Does single-subject mean exactly one person?

No; cases can be other units, and a study can include multiple cases.

Is every N-of-1 study identical to every single-case design?

No; the labels overlap, but specific randomization, measurement and phase structures differ.

Must an intervention always be withdrawn?

No; multiple-baseline and other designs can be appropriate when withdrawal is unsuitable.

Can repeated observations remove confounding automatically?

No; the schedule and replicated pattern must address plausible alternative explanations.

AI Glossary / Statistics and probability / Intermediate

Single-subject design

Definition

A single-subject design, often called a single-case design, studies outcomes repeatedly within individual cases while systematically varying an intervention or its introduction. A case may be one person, classroom, organization or another clearly defined unit. Well-designed single-case experiments use repeated comparisons to investigate a functional relationship between intervention and outcome. They are not simply anecdotes or ordinary one-time case reports, and a study can include several cases while retaining case-level analysis.

What it means

The design asks how a particular case's outcome changes when the intervention condition changes. Repeated baseline and intervention observations provide information about level, trend, variability and timing. Experimental control can come from replicated phase changes, staggered intervention starts across cases or other structured arrangements. Merely measuring someone before and after an intervention is weaker because unrelated events, maturation or existing trends may explain the difference. The logic depends on the design and replication of the pattern, not on sample size alone.

How it works

Define a measurable outcome, case, observation schedule and intervention. Establish repeated baseline measurements and choose a design suitable for the question, such as reversal, multiple baseline or alternating treatments. Where feasible, randomize relevant timing or sequences and ensure consistent measurement. Examine the pattern within and across phases, including immediacy, overlap and trend. Use statistical summaries appropriate to the design and serial dependence when helpful. Avoid withdrawing an intervention when doing so would be unsafe or when its effects are expected to be irreversible.

Three practical examples

Example 1: A workplace tests a reversible notification setting across repeated baseline and intervention periods for one team. Example 2: An education study introduces a support strategy at different planned times across several classrooms. Example 3: An AI usability project alternates two assistance modes across suitably designed sessions for individual users. These examples require attention to carryover, learning and external changes. A design that works for a reversible interface feature may be unsuitable for a permanently learned skill.

Think of it this way

Think of watching a process carefully through planned changes rather than comparing only two isolated photographs. If the outcome changes repeatedly in step with the intervention, that pattern can strengthen the explanation. But a convincing story requires more than a difference in averages: timing, prior trends and alternative events matter. Several well-planned demonstrations are more informative than simply collecting many closely related observations during one unchanged period.

Worked example

Consider a purely illustrative A-B-A-B sequence for a reversible workplace setting, where A is baseline and B is the intervention. Five observations in the first A phase are 8,9,8,10,10; the first B phase has 4,5,4,6,6; the second A phase has 9,8,10,9,9; and the second B phase has 5,4,5,6,5. Phase means are nine, five, nine and five. Each A-to-B transition therefore shows a four-unit reduction, about 44.44% relative to nine. The return towards baseline and second reduction create a replicated descriptive pattern. These 20 observations are not 20 independent participants, and their serial dependence cannot be ignored in inference. The example does not claim that a particular standards checklist is satisfied or that phase means alone establish causation.

Illustrative case study

In this fictional case study, Elmwood Operations evaluates an AI reminder setting intended to reduce missed routine checks. The setting can be safely switched on and off, and the team records the same operational outcome throughout planned phases. Analysts examine whether changes align with the switches while checking staffing and workload records for competing explanations. They report individual phase trajectories and note possible carryover. Before generalizing to other teams, they seek replication across additional settings rather than assuming one case represents the whole organization.

Common misconceptions and mistakes

Common mistakes include equating single-case research with an anecdote, treating repeated observations as independent participants and claiming that any A-B comparison establishes an intervention effect. Another error is using withdrawal designs for irreversible learning or when withdrawing support would be inappropriate. Mean differences can hide baseline trends, delayed effects or changing variability. Statistical significance alone does not replace the design's experimental-control logic, and a strong within-case result does not automatically establish broad population generalizability.

Business use

Businesses may use single-case designs when they need detailed evidence about a particular workflow, device or user and can manipulate conditions responsibly. AI applications with rapid reversible settings can sometimes support carefully planned within-case comparisons. Teams should document intervention fidelity, measurement consistency, phase decisions and concurrent changes. Generalization requires replication and a clear description of the cases and contexts studied. The design is most useful when the operational question fits its structure, not merely when recruiting a larger group is inconvenient.

Check your understanding

What are the four phase means in the worked example?

Answer: Nine, five, nine and five.

How many independent participants does the example contain?

Answer: It describes one case, not 20 participants.

Why is the return to baseline informative?

Answer: It contributes a repeated pattern linked to the condition change, subject to design assumptions.

Why might the same design be unsuitable for a learned skill?

Answer: Learning may persist when the intervention is withdrawn.

Related terms

N-of-1 trial
Baseline
Multiple baseline design
Related terms include single-case experiment
reversal design
alternating treatments
serial correlation and intervention fidelity

Learn next

Learn next how visual analysis considers level, trend, variability, immediacy and overlap. Compare reversal and multiple-baseline logic, then study randomization-based or time-series analyses appropriate to the chosen design. Plan phase changes and replication before data collection, and distinguish evidence about the studied case from evidence about a broader population.

Sources

Sources: What Works Clearinghouse, Single-Case Design Technical Documentation, https://ies.ed.gov/ncee/wwc/Docs/ReferenceResources/wwc_scd.pdf; Institute of Education Sciences, The Role of Effect Size in Conducting, Interpreting, and Summarizing Single-Case Research, https://ies.ed.gov/use-work/resource-library/report/research-report/role-effect-size-conducting-interpreting-and-summarizing-single-case-research. The phase data and fictional workplace case are original illustrations; no claim of compliance with a current review standard is made.

Previous: Single source of truth Next: Single version of the truth