family, not a single procedure: reinforce one class of behavior (or its absence) while withholding reinforcement from another. The variants differ in what, exactly, gets reinforced.

DRO — Differential Reinforcement of Other Behavior

DRO delivers reinforcement contingent on the target behavior's absence during an interval, and resets or withholds delivery if the behavior occurs. It is the only member of the family that reinforces a non-event rather than a specific response, which is both its strength and its limit. DRO doesn't require identifying or teaching a replacement, which makes it definable even when a clean alternative response is hard to specify or the target is topographically diffuse. The tradeoff is structural: DRO teaches the learner nothing to do instead, and it risks adventitiously reinforcing whatever happens to be occurring at interval's end, since "not the target behavior" describes an unlimited set of other responses (Jessel et al., 2015).

DRA — Differential Reinforcement of Alternative Behavior

DRA reinforces a specific, defined alternative response that is not physically incompatible with the target but produces access to a similar or the same reinforcer. Functional communication training is DRA's most common clinical application: teach the mand that gets the same outcome the problem behavior used to get, and reinforce it richly while the target extinguishes or moves to a leaner schedule (Drifke et al., 2020). Its main structural vulnerability is efficiency: if the alternative response takes more effort, more time, or produces a smaller or slower reinforcer than the problem behavior did, the learner keeps the behavior that used to work better (Weinsztok et al., 2022).

DRI — Differential Reinforcement of Incompatible Behavior

DRI is a special case of DRA where the alternative response is physically incompatible with the target, meaning both cannot occur at once. Sitting on hands is incompatible with hitting; keeping materials in a mouth is incompatible with spitting them out. Because the incompatible response occupies the same motor channel as the target, it functions as a built-in block without external physical intervention, but only for targets with a simple, single motor topography where a true incompatible response exists.

DRL — Differential Reinforcement of Low Rates

DRL reinforces the target behavior itself, but only when it occurs at or below a set rate, whether measured full-session, interval by interval, or as a minimum inter-response time (spaced-responding DRL) (Becraft et al., 2018). It fits behavior that isn't inherently dangerous and has a socially acceptable form, like frequent questions or bids for attention, where the goal is moderation rather than elimination. DRL is a poor fit for anything safety-critical, since it deliberately tolerates continued occurrence at a reduced rate rather than working toward zero.

DRH — Differential Reinforcement of High Rates

DRH reinforces the target when it occurs at or above a criterion rate, and is the direction to reach for when the behavior is desirable but underoccurring, such as on-task responding, work completion, or vocalization rate for a learner building fluency. In practice DRH often shares mechanics with a percentile-schedule shaping program, since both raise a rate criterion against the learner's own recent performance.

NCR — Noncontingent Reinforcement

NCR belongs in this family as the contrast case: it is deliberately non-differential. A functionally matched reinforcer is delivered on a fixed-time or variable-time schedule, independent of what the learner is doing, which breaks the response-reinforcer contingency that maintained the problem behavior rather than reinforcing anything specific instead of it. Because no discrimination is trained, satiation is the main structural risk, and the schedule usually needs to thin over time just like a differential one.

The decision a BCBA actually faces isn't whether to use differential reinforcement, it's which variant, and how hard to protect it once you thin. Start with function. A functional analysis, or at minimum a well-supported functional hypothesis, should exist before you pick a variant, not after. DRA and DRI are function-matched by design, and choosing one without knowing what maintains the behavior is a guess dressed up as a plan.

Default to DRA when a teachable alternative response exists that can access the same or a comparable reinforcer. This is the most common right answer in outpatient and school-based ABA. Reserve DRI for behavior with a clear, simple motor topography where an incompatible response is already close to fluent, particularly automatically maintained stereotypy where blocking isn't feasible or humane on its own. Fall back to plain DRO only when no clean alternative response can be defined or the target is too diffuse to pin down a single incompatible or alternative form, and know going in that DRO is the least durable member of the family once you start thinning.

Reach for DRL or DRH when the goal is rate moderation rather than elimination or acquisition, not as a weaker substitute for DRA. Treat NCR as a companion, not a competitor: layer it under a DRA plan when treatment integrity is genuinely at risk (a busy classroom, inconsistent caregivers) or you need an immediate reduction before staff can reliably catch and reinforce a specific alternative response.

Whatever you choose, the field's most common misstep is picking the variant that's easiest to write into a plan rather than the one that fits the function. A specific, teachable alternative response almost always outperforms reinforcing an absence, and the literature backs that as a direct finding rather than a theoretical preference (Hedquist et al., 2020). If extinction of the target behavior is part of the plan, treat the reinforcement side as the primary intervention and extinction as the supporting piece: differential reinforcement paired with extinction is both more humane and better supported than extinction run alone.

Once you've picked a variant, what determines whether it holds up in a real caseload is operational, not conceptual: how precisely the target and alternative are defined, how staff sample and score in the moment, and how the contingency is protected once a session gets busy.

If you're running DRO, decide upfront whether staff score it on a fixed-momentary or variable-momentary basis, checking for the target's absence at the same point in every interval versus at a randomized point each time. Fixed-momentary is easier to train and run with less experienced staff; variable-momentary is harder to game and tends to land better with caregivers who are watching for consistency. Both reduce automatically reinforced behavior about equally, so pick based on who is implementing it and how closely they'll be watched, not on an assumption that one is clinically superior (Wilder et al., 2023).

Protect the schedule during integrity lapses, don't just write a stricter plan. Staff will miss reinforcement opportunities. Building in a larger or higher-quality reinforcer for the alternative response gives the contingency more room to survive an occasional missed delivery, which matters more in real clinics than in a tightly controlled single-subject graph (Weinsztok et al., 2022).

Plan for resurgence and renewal before you thin, not after you see it. Two mitigation tactics have direct support: pairing the DRA context with a salient environmental cue distinct from the DRO context roughly halves resurgence (Craig et al., 2018), and fading the treatment context back toward the natural environment in deliberate steps (one stimulus at a time, not all at once) cuts renewal similarly (Jackson et al., 2026). Build the fade plan into the treatment plan from day one rather than reacting to a relapse after discharge or a setting transition.

If your differential reinforcement targets negatively reinforced behavior (escape, compliance with an aversive routine like protective equipment wear), the same logic holds: shape the criterion gradually with small, defined interval increases rather than large jumps, and hold each step until it's stable before advancing (Wheatley et al., 2020). Run generalization probes across staff and settings once the criterion holds in the original setting; a routine that only survives with one trained clinician isn't done yet (Wheatley et al., 2020).