3.1 Single-Case Experimental Designs (Reversal, Multiple-Baseline, Alternating-Treatments, Changing-Criterion)
Key Takeaways
- Internal validity measures control over confounding variables to attribute dependent variable changes to the independent variable, whereas external validity concerns generalizability established through systematic replication.
- Baseline logic consists of three fundamental components: prediction (projecting future baseline data paths), verification (demonstrating baseline stability without intervention), and replication (reproducing treatment effects).
- Reversal/withdrawal designs (ABAB, BAB) demonstrate powerful experimental control by withdrawing treatment, but require reversible behaviors and pose ethical risks for dangerous target behaviors.
- Multiple-baseline designs establish experimental control without withdrawing treatment by staggering intervention introduction across independent behaviors, subjects, or settings.
- Alternating-treatments designs rapidly alternate conditions to compare relative effectiveness, while changing-criterion designs evaluate stepwise performance criterion shifts using bi-directional controls.
Single-Case Experimental Designs
In Applied Behavior Analysis (ABA), single-case experimental designs (also called single-subject designs) serve as the primary methodology for evaluating the functional relationships between environmental manipulations and behavioral changes. Unlike traditional group designs that aggregate data across large sample populations and compare group means, single-case designs treat each individual participant as their own experimental control. By repeatedly measuring an individual's behavior across time and varying environmental conditions, behavior analysts can convincingly demonstrate whether a specific intervention (the independent variable [IV]) is responsible for changes in a targeted behavior (the dependent variable [DV]).
Internal Validity, External Validity, and Confounding Variables
The rigor of any experimental design is judged by its validity. Board-certified Qualified Behavior Analysts (QBAs) must balance two distinct forms of validity when conducting clinical evaluations:
- Internal Validity: The degree to which an experiment convincingly demonstrates that changes in the dependent variable are a direct function of the independent variable, and not the result of uncontrolled or extraneous factors (confounding variables). Common threats to internal validity include history (events occurring outside the clinical setting), maturation (developmental or physical changes within the learner), testing effects (repeated exposure to assessment materials), instrumentation (observer drift or faulty equipment), and regression to the mean.
- External Validity: The degree to which an experimental finding can be generalized to other subjects, settings, behaviors, or time periods. In single-case research, external validity is not established through random sampling or statistical inference, but rather through systematic replication across different clients, practitioners, and clinical environments.
Internal Validity ==> Measures control over confounding variables (Is IV causing DV change?)
External Validity ==> Measures generality across subjects/settings via systematic replication
The Three Pillars of Baseline Logic
To establish a functional relation without relying on inferential statistics, single-case methodology utilizes baseline logic. Baseline logic consists of three distinct, sequential reasoning components: prediction, verification, and replication.
| Baseline Logic Element | Methodological Definition | How It Is Demonstrated in Single-Case Designs |
|---|---|---|
| Prediction | The expected path of future data assuming no environmental changes occur. | Established by collecting continuous baseline data until a stable level, trend, and low variability are observed. |
| Verification | Demonstrating that the prior baseline level of responding would have continued unchanged had the independent variable not been introduced. | Demonstrated by withdrawing the treatment back to baseline conditions (in reversal designs) or showing that unintroduced baselines remain stable (in multiple-baseline designs). |
| Replication | Reproducing the observed behavioral change when the independent variable is re-introduced under identical conditions. | Demonstrated by re-applying the intervention in a second treatment phase (e.g., the second B in an ABAB design) or introducing it across subsequent baselines. |
Reversal and Withdrawal Designs (ABAB, BAB)
The reversal design (often used interchangeably with withdrawal design) is the most fundamental single-case design for demonstrating experimental control. The classic ABAB design alternates between baseline (A) and intervention (B) phases.
Mechanics of Experimental Control in ABAB
- Phase A1 (Initial Baseline): Data are collected until stability is achieved, allowing for prediction of future responding.
- Phase B1 (Initial Intervention): The independent variable is applied. A change in level, trend, or variability supports the hypothesis that the IV altered the DV.
- Phase A2 (Withdrawal/Reversal): The IV is removed. If responding returns to baseline levels, verification is achieved, ruling out history and maturation as confounding causes.
- Phase B2 (Re-intervention): The IV is re-introduced. If the behavior changes again in the same direction, replication is confirmed, establishing a definitive functional relation.
The BAB Variation
A BAB design begins immediately with the intervention phase, withdraws it to baseline, and then re-applies intervention. QBAs implement BAB designs under urgent clinical circumstances, such as when evaluating severe, dangerous target behaviors (e.g., severe self-injurious behavior or extreme aggression) where delaying treatment for an initial baseline is unethical, or when time constraints prevent a prolonged initial baseline.
Limitations and Ethical Constraints
- Non-Reversible Behaviors: Reversal designs cannot evaluate behaviors that cannot be unlearned once acquired (e.g., reading fluency, riding a bicycle, or mastered academic skills) or behaviors influenced by persistent environmental contingencies.
- Ethical Risks: Withdrawing an effective treatment for dangerous behaviors (e.g., severe self-injury, property destruction) poses unacceptable safety risks to clients and staff.
- Sequence Effects: The participant's experience in one phase may influence responding in subsequent phases.
Multiple-Baseline Designs
When target behaviors are non-reversible or when withdrawing treatment is ethically inappropriate, the multiple-baseline design is the gold standard alternative. In a multiple-baseline design, two or more independent baselines are established simultaneously, and the independent variable is introduced in a staggered fashion across time.
Primary Variations
- Multiple-Baseline Across Behaviors: One subject; two or more independent behaviors in the same setting.
- Multiple-Baseline Across Subjects: Two or more independent subjects; same target behavior in the same setting.
- Multiple-Baseline Across Settings: One subject; same target behavior across two or more distinct settings (e.g., home, school, community).
Experimental Control and Baseline Logic
Experimental control is demonstrated when each baseline changes only when the intervention is specifically applied to it, while untreated baselines remain stable at baseline levels. Verification is confirmed because untreated baselines serve as an ongoing control demonstrating what the treated baseline would look like without intervention. Replication is established as each subsequent baseline responds similarly upon treatment introduction.
Adaptations: Multiple-Probe and Delayed Baseline
- Multiple-Probe Design: Intermittent baseline measurements are taken rather than continuous data collection. Ideal for evaluating skill sequences (chaining) or long-term acquisition where continuous baseline testing would cause fatigue or practice effects.
- Delayed Baseline Design: Initial baselines are begun at different points in time rather than simultaneously. Useful when a new client enters a clinic late or when a new behavior emerges mid-treatment.
Alternating-Treatments / Multielement Designs
An alternating-treatments design (also known as a multielement design) rapidly alternates two or more independent variables (or intervention conditions) within the same phase, often counterbalancing conditions across time of day, staff members, or session order.
Session 1: Condition A (e.g., Praise) ==> Session 2: Condition B (e.g., Token)
Session 3: Condition B (e.g., Token) ==> Session 4: Condition A (e.g., Praise)
Key Advantages
- Does not require baseline withdrawal.
- Fast comparison of multiple interventions in a short period.
- Minimizes sequence effects through rapid, randomized alternation.
- Accommodates non-reversible behaviors.
Methodological Concerns
- Multi-Treatment Interference / Carryover Effects: The performance in one condition is influenced by the lingering effects of a adjacent condition. QBAs minimize carryover by associating distinct discriminative stimuli ($S^D$s) with each condition (e.g., different colored folders, different rooms) and allowing time intervals between sessions.
Changing-Criterion Designs
The changing-criterion design is used to evaluate interventions designed to produce gradual, step-wise changes in a single continuous behavior already in the learner's repertoire (e.g., increasing minutes of physical exercise or decreasing cigarettes smoked per day).
Structure and Verification Mechanisms
Following an initial baseline, a treatment phase is divided into multiple subphases. Each subphase sets a specific performance criterion. Experimental control is demonstrated when the behavior closely matches each newly established criterion.
To strengthen experimental control, QBAs utilize bi-directional shifts (mini-reversals) by temporarily changing the criterion back to a previous level. If the behavior follows the criterion back down (or up) and then resumes the desired direction in the next subphase, coincidental maturation or history threats are completely ruled out.
In single-case experimental design methodology, how is the baseline logic element of 'verification' specifically demonstrated within an ABAB reversal design?
A behavior analyst is designing an evaluation for a client who engages in dangerous high-intensity elopement across three different school environments (classroom, playground, cafeteria). The analyst cannot withdraw treatment once successful. Which single-case design is most appropriate?
When utilizing an alternating-treatments (multielement) design to evaluate two distinct reinforcement schedules, what primary methodological threat must the behavior analyst actively control for?