Skip to content
Back to search results

ABAB Design — Reversal / Withdrawal Design

Introduced by
Murray Sidman; Baer, Wolf & Risley (applied behavior analysis formalization)
Year
1960s (Sidman 1960; Baer et al. 1968)
Type
Single-subject experimental design
Family
Experimental design
Font size

What does it do?

The ABAB design is a single-subject experimental methodology that establishes causal control by repeatedly introducing and removing an intervention.

A baseline phase (A) is followed by an intervention phase (B), then a return to baseline (A), and a second intervention phase (B), allowing the researcher to demonstrate that observed behavior changes are produced by the intervention rather than by coincidental factors.

The intuition behind the method

Imagine you want to show that a reward system genuinely improves a student's on-task behavior — not merely that behavior happened to improve over time. The ABAB design achieves this by turning the intervention on, then off, then on again. If behavior improves each time the intervention is present and reverts each time it is removed, the logical inference is that the intervention is responsible for the change. The repeated demonstration of control across the four phases is what distinguishes this design from weaker single-subject alternatives such as the AB or ABA designs.

History

The reversal design traces its conceptual roots to B. F. Skinner's operant conditioning laboratory methods of the 1930s–1950s, in which individual organisms' behavior was studied under experimentally controlled schedules. Murray Sidman codified single-subject experimental logic in his landmark 1960 text Tactics of Scientific Research. The ABAB variant became the cornerstone of applied behavior analysis (ABA) following the foundational paper by Baer, Wolf, and Risley (1968) in the inaugural issue of the Journal of Applied Behavior Analysis, which established repeated measurement and demonstration of experimental control as defining features of the field.

When to use it

Use the ABAB design when (1) you want to demonstrate a causal relationship between a specific intervention and a behavior for an individual participant, (2) the target behavior is expected to reverse when the intervention is removed (i.e., the behavior has not been permanently learned), and (3) repeated measurement of the same outcome variable across all phases is feasible. It is particularly well-suited to applied behavior analysis, special education, clinical psychology, and rehabilitation research. Do NOT use it when the target behavior is unlikely to reverse (e.g., an acquired skill such as reading a word), when the intervention cannot ethically be withdrawn (e.g., life-sustaining treatment or crisis intervention), or when the research question concerns group-level generalization rather than individual-level functional control.

Limitations and considerations

  • External validity is limited: findings describe one individual and may not generalize to other participants, settings, or behaviors without systematic replication across cases.
  • Requires that the target behavior reverses when the intervention is withdrawn; non-reversible behaviors (skills, habits) are unsuitable.
  • Withdrawing an effective intervention raises ethical concerns when participants depend on the treatment for well-being or safety.
  • Carryover and history effects can prevent a clean return to baseline, complicating causal interpretation.
  • Phase length and the number of data points needed for stability are judgment calls that introduce researcher discretion.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Provides a strong within-subject demonstration of a causal (functional) relationship without requiring a control group.
  • Requires only one participant or a small number of participants, making it feasible when large samples are unavailable.
  • The double replication across two A–B pairs substantially strengthens internal validity compared to simpler AB or ABA designs.
  • Repeated measurement yields a dense, longitudinal data record that captures the dynamic of behavior change over time.
  • Visual analysis methods are transparent and widely understood in behavioral and clinical research communities.

Applications

  • Evaluating behavioral interventions (reinforcement schedules, token economies) for children with autism spectrum disorder or intellectual disabilities.
  • Assessing the effect of classroom management strategies on on-task behavior or disruptive conduct in special education settings.
  • Testing the efficacy of physical or occupational therapy protocols on motor performance in rehabilitation research.
  • Examining the impact of biofeedback or cognitive-behavioral techniques on observable symptoms such as tics, anxiety behaviors, or pain-related avoidance.
  • Measuring the effect of environmental or ergonomic modifications on worker productivity or safety behaviors in organizational settings.

How it works

How it works

1

During the first A phase, the target behavior is measured repeatedly under natural or standard conditions without any intervention. These data establish a stable baseline pattern — the pre-intervention level and trend — against which subsequent change will be judged. Data collection continues until the baseline shows stability (a consistent level and flat or predictable trend), which typically requires a minimum of three to five data points.

2

The intervention is introduced while measurement of the target behavior continues at the same frequency. The researcher examines whether behavior changes from the established baseline pattern. A clear, replicated change in level, trend, or variability that coincides with the onset of the intervention constitutes the first demonstration of a functional relationship.

3

The intervention is withdrawn and conditions return to the original baseline state (the second A phase). If the behavior reverts toward baseline levels, this reversal strengthens the inference that the intervention — and not some confounding variable — was responsible for the earlier change. A meaningful reversal is the critical test of experimental control in this design.

4

The intervention is reintroduced (the second B phase). Behavior is expected to improve again, providing a second, independent replication of the functional relationship within the same participant. This replication substantially increases confidence in a causal interpretation and is the feature that distinguishes the ABAB design from the simpler ABA reversal.

5

Data are analyzed visually by inspecting level, trend, variability, immediacy of change, overlap between phases, and consistency of effect across the two A–B phase pairs. Effect size metrics such as the non-overlap of all pairs (NAP) or Tau-U are often reported alongside visual analysis to provide a quantitative summary of the functional relationship.

Common pitfalls

  • Withdrawing an effective intervention without ethical justification — always assess whether reversal is permissible and obtain informed consent that includes disclosure of the withdrawal phase.
  • Using too few data points per phase — a minimum of three data points is the conventional floor, but five or more are needed to assess true stability; premature phase changes inflate the risk of false positives.
  • Ignoring trend at baseline — if baseline data show a pre-existing improving trend, apparent intervention effects may reflect natural recovery rather than treatment; the intervention should ideally be introduced on a stable or worsening trend.
  • Relying solely on visual analysis without supplementary effect-size metrics — visual analysis can be unreliable when data are variable or overlap between phases is moderate; reporting NAP or Tau-U alongside graphs improves rigor.
  • Confusing ABA with ABAB — ending the design at the second A phase (ABA) leaves the participant in a withdrawn-treatment condition and provides only one replication of the functional relationship, which is both weaker scientifically and potentially less ethical.

Frequently asked questions

How is the ABAB design different from the ABA design?

Both are reversal designs, but the ABAB design adds a second intervention phase after the withdrawal. This second B phase provides an independent replication of the functional relationship within the same study, substantially strengthening causal inference. The ABA design ends with the withdrawal phase, leaving the participant in the no-treatment condition and offering only one demonstration of experimental control.

What should I do if behavior does not reverse during the withdrawal phase?

A failure to reverse can mean several things: the behavior may have been permanently acquired (making reversal design inappropriate), a history or carryover effect may be masking the reversal, or the baseline was not stable enough to detect change. Document the non-reversal carefully. If reversal is conceptually expected but does not occur, consider switching to a multiple baseline design, which does not require withdrawal, for future work.

Is it ethical to withdraw a working intervention?

This is a genuine concern. Ethics committees and participants must be informed in advance that the intervention will be temporarily withdrawn. The withdrawal is generally acceptable when (a) the behavior poses no immediate safety risk, (b) the withdrawal phase is kept as brief as scientifically necessary, and (c) the intervention is fully reinstated in the final B phase. Some behaviors — crisis-level self-injury, life-sustaining treatments — make any withdrawal unacceptable, and an alternative design such as a multiple baseline should be used.

How many data points per phase are required?

There is no fixed rule, but the conventional minimum is three data points per phase. In practice, researchers continue collecting data until the series shows a stable pattern — consistent level, predictable trend, and low variability. Five or more data points per phase is a widely recommended target. Stability criteria should be defined a priori to prevent selective stopping.

Can the ABAB design be used with more than one participant?

Yes. Running the ABAB design across multiple participants independently constitutes systematic direct replication, which is how single-subject research builds generalizable evidence. Each participant's data are analyzed individually; the cross-participant consistency of effects is then evaluated as evidence of external validity. This is different from averaging data across participants, which would obscure the individual-level functional relationships the design is built to reveal.

Related methods

Sources

  1. Sidman, M. (1960). Tactics of Scientific Research: Evaluating Experimental Data in Psychology. Basic Books.

    Open publication reference
    Copied.
  2. Baer, D. M., Wolf, M. M., & Risley, T. R. (1968). Some current dimensions of applied behavior analysis. Journal of Applied Behavior Analysis, 1(1), 91–97.

    Open DOI reference
    Copied.