Reversal design

Repeated condition changes test the same relation

In a basic ABAB arrangement:

1. A1: repeated baseline or comparison observations support a prediction.
2. B1: the independent variable is introduced and the outcome is measured.
3. A2: the comparison condition returns to test whether the outcome moves toward its earlier pattern.
4. B2: the intervention returns to test whether the effect replicates.

The WWC Single-Case Design Technical Documentation describes prediction, repeated manipulation, and visual analysis as foundations for single-case causal inference. Each phase transition creates an opportunity to compare level, trend, variability, immediacy, overlap, and consistency.

An A-B sequence shows that change followed intervention onset. The A2 and B2 phases make history or maturation less plausible when responding repeatedly tracks condition. Replication strengthens the inference only when the independent variable is implemented faithfully and other important features remain comparable.

Reversibility is the first design gate

The dependent variable should respond quickly enough to the condition change and be able to move toward its prior pattern. Reversible examples may include performance supported by an immediately available cue, equipment feature, or consequence arrangement. The expected latency and carryover should fit the phase length.

Durable learning makes return to baseline unlikely. Once someone acquires a skill, removing the teaching procedure may leave the skill intact. Long-acting medication, surgery, environmental renovation, and cumulative training also resist rapid reversal. A failed A2 phase can reflect persistence of benefit rather than absence of an original effect.

A review of single-case designs in health research describes reversal or withdrawal designs as repeated-measure arrangements that replicate conditions and effects within a case. A 2024 rehabilitation review likewise identifies a reversible dependent variable as a core fit consideration.

Withdrawal and reversal terminology overlaps

Research sources often combine the terms reversal and withdrawal. A withdrawal arrangement specifically removes or reduces the independent variable and observes what follows. “Reversal” emphasizes the repeated condition sequence and expected change in the dependent variable. An ABAB design commonly does both.

State the actual phases instead of relying on the label. A design might compare A-B-A-B, B-C-B-C, or another justified sequence. The interpretation depends on which condition changed, what remained available, and whether the outcome could reverse.

Ending in an effective and acceptable intervention phase is often preferable. The final B phase also supplies a replication. A person's current preference, risk, service access, and clinical need govern the endpoint.

Phase rules and stopping rules should be prospective

Define the minimum exposure, phase-change criteria, maximum acceptable duration, and rescue or stop conditions before data collection. A phase should last long enough to assess its pattern while remaining within the person's agreed and safe limits.

Avoid changing phases solely when the most persuasive data point appears. That can bias the graph. Record early transitions and their reasons, including assent withdrawal, distress, clinical deterioration, external service changes, or technical failures.

The SCRIBE 2016 statement calls for reporting the phase sequence, decision rules, procedural changes, planned replication, intervention delivery, fidelity, raw outcomes, adverse events, and early stopping. It is a reporting guideline rather than a design-selection rule.

Carryover and sequence can weaken reversal

Carryover occurs when B continues to influence responding in A2. Practice, satiation, emotional responding, fatigue, or learned stimulus control may persist. Sequence effects arise because A2 follows treatment and may differ from A1 for that reason alone.

Plan transition intervals when they are safe and scientifically justified. Preserve the true time sequence and report latency. A return toward baseline need not match A1 point for point, though persistent trend, high overlap, or delayed change should temper the conclusion.

Repeated exposure can also change acceptability. Ask the person about burden and preference throughout the design. A technically reversible outcome does not make repeated switching acceptable.

Measure the condition and outcome separately

Define the independent variable for replication, including materials, timing, implementer actions, dose, and comparison condition. Measure procedural integrity in every phase. Define the dependent variable, opportunity, observation window, prompts, exclusions, and missing-data rule. Train and calibrate observers and sample agreement where appropriate.

Context logs should cover staff, setting, schedule, health, medication, access, and unusual events. A staff change aligned with B or a measurement change aligned with A creates a rival explanation. Graph raw observations and phase boundaries rather than relying only on phase averages.

The current WWC handbook page identifies Version 5.0 as its current standards. Its three-demonstration convention requires at least three phase changes in a reversal or withdrawal design, producing at least four phases. Additional observation rules determine the research rating. These thresholds cannot rescue a confounded or unsafe comparison.

A fictional reversal example

Jordan is a fictional adult who chooses to test orientation labels during a preferred, low-risk makerspace cleanup routine. Four matched opportunities occur in each phase, with five bins scored per opportunity. Communication access, staff, materials, session length, and safety supports remain constant. Jordan can end the comparison at any time.

Without labels, Jordan places 8 of 20 items correctly in A1. With labels, the result is 18 of 20 in B1. During the agreed brief return to unlabeled bins, Jordan places 9 of 20 correctly in A2. Labels return and performance reaches 19 of 20 in B2. The assigned condition is implemented correctly in 16 of 16 opportunities.

The outcome moves with three condition changes and similar conditions produce similar results. With credible raw data, measurement, context, and fidelity, the pattern supports a functional relation for this label package and routine. It establishes neither generalized independence nor benefit for another person. Jordan's preference and everyday maintenance remain separate outcomes.

Safety and dignity can rule the design out

Avoid reversal when removing benefit could increase serious injury, crisis, medical instability, communication loss, or other unacceptable harm. Preserve AAC, food, water, bathroom access, mobility, prescribed care, pain care, rest, emergency help, and effective safety protections. Multiple-baseline, multiple-probe, or another design may offer a safer route.

The current BACB Ethics Code applies to BCBA and BCaBA certificants and people who completed an application. It addresses competence, client involvement, informed consent and assent when applicable, medical needs, risk, data, and evaluation. Consent or assent withdrawal and changing risk can override a planned replication.