03 · What You Need to Know
How Multiple-Baseline Designs Establish Experimental Control
What is a multiple-baseline design?
A multiple-baseline design is a single-case experimental arrangement in which an intervention is introduced at different points across two or more tiers, such as participants, behaviors, or settings.
Researchers repeatedly measure the target outcomes before and after intervention introduction. The intervention begins in one tier while other tiers remain under baseline conditions, allowing the researcher to observe whether their outcomes change before receiving treatment.
For example, a prompting intervention may begin with Student A after five sessions, Student B after eight sessions, and Student C after eleven sessions.
If each student's performance changes following their respective intervention introduction, while untreated tiers do not show comparable changes beforehand, the pattern may strengthen evidence of a functional relationship.
However, the design must provide adequate observations and opportunities for replication. Merely assigning different intervention start dates does not establish experimental control.
Why might multiple baseline be preferable to ABAB?
ABAB designs depend on intervention introduction, withdrawal, and reintroduction to create repeated experimental contrasts.
This approach can be informative when treatment effects are sufficiently reversible. It becomes problematic when an intervention produces durable learning, irreversible change, or benefits that should not be withdrawn.
Multiple-baseline designs use staggered intervention introduction instead of reversal. Participants who receive the intervention can ordinarily continue receiving it while other tiers remain under baseline conditions until their scheduled introduction.
This may be especially appropriate for instructional interventions designed to develop lasting knowledge or skills.
Nevertheless, delaying treatment can create ethical concerns of its own. The comparison should therefore consider both the scientific value of staggered introduction and the consequences of waiting.
The broader ethical distinction is examined in when withdrawing an intervention is unethical or impractical.
How does staggered intervention introduction support causal inference?
The design creates a series of planned opportunities to examine whether outcome changes correspond to intervention introduction.
Suppose three students are measured repeatedly. Student A receives treatment first, while Students B and C continue baseline measurement.
If A improves but B and C remain near their existing patterns, the result is consistent with an intervention effect. When B subsequently receives treatment, researchers examine whether B changes while C remains under baseline conditions.
A comparable change following C's later intervention introduction provides another opportunity for replication.
The staggered sequence helps address some historical explanations because a single event affecting everyone at the same time would not ordinarily produce changes aligned with three different intervention start points.
However, tier-specific events, anticipation, contamination, or coincident changes may still threaten inference.
What is a tier in a multiple-baseline design?
A tier is a distinct experimental series across which intervention introduction is staggered.
Depending on the design, tiers may represent different participants, behaviors, or settings.
| Tier structure |
Illustrative arrangement |
Important consideration |
| Across participants |
Introduce prompting to different students at different sessions. |
Students should not receive the intervention indirectly before their scheduled introduction. |
| Across behaviors |
Introduce an intervention sequentially to distinct target behaviors. |
Changes in one behavior should not automatically produce changes in untreated behaviors. |
| Across settings |
Introduce a procedure sequentially in different instructional environments. |
Intervention effects should not spread immediately into untreated settings. |
The tiers must be sufficiently distinguishable for staggered intervention introduction to produce meaningful experimental contrasts.
For a detailed comparison, see multiple-baseline designs across participants, behaviors, and settings.
How many tiers are needed?
Three tiers are commonly used because they can provide three opportunities to demonstrate an intervention effect at different points in time.
However, the number of tiers alone does not establish that a study meets a particular methodological standard.
The What Works Clearinghouse single-case design standards emphasize demonstrations of effect at different points in time and sufficient observations within relevant phases.
For example, three nominal tiers may be inadequate if intervention introductions occur almost simultaneously or if outcomes change before the intervention begins.
Conversely, additional tiers may strengthen replication but increase study burden and the time some participants spend awaiting treatment.
Researchers should determine the required design features using the applicable standards and the anticipated outcome patterns rather than assuming that a fixed number of participants guarantees validity.
How far apart should intervention start points be?
Intervention start points should be separated sufficiently to create distinguishable opportunities to evaluate changes.
For example, beginning interventions after sessions five, six, and seven may provide less temporal separation than beginning after sessions five, nine, and thirteen, depending on the outcome's response time.
However, there is no universally appropriate interval.
The spacing should reflect baseline variability, the expected delay before treatment effects become observable, the measurement schedule, and ethical constraints.
Researchers should avoid selecting start points merely because they create a visually attractive graph.
The schedule should be prespecified or governed by documented phase-change criteria established before examining favorable outcome patterns.
What should happen in tiers that have not received the intervention?
Tiers awaiting intervention should continue under their defined baseline conditions while outcome measurement proceeds.
The purpose is to determine whether those outcomes remain consistent with their existing patterns while another tier receives treatment.
For example, if Student A begins improving after intervention introduction, researchers examine whether Students B and C also improve before receiving treatment.
If all three students improve at the same time, a shared historical event, instructional change, or measurement effect may be a plausible explanation.
However, untreated tiers need not display perfectly flat baselines. Researchers should evaluate level, trend, variability, and whether changes are consistent with the hypothesized intervention effect.
These observations provide the comparison structure central to the design.
How stable must baselines be?
Baseline observations should be sufficiently informative to characterize the outcome before intervention introduction.
Researchers commonly examine baseline level, trend, and variability, particularly whether the outcome is already changing in the direction expected from treatment.
A baseline that improves rapidly before intervention may make subsequent improvement difficult to attribute.
However, requiring a perfectly stable baseline may be unrealistic for naturally variable behaviors.
Additional baseline measurement may clarify the pattern, but extending baseline can increase burden or delay beneficial treatment.
The decision should follow the principles discussed in baseline stability in single-case research.
What is the difference between concurrent and nonconcurrent multiple baselines?
In a concurrent multiple-baseline design, tiers are observed over overlapping calendar periods, allowing researchers to compare treated and untreated tiers during the same general time frame.
For example, three students may be monitored simultaneously while treatment begins at different sessions.
In a nonconcurrent multiple-baseline design, tiers may be studied during different calendar periods.
Nonconcurrent arrangements can be useful when participants become available at different times or simultaneous recruitment is impractical.
However, they generally provide less direct protection against historical events because tiers do not necessarily experience the same external conditions.
Randomizing eligible intervention start points and documenting contextual conditions may strengthen selected nonconcurrent designs, but neither procedure eliminates all threats.
Researchers should identify whether the design is concurrent or nonconcurrent and justify the implications for experimental control.
Can intervention start points be randomized?
Yes. Researchers may randomly select intervention start points from a prespecified set of permissible occasions.
Randomization can reduce researcher discretion over when treatment begins and may support randomization-based statistical inference.
However, the permissible start points must satisfy ethical and methodological constraints.
For example, a participant requiring urgent support should not be assigned to an excessively long baseline solely to create a randomization distribution.
The assignment mechanism must also be documented and reflected in any randomization test.
Randomization strengthens selected aspects of design credibility but does not replace reliable measurement, intervention fidelity, or sufficient experimental contrasts.
When is a multiple-probe design preferable?
A multiple-probe design is related to multiple baseline but uses intermittent baseline assessments rather than continuous measurement throughout every untreated phase.
It may be useful when repeated assessment is burdensome, impractical, or likely to influence performance.
For example, repeatedly testing a student's ability to complete a complex task may itself provide practice that changes the outcome.
Intermittent probes can reduce unnecessary assessment while retaining strategically timed observations needed to evaluate intervention effects.
However, insufficient probing may make it difficult to characterize baseline patterns or identify changes occurring before intervention.
Researchers should follow design-specific standards rather than treating multiple probe as permission to collect only occasional measurements without an experimental rationale.
How can contamination undermine the design?
Contamination occurs when intervention exposure or its effects reach tiers that have not formally entered the intervention phase.
For example, a teacher may begin using the prompting strategy with all students after observing its apparent success with the first participant.
Students may also share instructional materials or strategies.
If untreated tiers receive meaningful elements of the intervention, their baseline conditions no longer represent the intended comparison.
Researchers should monitor intervention fidelity and document whether procedures remain distinct across tiers.
Where contamination is unavoidable, a multiple-baseline arrangement across those tiers may not provide credible experimental control.
What happens when the intervention affects multiple behaviors?
Some interventions are designed to produce generalization across related behaviors.
For example, teaching a general problem-solving strategy may improve several academic tasks simultaneously.
That outcome may be educationally desirable, but it can undermine a multiple-baseline design across behaviors if untreated behaviors change before their scheduled intervention introduction.
Researchers should distinguish intervention generalization from a failure of the measurement system.
The appropriate design depends on whether the behaviors are sufficiently independent for staggered treatment to provide meaningful contrasts.
When substantial cross-behavior generalization is expected, a design across participants or settings may be more suitable, provided those tiers are appropriately independent.
How should outcomes be analyzed?
Visual analysis commonly examines level, trend, variability, immediacy, overlap, and consistency across comparable phases.
The graph should display repeated observations in chronological order and identify intervention start points clearly.
Researchers should examine whether each tier changes after its intervention introduction and whether untreated tiers remain consistent with their existing patterns.
Statistical methods may supplement visual analysis, including appropriate randomization tests or effect-size estimators.
However, temporal dependence, baseline trends, and the small number of tiers may complicate inference.
The design should not be judged solely by a statistically significant test or a difference between phase averages.
Can multiple baseline establish causality?
A well-designed multiple-baseline experiment can provide evidence of a functional relationship when outcome changes correspond systematically with staggered intervention introductions.
However, the strength of causal inference depends on the adequacy of the contrasts and the exclusion of plausible alternative explanations.
For example, a shared event affecting all participants at the same time would be less consistent with changes appearing only after distinct treatment start points.
Yet tier-specific events, intervention contamination, and measurement changes may still threaten interpretation.
Researchers should distinguish a convincing pattern within the studied cases from evidence that the same effect will occur in every other population or setting.
When is multiple baseline not appropriate?
A multiple-baseline design may be unsuitable when the intervention cannot be introduced separately across tiers, when untreated tiers cannot remain meaningfully unexposed, or when intervention effects spread immediately across all tiers.
It may also be inappropriate when delaying intervention would create unacceptable harm.
For example, a schoolwide policy implemented simultaneously across every classroom may not permit genuine staggered introduction unless the institution can ethically and operationally phase implementation.
Likewise, a highly urgent intervention may not permit extended baseline observation.
Researchers should not force the design merely because withdrawal is undesirable. An alternative experimental or observational approach may be more defensible.
What should be planned before data collection?
The protocol should define the tiers, outcome measures, intervention conditions, start-point rules, and expected timing of effects.
Researchers should establish how baseline patterns will be evaluated, how intervention fidelity will be monitored, and how missing observations will be documented.
The design should provide sufficient opportunities to demonstrate an effect at different points in time under the methodological standards being used.
Ethical review should consider the consequences of delaying intervention for tiers awaiting treatment.
Where relevant, the protocol should also specify whether start points are randomized and how the analysis will respect that mechanism.
A multiple-baseline design should be selected because its staggered contrasts are scientifically and ethically meaningful, not simply because it avoids withdrawal.
