Withdrawal design

Withdrawal is a condition change

In a withdrawal phase, the independent variable is discontinued or returned toward a planned comparison condition while the dependent variable continues to be measured. If responding also moves toward its earlier pattern, the change can strengthen the case that the intervention mattered.

Common sequences include:

• A-B-A: baseline, intervention, then withdrawal.
• A-B-A-B: baseline, intervention, withdrawal, then reintroduction.
• B-A-B: an intervention already in place, a justified withdrawal, then restoration.

An A-B-A design contains a withdrawal, though it usually supplies fewer demonstrations and ends without the intervention restored. ABAB adds a reintroduction and lets the outcome reproduce under B. When B is effective and acceptable, ending in B often fits the person's interests better.

The WWC Single-Case Design Technical Documentation treats withdrawal or reversal designs as repeated phase arrangements used to demonstrate an effect at different points in time. Under its three-demonstration convention, the design needs at least three phase changes and four phases before additional observation rules can support its rating.

The labels overlap in research

Authors frequently write reversal or withdrawal design as one family. “Withdrawal” names the removal manipulation. “Reversal” highlights the broader logic that the outcome repeatedly changes with the conditions. Neither label reveals the full procedure.

A paper may call an ABAB sequence a withdrawal design because B is removed in A2. Another may call the same sequence a reversal design because responding is expected to reverse. Report A1, B1, A2, and B2, define each condition, and state what was removed. That description is more useful than debating the preferred label.

A health-research overview and a 2024 rehabilitation review both describe this design family through repeated condition manipulation and within-case replication. They also identify reversibility and practical acceptability as central selection issues.

Use withdrawal only when change can reverse

The outcome should respond within a reasonable phase length, and the intervention's effect should be able to recede when the condition changes. An immediately available prompt, cue, device setting, or staff response may meet that test in a low-risk context.

Learned skills often persist after teaching ends. Surgery, long-acting medication, cumulative instruction, and durable environmental changes may also have lasting effects. In those situations, A2 may fail to resemble A1 even when B produced a real benefit. A multiple-baseline, multiple-probe, changing-criterion, or another suitable design may fit better.

Carryover also matters. Practice, fatigue, satiation, emotional responding, or stimulus-control history can extend beyond B. A2 follows treatment and may differ from A1 because of sequence or history. Record latency and relevant context, and temper the conclusion when the outcome changes slowly or incompletely.

Define removal and stop rules in advance

Specify what remains available in A2. “Treatment withdrawn” is too vague. Identify materials, prompts, consequences, staff actions, dose, schedule, and ordinary supports. A comparison condition can remove the experimental component while preserving access and safety.

Predefine minimum observations, phase-change criteria, maximum duration, rescue thresholds, and authority to stop. Include what happens if the person withdraws assent, shows distress, experiences deterioration, or asks to resume the preferred condition. Record deviations and reasons.

The SCRIBE 2016 reporting guideline calls for clear phase sequences, decision rules, planned replication, procedural changes, fidelity, raw outcomes, adverse events, and early stopping. Transparent reporting cannot make an unsuitable withdrawal acceptable, but it exposes how the evidence was produced.

Measure implementation and outcome separately

Define the dependent variable, opportunity, observation window, prompts, exclusions, and missing-data treatment. Graph every observation with true phase boundaries. Phase averages alone can hide trend, variability, overlap, delayed effects, and one influential point.

Measure procedural integrity for both A and B. If staff continue part of B during A2, the intended withdrawal did not occur. If staff, setting, schedule, equipment, or measurement change with the phase, those differences can compete with the intervention as explanations. Sample observer agreement when scores influence design decisions.

The current WWC handbook page identifies Version 5.0 as current. Meeting its phase-count and observation rules is one research-rating question. Clinical usefulness also depends on fit, meaningful outcomes, implementation, acceptability, and evidence beyond the graph.

A fictional low-risk withdrawal example

Samira is a fictional employee who chooses to test a removable electronic packing checklist during supervised practice with nonsensitive supplies. Each phase contains five simulated orders with four required steps, creating 20 scored steps. Training, supervision, equipment, time, and the error-correction procedure stay constant.

With the ordinary paper work instruction in A1, Samira completes 11 of 20 steps independently. The electronic checklist is added in B1, and performance reaches 18 of 20. During an agreed brief withdrawal to the paper instruction in A2, performance is 12 of 20. The checklist returns in B2, and performance is 19 of 20. Staff implement the assigned condition correctly in 20 of 20 orders.

The outcome tracks three condition changes and replicates under both conditions. With credible measurement and stable context, the pattern supports a functional relation for this checklist, task, and setting. It does not establish lasting learning, transfer to actual client records, or effectiveness for another employee. Samira's preference for the checklist remains a separate reason to retain it after the comparison.

Ethics may rule withdrawal out

Do not withdraw AAC, food, water, bathroom access, mobility, prescribed care, pain care, rest, emergency help, or effective safety supports. Avoid withdrawal when it could increase severe injury, crisis, medical instability, loss of essential communication, or another unacceptable harm. A physician or other authorized prescriber governs medication changes within that professional's scope.

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 ongoing evaluation. A publishable graph never outranks the person's welfare or right to stop.