Written and maintained by CASRAI Editorial Board
Published · Last updated
On this page
Definition of Single Subject Design (Single Case Design)
A single subject design (also known as a single case design, small-n design, or single-case experimental design) is an idiographic quantitative research methodology in which a single entity—such as an individual participant, classroom, or clinical unit—serves as its own experimental control through repeated, prospective measurement of a target behavior across systematically manipulated conditions. Rather than inferring treatment efficacy from aggregate mean differences across randomized cohorts, single-case methodology demonstrates causal experimental control through intra-subject replication across distinct baseline and intervention phases. In behavioral, special education, and neurorehabilitation research, this idiographic approach complements traditional group models, such as within-subjects vs. between-subjects designs, establishing high internal validity within established categories of research design when studying low-incidence populations or individualized interventions.
Core Types of Single Subject Design
Single subject research uses distinct structural configurations to establish internal validity across clinical and educational environments:
- Reversal Design and ABA Design: In a basic three-phase ABA design, an initial baseline (A) is recorded, an intervention (B) is applied, and the intervention is withdrawn back to baseline (A). Demonstrating that the target behavior deteriorates upon withdrawal rules out history, maturation, and reverse causality.
- ABAB Design: To prevent leaving participants in an untreated state, an ABAB design reintroduces the intervention in a fourth phase. Replicating the effect across two separate phase shifts (A1 to B1, A2 to B2) provides robust evidence of experimental control. Under What Works Clearinghouse (WWC) Version 5.0 standards, meeting criteria without reservations requires at least four phases with a minimum of 5 data points per phase.
- Multiple Baseline Design: When a target behavior is irreversible (e.g., acquired academic skills) or withdrawal is unethical, researchers employ a multiple baseline design. Baseline measurements are initiated simultaneously across three or more participants, settings, or behaviors, with the intervention introduced in a staggered, time-lagged sequence. WWC standards require at least three baseline-intervention comparisons with minimum phase lengths.
- Alternating Treatments Design: Rapidly alternates two or more distinct interventions within the same subject across consecutive observation sessions, establishing experimental control through vertical separation of graphed data paths without requiring withdrawal phases.
- Changing Criterion Design: Applied to stepwise, unidimensional behavioral goals (e.g., progressive exercise duration). Following baseline, successive intervention sub-phases establish stepwise performance targets, demonstrating control as behavior tracks each benchmark.
Formula for Non-Overlap: Percentage of Non-Overlapping Data (PND)
Alongside systematic visual analysis, researchers calculate quantitative non-overlap effect sizes. The most widely established non-overlap metric is the Percentage of Non-overlapping Data (PND; Scruggs, Mastropieri, & Casto, 1987). When the therapeutic objective is to increase a desired behavior, PND is calculated using the following formula:
$$text{PND} = left( frac{n_{text{exceed}}}{N_{B}} right) times 100$$
Where every mathematical component is defined as:
- PND: Percentage of Non-overlapping Data, expressed as a score between 0% and 100%.
- n_exceed: The number of intervention phase (B) data points whose values are strictly greater than the maximum value observed during baseline ($y_i > max(y_A)$). For behavior reduction goals, this represents points strictly below the baseline minimum ($y_i < min(y_A)$).
- N_B: The total number of observation sessions collected throughout intervention phase B.
Worked Example with Real Numbers and Arithmetic Shown
Consider an educational psychologist evaluating a self-monitoring intervention to increase on-task academic engagement (percentage of 30-minute intervals on-task) for an elementary student. The evaluation uses an AB phase structure adhering to WWC data-point criteria:
- Baseline Phase (Phase A, 5 sessions): 32%, 38%, 35%, 40%, 36%
- Intervention Phase (Phase B, 7 sessions): 38%, 42%, 55%, 52%, 64%, 60%, 66%
The arithmetic calculation proceeds through four explicit steps:
- Identify Baseline Maximum: Across baseline sessions ${32, 38, 35, 40, 36}$, the maximum value is $max(y_A) = 40%$.
- Evaluate Intervention Data Points: Compare each of the 7 intervention values against the baseline maximum of 40%:
- Session 6: 38% (does not exceed 40%; non-overlap = 0)
- Session 7: 42% (exceeds 40%; non-overlap = 1)
- Session 8: 55% (exceeds 40%; non-overlap = 1)
- Session 9: 52% (exceeds 40%; non-overlap = 1)
- Session 10: 64% (exceeds 40%; non-overlap = 1)
- Session 11: 60% (exceeds 40%; non-overlap = 1)
- Session 12: 66% (exceeds 40%; non-overlap = 1)
- Tally Exceeding Points: Exactly 6 out of 7 intervention sessions exceed the baseline maximum ($n_{text{exceed}} = 6$, $N_B = 7$).
- Calculate PND:
$$text{PND} = left( frac{6}{7} right) times 100 = 0.85714 times 100 = 85.71%$$
Interpretation and Reading Guidance
Visual analysis evaluates six core data dimensions: level, trend, variability, immediacy of effect, overlap, and consistency across phases. To interpret quantitative non-overlap, Scruggs and Mastropieri (1998) established standardized empirical benchmarks for PND scores:
| PND Score Range | Effectiveness Classification | Visual Analysis Corroboration |
|---|---|---|
| PND > 85% | Highly Effective | Pronounced level shift, rapid immediacy of change, and minimal data overlap between phases. |
| 70% ≤ PND ≤ 85% | Moderately Effective | Clear therapeutic change with modest overlap; stable trend in desired direction. |
| 50% ≤ PND < 70% | Minimally Effective / Questionable | Substantial phase overlap; weak or delayed level shift requiring cautious interpretation. |
| PND < 50% | Ineffective / Unreliable | Extensive overlap; intervention data fall within baseline variability, showing no functional relation. |
In our worked example, the PND score of 85.71% falls within the “Highly Effective” range (> 85%), corroborated by an immediate level increase from a baseline mean of 36.2% to an intervention mean of 53.86%.
Assumptions and Common Mistakes
Reliable causal inference in single subject research depends on key methodological assumptions:
- Baseline Stability: Baselines must demonstrate stable level and absence of therapeutic trend before introducing interventions. Introducing treatment during an improving baseline confounds intervention effects with natural recovery.
- Measurement Reliability (IOA): Data must be collected using operational definitions with inter-assessor agreement (IOA) assessed on at least 20% of sessions per phase, achieving mean agreement ≥ 80% or Cohen’s kappa ≥ 0.60.
- Direct Independent Variable Manipulation: The researcher must actively control intervention onset and withdrawal without co-occurring historical confounds.
Common pitfalls include premature phase transitions prior to data stabilization, ignoring serial autocorrelation when applying statistical tests, and over-relying on PND when baseline data contain anomalous outlier spikes. Unlike group designs utilizing blocking in statistics or a placebo-controlled study design, single-case validity relies entirely on verified phase transitions over time.
How to Report It (APA Example)
Reporting single subject designs must adhere to the Single-Case Reporting Guideline In BEhavioural Interventions (SCRIBE 2016 Statement; Tate et al., 2016) and APA 7th edition standards:
“A concurrent multiple baseline design across three participants was conducted to evaluate the efficacy of the self-monitoring intervention on academic on-task engagement. Baseline sessions across participants showed stable responding (baseline mean = 36.2%, range = 32%–40%). Inter-assessor agreement was recorded across 28.5% of sessions with a mean agreement of 92.4% (range = 88%–100%). Following intervention implementation, Participant 1 demonstrated an immediate level shift (+17.66 percentage points) and an accelerating trend (intervention mean = 53.86%, range = 38%–66%). Quantitative non-overlap analysis yielded a Percentage of Non-overlapping Data (PND) of 85.71% (6 of 7 intervention sessions exceeding baseline maximum), indicating a highly effective intervention according to Scruggs and Mastropieri (1998) benchmarks.”
How to Calculate It in Excel, R, or Python
Calculating PND across analytical platforms requires simple, verified syntax:
Excel
With baseline values in cells A2:A6 and intervention values in cells B2:B8:
=(COUNTIF(B2:B8, ">" & MAX(A2:A6)) / COUNT(B2:B8)) * 100
R
Using base R vector operations:
baseline <- c(32, 38, 35, 40, 36)
intervention <- c(38, 42, 55, 52, 64, 60, 66)
base_max <- max(baseline)
pnd <- (sum(intervention > base_max) / length(intervention)) * 100
cat(sprintf("Baseline Max: %.1f%% | PND: %.2f%%n", base_max, pnd))
Python
Using standard Python operations:
baseline = [32, 38, 35, 40, 36]
intervention = [38, 42, 55, 52, 64, 60, 66]
base_max = max(baseline)
pnd = (sum(1 for val in intervention if val > base_max) / len(intervention)) * 100
print(f"Baseline Max: {base_max}% | PND: {pnd:.2f}%")
Frequently Asked Questions
What is the difference between a single subject design and a case study?
A case study is an observational, retrospective narrative describing a subject’s history without controlled experimental manipulation. A single subject design is an experimental methodology with prospectively planned condition manipulations, repeated time-series measurements, and verified intra-subject replication to confirm causality.
How many participants are needed in a single subject design?
While an individual case is the unit of analysis, single subject studies frequently include 3 to 8 participants within a multiple baseline design. Replicating effects across several individuals establishes external validity while preserving idiographic precision.
What are the What Works Clearinghouse data point requirements?
Under WWC Standards (Version 5.0), receiving a rating of “Meets WWC Design Standards Without Reservations” requires at least four phases with a minimum of 5 data points per phase for an ABAB design, or at least three series with 5 data points per phase for multiple baseline designs. Studies with at least 3 data points per phase can meet standards with reservations.
When should researchers avoid an ABAB reversal design?
Reversal designs should be avoided when target behaviors cannot be unlearned (e.g., acquired reading or math skills) or when withdrawing intervention causes immediate ethical risks (e.g., severe self-injury). Multiple baseline designs are the preferred alternative.
Sources
- Institute of Education Sciences (IES), What Works Clearinghouse. (2022). What Works Clearinghouse Procedures and Standards Handbook, Version 5.0, Chapter VI: Single-Case Design Standards. U.S. Department of Education. URL: https://ies.ed.gov/ncee/wwc/Handbooks
- Tate, R. L., Perdices, M., Rosenkoetter, U., Shadish, W., Vohra, S., Barlow, D. H., Horner, R., Kazdin, A. E., Kratochwill, T., & McDonald, S. (2016). The Single-Case Reporting Guideline In BEhavioural Interventions (SCRIBE) 2016 statement. Journal of Clinical Epidemiology, 73, 142–152. DOI: 10.1016/j.jclinepi.2016.04.006
- Scruggs, T. E., & Mastropieri, M. A. (1998). Summarizing single-subject research: Issues and applications. Behavior Modification, 22(3), 221–242. DOI: 10.1177/01454455980223001
- Appelbaum, M., Cooper, H., Kline, R. B., Mayo-Wilson, E., Nezu, A. M., & Rao, S. M. (2018). Journal article reporting standards for quantitative research in psychology: The APA Publications and Communications Board task force report. American Psychologist, 73(1), 3–25. DOI: 10.1037/amp0000191







