Schedules of Reinforcement: A Practitioner's Guide to CRF, FR, VR, FI, VI, Compound Schedules, and Schedule Thinning

A schedule of reinforcement is the rule that specifies which occurrences of a response produce a reinforcer. The four basic schedules are building blocks; nearly everything used in applied practice is either one of them or a defined combination of two or more.

CRF: Continuous Reinforcement

Every occurrence of the target response is reinforced. CRF is the schedule of choice for establishing a brand-new response, since the correlation between response and reinforcer is at its clearest under continuous delivery, and behavioral control tracks that response-reinforcer correlation directly, not merely how close in time the two events fall (Kuroda et al., 2018). The tradeoff is that CRF also satiates a learner fastest of any schedule and produces the least extinction-resistant responding of the basic four, which is exactly why acquisition programs move off it as soon as the response stabilizes.

FR: Fixed Ratio

Reinforcement follows every nth response, counted rather than timed. FR schedules generally produce a high rate of responding, but a requirement that outpaces the learner's endurance produces ratio strain, a break in responding or a full stop before the count is met. Reinforcer quality is one lever for managing that: in basic research with domestic hens, raising reinforcer quality increased behavioral activation under fixed-ratio requirements (Bruce et al., 2019), a principle that, where it carries into applied work, would soften the practical cost of a leaner ratio. Post-reinforcement pausing is FR's other signature feature. In the same basic literature with nonhuman subjects, both the size of the reinforcer just delivered and whether the next one is signaled changed how long that pause ran (Young et al., 2017), a mechanism worth holding in mind, as a working hypothesis rather than a settled human finding, when reading whether a learner's pause reflects fatigue, satiation, or simply tracking the schedule's structure.

VR: Variable Ratio

Reinforcement follows an average number of responses that varies unpredictably around a set mean, so the learner cannot count toward a known target. VR schedules typically produce steadier, higher-rate responding than FR at a comparable mean requirement, and VR-maintained behavior tends to be more persistent under extinction or thinning. That persistence traces to reinforcement rate specifically rather than reinforcer magnitude, which is one reason a thinning plan should protect the rate of the richer schedule component rather than simply upsizing the reinforcer to compensate (Costa et al., 2025).

FI: Fixed Interval

Reinforcement becomes available for the first response after a fixed span of time has elapsed; responding earlier does nothing. FI schedules classically produce a scalloped pattern, little or no responding early in the interval, accelerating as the interval's end nears, though the pattern depends on the learner having some way to discriminate elapsed time. Multiple and chained arrangements often build in FI-based components deliberately, precisely because the low early-interval response rate is a known, plannable feature rather than a flaw to engineer out.

VI: Variable Interval

Reinforcement becomes available after an average span of time that varies unpredictably around a set mean, so unlike FI, the moment of availability can't be tracked. VI schedules produce a moderate, steady response rate without FI's scallop, which is what makes VI useful as a foundation for stable baselines in both applied and experimental work. VI's own value may bear on how durable responding is under disruption. A laboratory demonstration in rats deprived of REM sleep illustrates the general principle: thinner VI schedules left responding more vulnerable to the disruption, while richer VI values protected response rate under the same challenge (Kirby et al., 2003). Establishing control under a synchronous, rule-stated VI contingency also depends on the accuracy of the rule delivered before the session; inaccurate presession rules degraded the schedule control that accurate rules established cleanly (Sheridan et al., 2025).

Concurrent, multiple, mixed, and chained schedules

Beyond the four basic schedules, applied programs commonly run compound arrangements. Concurrent schedules offer two or more response options at the same time, each on its own schedule, and allocation across them is governed by relative, not absolute, reinforcement rate; boosting absolute reinforcement rate alone did not improve human sensitivity to the relative-rate differences that actually drive choice in a matching-law preparation (Morris et al., 2025). Multiple and mixed schedules alternate two or more components across time, distinguished by whether each component carries its own discriminative stimulus (multiple) or not (mixed). Chained schedules link several components end to end, with completion of an earlier link, often marked by a conditioned reinforcer, setting the occasion for the next. Restricting an available response option in a multiresponse environment does not reliably send responding to one predictable alternative. In basic research with rats, measured time reallocation rather than an assumed next-most-probable response was what tracked where responding went (Lyons et al., 1984), a caution against assuming the substitute response can be predicted in advance.

Schedule thinning

Thinning moves a schedule from denser to leaner over the course of treatment, and how the thinning is structured matters as much as the end value. Distributing access to reinforcement across time, versus letting a learner accumulate several responses' worth and take it all at once, can work equally well, and learners sometimes prefer the accumulated format over the frequent, small-delivery arrangement that intuition points to first. That gives a thinning plan a second lever beyond simply raising the ratio or interval requirement (DeLeon et al., 2014).

The decision a BCBA actually faces with schedules is rarely which basic schedule exists in the abstract, it's what values to set and how to move them as a program advances. Start acquisition on CRF or a very dense FR or VR to build the response and establish a clean correlation between it and the reinforcer, then move off continuous delivery as soon as the response is reliable, since that same clean correlation is also what makes CRF the most extinction-fragile of the basic schedules.

Choose ratio schedules when the target is a countable, discrete response and rate of output is the clinical goal; choose interval schedules when the target is closer to sustained engagement or when counting individual responses isn't the point. Within ratio schedules, default to VR over FR once the response is established and durability matters, since VR's unpredictable requirement produces steadier output, and its persistence traces to reinforcement rate rather than reinforcer size. That distinction should also guide how a schedule is protected during thinning: hold the rate of the richer, protected component up rather than assuming a larger reinforcer will substitute for it (Costa et al., 2025).

Be honest about what this literature does and doesn't hand a clinician. Most of the direct empirical support here comes from tightly controlled single-case or laboratory preparations, several of them run with nonhuman subjects, testing matching-law sensitivity, ratio strain, and pause structure, rather than from applied group-design trials validating a specific clinical thinning protocol. Where a finding rests on animal work, treat it as a basic principle to carry into practice as a hypothesis, not as direct clinical guidance. Roughly half the studies reporting a clear direction found positive effects, and a genuine minority found null, mixed, or inconclusive results, so a specific parameter choice, exactly how many responses to add per FR step or how much to widen a VI's range, is closer to informed clinical judgment applied to general principles than to a settled finding. Where the evidence is strongest is at the level of those general principles: presession rule accuracy matters for establishing schedule control at all (Sheridan et al., 2025), and reinforcer deprivation, not just schedule value, has to be in place for reinforcement to function as reinforcement in the first place (Konarski et al., 1980).

Once a schedule is set, execution comes down to how it's monitored session to session and what a clinician does when responding doesn't match the textbook pattern.

Before trusting any schedule value on a data sheet, confirm the reinforcer is actually functioning as a reinforcer that day. A preferred activity delivered without any period of restricted access beforehand may not be reinforcing in that moment regardless of the schedule wrapped around it, and treating "high-preference" as a stand-in for "currently deprived" is a common way a program looks like a schedule problem when it's really a reinforcer-access problem (Konarski et al., 1980).

Watch post-reinforcement pausing on FR programs as data, not noise. A pause that lengthens across sessions can reflect ratio strain building toward a break in responding. In basic research with nonhuman subjects, both reinforcer size and whether the next reinforcer was signaled changed how long that pause ran (Young et al., 2017); treated as a working hypothesis rather than a settled human finding, that points toward checking reinforcer magnitude and signaling before increasing the ratio requirement, instead of assuming the learner simply needs to push through.

Whenever a spoken or written rule accompanies the contingency, whether it's "finish this, then you get the tablet" or a visual schedule card, make sure the rule matches the actual contingency exactly. Inaccurate presession rules degrade schedule control that an accurate rule establishes cleanly, so a rule that oversells or undersells what's coming is worth fixing before troubleshooting anything else about the schedule itself (Sheridan et al., 2025).

Building a concurrent arrangement calls for measuring allocation directly rather than predicting it. In basic research with rats, removing or restricting one response option did not reliably send responding to a single obvious alternative (Lyons et al., 1984), so track where responding actually goes, not where it was assumed to go, especially early in a new arrangement.

When thinning any schedule, decide upfront whether the plan moves toward a leaner ratio or interval value, toward a more distributed or more accumulated access format, or both, and log that decision explicitly in the plan. Distributed and accumulated formats can perform equally well, and a learner's own preference between them is worth checking directly rather than assumed on the clinician's behalf (DeLeon et al., 2014).