Yes. Single subject design requires baseline data, and not merely as a formality. Baseline data establish the pre-intervention level of the target behavior against which every later phase is compared. Without repeated baseline measurement, a single subject study cannot demonstrate experimental control, cannot separate treatment effects from maturation or drift, and cannot meet the evidence standards most journals and review bodies apply.
The requirement is structural, not stylistic. Single subject research asks whether a specific intervention changed a specific behavior in a specific individual. That question is unanswerable without a stable, repeated pre-intervention record. The baseline is the comparison condition, and in the strongest designs it doubles as the control. A study that skips it may still be a case report, but it is not a single subject experiment.
The Mechanism: Why Baseline Functions as the Control
Single subject designs achieve experimental control through replication within and across individuals, not through randomization to groups. The baseline phase supplies the counterfactual: what the behavior looked like when the intervention was absent. Each subsequent phase is compared to that record.
Baseline logic rests on prediction. A stable baseline allows the researcher to predict what would have happened without intervention. If the behavior changes sharply when the intervention begins, and returns when it is withdrawn, the prediction is violated in a pattern that implicates the intervention. This is the reasoning behind the withdrawal (A-B-A-B) design. The reversibility test only works if the first A phase is measured well enough to establish a credible prediction.
Baseline stability is judged by level, trend, and variability. Level is the average performance across baseline sessions. Trend is the direction and slope of change across those sessions. Variability is the spread of scores around the trend line. A baseline that is flat and low in variability permits a confident prediction. A baseline with a steep upward trend predicts continued improvement regardless of treatment, which undermines any later claim.
Repeated measurement is what separates single subject designs from pre-post case reports. A single pre-test gives one number and no variability estimate. Five or more baseline sessions give a distribution, a trend, and a basis for judging whether change exceeds normal fluctuation. That is the mechanism. The baseline is not paperwork; it is the comparator.
Quantified Context: How Much Baseline Is Enough
There is no universal session count, and any source that states one number as a rule is oversimplifying. The quantity depends on the behavior's natural variability and the design used. Still, several concrete thresholds recur in methodology guidance and reporting standards.
A commonly cited minimum is three baseline data points, but most methodologists treat three as a floor for low-variability behaviors and describe five or more as the practical target. For behaviors with high day-to-day fluctuation, such as disruptive behavior in a classroom, longer baselines are needed to estimate variability reliably.
The What Works Clearinghouse standards, used in education research, require at least three data points per phase to meet design standards, and a design with fewer may be classified as meeting standards with reservations or not meeting them. The single-case reporting guideline known as SCRIBE asks authors to report the number of data points per phase, the baseline trend, and any baseline variability. Journals applying these standards routinely return manuscripts with two-point baselines.
Stability criteria matter as much as count. A common heuristic treats baseline as stable when the majority of data points fall within a narrow band around the mean, or when the trend is flat or in the direction opposite to the expected treatment effect. A baseline trending toward the treatment goal is a red flag, because improvement after intervention could simply continue an existing trajectory.
These thresholds must be reconciled explicitly. Three points satisfies a minimum design standard only if variability is low and the trend is flat. Three points with a steep trend or wide scatter does not establish a prediction. Five to eight points is the safer target when behavior is variable. The count is necessary but not sufficient; stability and trend are the deciding criteria.
Design Alternatives and Their Baseline Demands
Different single subject designs place different demands on baseline. The table below compares the main designs by baseline requirement and the inference each supports.
| Design | Baseline requirement | What the baseline establishes |
|---|---|---|
| A-B (simple phase change) | Repeated baseline, typically 5+ sessions | Pre-intervention level; weakest design, vulnerable to confounds |
| A-B-A withdrawal | Stable baseline plus return to baseline | Reversibility; stronger causal inference |
| A-B-A-B reversal | Two baselines, initial and return | Replication of effect within one participant |
| Multiple baseline across behaviors | Concurrent baselines for each behavior | Staggered introduction controls for history and maturation |
| Multiple baseline across participants | Concurrent baselines for each person | Replication across individuals; strongest when baselines are staggered |
| Alternating treatments | Baseline plus rapid condition alternation | Compares two or more interventions against the same baseline |
| Changing criterion | Baseline plus stepwise criteria | Gradual behavior change against progressive goals |
The multiple baseline design deserves emphasis because it uses baseline data in a distinctive way. Several behaviors, settings, or participants are measured simultaneously. The intervention is introduced to one while the others remain in baseline. If the target behavior changes only when its intervention begins, while untreated baselines stay flat, the design controls for history and maturation. That logic fails if the untreated baselines are not measured, which is why concurrent baselines are the defining feature of the design.
Alternating treatment designs still require a baseline to anchor comparisons, even though they rely on rapid condition switching. The baseline provides the reference level against which alternating conditions are judged.
Edge Cases: When Baseline Is Hard or Impossible
Some clinical and applied settings make baseline measurement difficult, and the field has developed responses. Each response carries a cost.
When a behavior is dangerous, withholding intervention to collect a long baseline raises ethical concerns. In these cases researchers shorten the baseline to the minimum needed to establish level and trend, document the rationale, and sometimes use a multiple baseline across participants so that some individuals begin treatment sooner. The ethical constraint is real, but it does not eliminate the requirement; it compresses it.
When a behavior is low-frequency, baseline sessions may need to be longer or more numerous to capture enough occurrences. A behavior that occurs once per week cannot be characterized in three short sessions. Researchers extend observation windows or use interval recording to generate a usable baseline record.
When behavior is highly variable, a stable baseline may never emerge. In that case the researcher documents the variability, may extend the baseline, and proceeds with caution, reporting the limitation. A variable baseline weakens the inference but does not make the study worthless; it makes the claim more modest.
A counterintuitive professional caveat: a perfectly flat baseline can be a warning sign rather than a strength. Zero variability across many sessions sometimes indicates measurement insensitivity, a ceiling or floor effect, or a recording procedure that cannot detect change. Reviewers increasingly ask whether the measure could have registered improvement at all. A baseline with modest, interpretable variability is often more credible than one that is suspiciously uniform.
Misconceptions and Red Flags
Several persistent misconceptions lead researchers astray. Recognizing them prevents the most common design failures.
The misconception that a single pre-test score counts as baseline. One data point cannot establish level, trend, or variability, and a pre-post comparison is a case report, not a single subject experiment. The misconception that baseline is only needed for reversal designs. Every single subject design compares later phases to a pre-intervention record, including multiple baseline and changing criterion designs. The misconception that more baseline sessions always strengthen a study. Excessively long baselines can introduce drift, raise ethical concerns, and inflate the influence of history. The misconception that a stable baseline guarantees a valid study. Stability without measurement sensitivity, interobserver agreement, and procedural fidelity still yields weak evidence. The misconception that baseline data can be reconstructed after intervention begins. Retrospective baseline, drawn from records collected for another purpose, is acceptable only when the recording procedures match and are documented. The misconception that statistical significance replaces visual analysis. Single subject designs rely on visual analysis of level, trend, and variability, supported by effect size indices such as tau or percentage of nonoverlapping data.
A second warning concerns baseline trend. If the baseline is already improving before intervention, a later improvement cannot be attributed to treatment without an appropriate design control, such as a multiple baseline that shows untreated cases remaining flat. Researchers should report baseline trend explicitly and avoid claiming an effect when the trend predicts it.
Practical Actions: Building a Defensible Baseline
The following steps translate the requirements into concrete practice. They apply across education, behavior analysis, clinical psychology, and rehabilitation research.
Define the target behavior operationally before collecting any data. Specify what counts as an occurrence, how it is recorded, and in what settings. Ambiguity here contaminates every later phase.
Collect at least five baseline sessions when behavior is variable, and no fewer than three when it is demonstrably stable. Document the rationale for the number chosen. Report level, trend, and variability for the baseline phase in the manuscript.
Assess stability before proceeding. If the baseline trends toward the treatment goal or shows wide scatter, extend it or reconsider the design. Do not begin intervention on a baseline you cannot use as a prediction.
Establish interobserver agreement during baseline, not only during intervention. Agreement of 80 percent or higher is a common reporting benchmark. Baseline data of unknown reliability cannot anchor a comparison.
Plan the design before data collection. Decide whether a withdrawal, multiple baseline, alternating treatment, or changing criterion design fits the question and the setting. Each imposes specific baseline requirements that must be anticipated.
Document any deviation. If ethics or setting constraints shorten the baseline, record the reason and address it as a limitation. Transparency about a constrained baseline is stronger than an unacknowledged one.
Limitations and What Baseline Cannot Do
Baseline data are necessary but not sufficient for a credible single subject study. Several limits deserve honest statement.
Baseline cannot eliminate all threats to internal validity on its own. History, maturation, and testing effects are controlled by design features such as reversal and multiple baseline, not by baseline measurement alone. A strong baseline paired with a weak design still yields weak inference.
Baseline cannot establish generality. A stable baseline in one setting says nothing about other settings, behaviors, or people. Generality requires replication across those dimensions, which is why multiple baseline designs are valued.
Baseline data are vulnerable to measurement problems. If the observation procedure changes between baseline and intervention, apparent effects may reflect the procedural shift rather than the treatment. Procedural fidelity must be documented across all phases.
Baseline cannot substitute for statistical rigor where inference is contested. Visual analysis remains primary in single subject research, but effect size indices and randomization tests are increasingly expected. These methods still depend on adequate baseline data.
Finally, the field's standards continue to evolve. Reporting guidelines, clearinghouse standards, and journal policies differ in their specific thresholds, and researchers should check the requirements of the outlet or review body they target. The core requirement, however, is stable across all of them: without repeated, reliable baseline measurement, a single subject design cannot support the causal claims it is built to make. The baseline is the study's control, its prediction, and its anchor.