Skip to content
Back to search results

Adaptive AB Design — Adaptive AB Single-Subject Experimental Design

Introduced by
Baer, Wolf & Risley (AB foundation); Kratochwill & Levin (adaptive single-case extensions)
Year
1968 (AB foundation); 2000s (adaptive extensions)
Type
Single-subject experimental design with adaptive phase-change rules
Family
Experimental design
Font size

What does it do?

The adaptive AB design is a single-subject experimental design that retains the two-phase baseline-then-intervention structure of the classic AB design but replaces fixed session-count rules with pre-specified data-driven criteria — such as stability thresholds or trend benchmarks — that determine when to transition between phases.

This adaptive logic allows the phase boundary to move in response to the individual participant's actual performance trajectory rather than a predetermined schedule.

The intuition behind the method

In a standard AB design you might decide in advance to collect ten baseline sessions and then switch to the intervention regardless of what the data show. In an adaptive AB design you instead write down a rule before the study begins — for example, 'move to the intervention phase only when five consecutive data points are within ten percent of each other and show no upward trend.' The design responds to the participant's own data rather than a fixed calendar, making the baseline more defensible and the phase transition more scientifically justifiable. The core trade-off remains: without a reversal phase, the design still cannot rule out coincidental events as the cause of any change observed.

History

The AB design was foundational to applied behavior analysis following Baer, Wolf, and Risley's seminal 1968 paper. Single-subject researchers recognized early that fixed session counts for baselines were arbitrary, and methodologists advocated stability criteria as the scientifically appropriate stopping rule — a position consolidated in influential texts such as Cooper, Heron, and Heward's Applied Behavior Analysis. The explicit framing of data-driven phase-change criteria as an adaptive feature of single-case designs gained traction in the 2000s alongside broader interest in adaptive experimental methodology, as reviewed by Kratochwill and Levin (2010) in their synthesis of single-case methodological advances.

When to use it

Use the adaptive AB design when you need a single-subject baseline-intervention framework but cannot determine in advance exactly how many sessions the individual participant will require to achieve a stable baseline — common in applied behavior analysis, special education, rehabilitation, and clinical case studies where participant variability is high. It is preferable to a fixed-session AB design when baseline instability would otherwise force an early, unjustified phase transition or an unnecessarily prolonged one. Do not use the adaptive AB design when causal evidence is required: like the standard AB design, it cannot rule out history or maturation effects without a reversal or replication phase. When internal validity is the priority, use an ABA, ABAB, or multiple-baseline design. Also avoid it when the team cannot commit to the decision rule before data collection, as post-hoc phase-boundary adjustment invalidates interpretation.

Limitations and considerations

  • Like the standard AB design, it cannot establish causality: without a reversal phase, history, maturation, and regression to the mean remain plausible alternative explanations.
  • The adaptive phase-change rule introduces an additional methodological decision point that must be fully pre-specified and transparently reported to avoid researcher degrees of freedom.
  • Results apply only to the individual studied; external generalizability requires systematic replication across multiple participants and settings.
  • If the stability criterion is set too stringently, baseline data collection may extend impractically; if too loosely, the baseline is no more defensible than a fixed-session rule.

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

Strengths

  • The data-driven phase transition produces a more scientifically defensible baseline than an arbitrary fixed session count.
  • Flexibility accommodates individual differences in baseline stability without compromising the pre-specification principle.
  • Retains the ethical advantage of the AB design — no withdrawal of a presumably beneficial intervention is required.
  • Compatible with visual analysis and Tau-U effect-size estimation used across single-subject research.
  • Practical in clinical and educational settings where session timing is constrained and participant trajectories are unpredictable.

Applications

  • Evaluating the effect of a reading intervention on an individual student whose pre-intervention fluency is highly variable, using a stability criterion to determine when baseline data are sufficient.
  • Monitoring behavioral outcomes in applied behavior analysis practice where client improvement rates during baseline would distort a fixed-session phase transition.
  • Documenting treatment response in rehabilitation medicine for a single patient whose motor recovery trajectory during the no-treatment phase must meet a plateau criterion before physical therapy is introduced.
  • Assessing the impact of a classroom accommodation in special education, where the practitioner needs a principled rule for when baseline behavior is stable enough to justify the intervention.
  • Piloting a new psychosocial intervention with a single case in clinical psychology before committing to a group design, using adaptive baseline criteria to protect internal validity.
  • Tracking self-monitoring outcomes in occupational health for an individual employee, with the phase transition governed by pre-agreed behavioral stability thresholds.

How it works

How it works

1

Operationally define the dependent variable — the specific, observable, and measurable behavior or outcome to be tracked across both phases. Clear operationalization ensures that measurements in the baseline and intervention phases are strictly comparable and that the adaptation criteria can be applied consistently.

2

Collect repeated measurements of the target behavior under natural, no-intervention conditions (the A phase). Rather than stopping at a fixed number of sessions, continue baseline data collection until the pre-specified stability criterion is met — for example, no more than ten percent variability across the last five data points and no systematic trend in the desired direction. This data-responsive stopping rule produces a more defensible pre-intervention estimate than an arbitrary fixed-session rule.

3

Evaluate the accumulated baseline data against the pre-specified decision criteria at each session. If the stability threshold is not yet met, extend the baseline. If the trend criterion is not yet satisfied — for example, if the baseline is already improving in the desired direction — reassess whether the intervention phase can be informatively interpreted or whether the design should be redesigned. The decision rule must be fully specified before data collection begins to prevent post-hoc rationalization.

4

Once the decision rule signals that the baseline criterion has been satisfied, introduce the treatment at a clearly defined session boundary while holding all other conditions constant. Document the exact session number and rationale for the phase transition. No concurrent procedural changes should accompany the intervention onset.

5

Plot all time-series data with a phase-change line at the adaptive transition point. Evaluate changes in level, trend, and variability between phases using standard visual analysis conventions. Supplement visual analysis with effect-size statistics such as Tau-U, which accounts for baseline trend and is applicable to short series. Interpret any observed change cautiously given the absence of a reversal phase.

Common pitfalls

  • Specifying or adjusting the decision rule after viewing baseline data — this post-hoc flexibility destroys the interpretive validity of the phase transition and constitutes an undisclosed researcher degree of freedom.
  • Proceeding to the intervention phase despite a baseline that is trending in the desired direction — an improving baseline makes any further improvement during the B phase uninterpretable as an intervention effect.
  • Too few data points before the criterion is met: if the stability rule is satisfied by only three or four sessions, the baseline is too short for reliable visual analysis; the criterion should require a minimum number of observations as well as a stability threshold.
  • Conflating the adaptive phase-change mechanism with adaptive designs in clinical trials — the adaptive AB design borrows the word 'adaptive' in a single-subject context; it does not involve the alpha-spending functions or interim analyses of clinical adaptive designs.

Frequently asked questions

What makes this design 'adaptive' compared to a standard AB design?

In a standard AB design the number of baseline sessions is typically fixed before data collection begins. In the adaptive AB design the phase transition is governed by a pre-specified data-driven criterion — such as a stability threshold or trend benchmark — so the baseline continues until the participant's own data satisfy the rule. The word 'adaptive' refers to this data-responsive phase-boundary decision, not to clinical trial interim analyses.

How do I specify the decision rule without biasing the results?

The stability criterion and any minimum session count must be written into the study protocol before any data are collected and, ideally, registered or documented with a timestamp. Common criteria include a specified maximum coefficient of variation across the last N data points, the absence of a monotone trend of a given slope, or a combination of both. Post-hoc modification of the criterion is not acceptable.

Does the adaptive phase-change rule solve the causal inference problem of the AB design?

No. The adaptive rule strengthens the pre-intervention estimate by ensuring a genuinely stable baseline, but it does not add a reversal or replication element. History and maturation remain alternative explanations for any change observed after the phase transition. Causal inference in single-subject research requires a reversal (ABA, ABAB) or a multiple-baseline replication structure.

Can I use Tau-U with adaptive AB data?

Yes. Tau-U is appropriate for short time-series data from single-subject designs and accounts for baseline trend when computing the effect size. It can be applied to adaptive AB data in the same way as standard AB data, using all data points from each phase. The phase-boundary location (determined adaptively) should be reported transparently alongside the Tau-U statistic.

When should I upgrade from an adaptive AB design to an ABA or ABAB design?

Upgrade when the research question requires causal evidence rather than preliminary documentation, when the behavior being targeted is not expected to be irreversible with learning, and when withdrawing the treatment is ethically feasible. The adaptive AB design is best suited to pilot investigations, practitioner case documentation, or situations where reversal is ethically or practically impossible.

Related methods

Sources

  1. 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.
  2. Kratochwill, T. R., & Levin, J. R. (Eds.). (2010). Single-Case Intervention Research: Methodological and Statistical Advances. American Psychological Association.

    ISBN: 978-1433807428
    Copied.