Skip to content
Back to search results

Single-Case Experimental Design

Introduced by
Thomas R. Kratochwill and the What Works Clearinghouse single-case design panel
Year
2013
Type
Within-subject experimental pipeline for evaluating interventions on individuals
Family
Single-case / single-subject experimental methodology
Font size

What does it do?

Single-case experimental design (SCED) is a family of rigorous within-subject experimental methodologies for evaluating whether an intervention causes change in an individual, widely used in rehabilitation, special education, and applied behavior analysis.

Rather than averaging across a large sample, SCED measures a defined target behavior repeatedly across a baseline (A) phase and an intervention (B) phase, and infers a causal effect when the change is replicated at three or more different points in time within the same case. Internal validity is built into the design itself through systematic manipulation of the independent variable and repeated demonstrations of effect, not through a control group. The 2013 What Works Clearinghouse single-case design standards, formalized by Kratochwill and colleagues, codified what counts as a credible SCED, including requirements for systematic manipulation, at least three attempts to demonstrate an effect, and minimum data points per phase. SCED is the experimental backbone of evidence-based practice for individuals whose conditions, contexts, or low incidence make group designs impractical.

The intuition behind the method

Imagine a clinician working with a single child whose self-injurious behavior they want to reduce. A randomized trial would need many children and would tell you the average effect, but the clinician faces one person here and now. SCED solves this by making the individual their own control: you watch the behavior carefully over time until it is stable and predictable, then you introduce the intervention and watch again. If the behavior changes right when the intervention starts, that is suggestive but not yet convincing, because something else might have changed at the same moment. The trick is replication within the case: by turning the intervention on and off, or staggering it across behaviors, settings, or people, you create three or more separate moments where the behavior changes only when the intervention is present. When change tracks the manipulation that reliably, coincidence becomes an implausible explanation, and you can credibly say the intervention worked for this person.

History

Single-case experimentation grew out of the experimental analysis of behavior pioneered by B. F. Skinner and developed within applied behavior analysis from the 1960s onward, where researchers documented intervention effects on individuals through repeated measurement and reversal designs. For decades the field relied on visual analysis and an implicit craft tradition, which left study quality difficult to compare. The pivotal modern development came when the What Works Clearinghouse convened a panel to formalize design standards; Kratochwill, Hitchcock, Horner, Levin, Odom, Rindskopf, and Shadish published the resulting single-case intervention research design standards in 2013, specifying systematic manipulation, three demonstrations of effect, and minimum data-point requirements. These standards reoriented special education, rehabilitation, and behavioral research toward a gradeable, reproducible methodology and spurred parallel work on effect-size indices and evidence aggregation for single-case data.

When to use it

Use single-case experimental design when you need rigorous causal evidence about whether an intervention works for a specific individual or a small set of individuals, and when a group design is impractical or inappropriate. It is the method of choice for low-incidence conditions, heterogeneous populations, and clinical or educational settings where each person's response matters and large samples cannot be assembled — common throughout rehabilitation, special education, and behavior analysis. SCED requires a target behavior that can be measured repeatedly and a context in which the intervention can be systematically introduced, withdrawn, or staggered to create replicated demonstrations of effect. It is less suitable when the outcome cannot be measured frequently, when withdrawal of an effective treatment would be unethical and no staggered alternative exists, or when the research question is genuinely about average effects across a population, where a group design is preferable.

Limitations and considerations

  • External validity is limited; demonstrating an effect in one or a few cases does not establish generalization without systematic replication.
  • Requires behaviors that can be measured repeatedly and reliably, which excludes outcomes that change slowly or cannot be sampled often.
  • Designs relying on withdrawal of treatment can be ethically problematic or impossible for irreversible learning.
  • Strong serial dependence in within-person time series complicates statistical analysis and visual interpretation.

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

Strengths

  • Establishes experimental control and causal inference for a single individual without requiring a control group or large sample.
  • Ideal for low-incidence, heterogeneous, or clinical populations where group randomized trials are infeasible.
  • Repeated within-person measurement reveals trajectory, immediacy, and variability of response that group means obscure.
  • Backed by formal design standards that make studies appraisable, comparable, and aggregable into evidence reviews.

Applications

  • Evaluating behavioral and instructional interventions for individual students in special education.
  • Testing rehabilitation therapies for patients with low-incidence or heterogeneous disabilities.
  • Assessing the effect of assistive technology or environmental adaptations on a single user's functioning.
  • Demonstrating the efficacy of behavior-reduction procedures for self-injury or challenging behavior in clinical settings.
  • Building evidence reviews by appraising and aggregating multiple single-case studies against design standards.

How it works

How it works

1

The pipeline begins by operationally defining a single target behavior and a measurement system that can be applied repeatedly and consistently over time. The behavior must be observable and quantifiable — frequency, duration, latency, percentage of intervals, or a rate — so that the same metric can be recorded across every session. Repeated measurement is the defining feature that separates SCED from a one-shot pretest-posttest: the dependent variable is sampled many times under each condition, generating a time series for the individual rather than a single number. A clear operational definition and reliable, often inter-observer-verified, measurement are prerequisites because all later inferences rest on the integrity of this repeated signal.

Formula
y_t,\ t=1,\dots,T,\quad y_t \text{ a defined, operationally measured target behavior sampled repeatedly over time}
2

Before any intervention, the analyst collects a baseline (A) phase to document the behavior's natural level, trend, and variability. The goal is stability and predictability: a baseline with little trend and low variability lets the analyst forecast what would happen in the absence of intervention, which is the implicit counterfactual. The standards recommend a minimum number of data points per phase — typically at least three, and preferably five, to meet the strongest rating — so that the pattern is not an artifact of one or two observations. If the baseline is trending in the therapeutic direction or is highly variable, the analyst extends it until a defensible prediction can be made, because a noisy or improving baseline undermines any later claim of effect.

Formula
\text{Phase } A:\ \{y_t\}_{t\in A},\quad n_A \ge 3\text{–}5,\ \widehat{\text{trend}}(A)\approx 0,\ \text{low variability before } B
3

Experimental control in SCED comes from the experimenter systematically and deliberately manipulating the independent variable — introducing, withdrawing, or changing the intervention at planned points — rather than letting it vary naturally. The transition from baseline (A) to intervention (B) is timed by design, and the analyst compares the level, trend, and immediacy of change between adjacent phases. This active manipulation is what the design standards treat as a non-negotiable feature: a study that merely observes a behavior before and after a naturally occurring change is correlational, not experimental. Documenting fidelity of implementation here ensures that the manipulation actually delivered the intended independent variable.

Formula
A \rightarrow B:\ \text{independent variable set by the experimenter at a planned point},\ \Delta = \bar y_B - \bar y_A
4

The core inferential engine is replication of the effect at three or more different points in time within the same study. A single A-to-B change could coincide with maturation, history, or a regression artifact, so SCED designs build in repeated opportunities to demonstrate the effect — by reversing to baseline and reintroducing intervention (ABAB), by staggering introduction across multiple baselines, or by alternating conditions. Kratochwill and colleagues set three demonstrations as the threshold for meeting design standards, with the logic that each successful replication multiplies the implausibility of a rival explanation. When the behavior reliably changes only when and where the intervention is present, the design itself rules out confounds and licenses a causal claim for the individual.

Formula
\#\{\text{demonstrations of effect}\} \ge 3 \text{ at three different points in time} \Rightarrow \text{experimental control}
5

Finally, the study is appraised against the formal design standards to determine how much confidence its evidence warrants. The What Works Clearinghouse framework rates a single-case study as Meeting Standards, Meeting With Reservations, or Not Meeting Standards based on whether the experimenter systematically manipulated the independent variable, whether outcomes were measured reliably by more than one assessor, whether there were at least three attempts to demonstrate an effect, and whether phases had enough data points. This appraisal step turns SCED from a craft into a reproducible, gradeable methodology, allowing individual studies to be aggregated into evidence reviews and ensuring that weak designs are not mistaken for strong causal evidence.

Formula
\text{rating} \in \{\text{Meets},\ \text{Meets with Reservations},\ \text{Does Not Meet}\}\ \text{per WWC criteria}

Common pitfalls

  • Introducing the intervention before the baseline is stable, so the prediction of no-treatment behavior is indefensible.
  • Claiming an effect from a single A-to-B change without the three demonstrations needed to rule out coincidence.
  • Using too few data points per phase, leaving level and trend estimates at the mercy of one or two observations.
  • Neglecting treatment fidelity and inter-observer agreement, so the measured signal cannot be trusted to reflect the manipulation.

Frequently asked questions

How is a single-case experimental design different from a simple before-and-after comparison?

A before-and-after comparison measures the outcome once before and once after an intervention and is correlational: any change could be due to maturation, history, or chance. A single-case experimental design measures the behavior repeatedly to establish a stable baseline, has the experimenter systematically manipulate the intervention, and replicates the effect at three or more points in time. That replication within the same case is what rules out rival explanations and licenses a causal claim, which a one-shot pre-post comparison cannot support.

Why does the design require three demonstrations of an effect?

A single change in behavior when an intervention starts might coincide with some other event, so one demonstration is not convincing. Each additional successful demonstration — through reversal, staggered introduction across baselines, or alternation — creates another moment where the behavior changes only when the intervention is present, and each one multiplies the implausibility of a coincidental explanation. Kratochwill and colleagues set three as the threshold in the 2013 design standards because three independent demonstrations provide a credible experimental basis for inferring control within a single case.

Can single-case experimental designs contribute to evidence-based practice?

Yes. Because the 2013 design standards make individual studies appraisable and gradeable, well-conducted single-case studies can be rated for quality and aggregated across cases and studies to support claims of generalization. This is essential in fields like special education and rehabilitation where group randomized trials are often infeasible. Systematic replication across individuals, settings, and researchers, combined with effect-size summaries, allows single-case evidence to inform practice guidelines for populations that group designs struggle to reach.

Related methods

Sources

  1. Kratochwill, T. R., Hitchcock, J. H., Horner, R. H., Levin, J. R., Odom, S. L., Rindskopf, D. M., & Shadish, W. R. (2013). Single-case intervention research design standards. Remedial and Special Education, 34(1), 26-38.

    Open DOI reference
    Copied.