Must the tiers be different people?

No. They can be participants, behaviors or settings, provided the design's independence and measurement requirements are addressed.

Does the intervention have to be withdrawn?

No. Avoiding withdrawal is one reason to consider this design.

Can every tier begin treatment on the same date?

That would remove the central staggered-start feature of a multiple baseline design.

AI Glossary / Statistics and probability / Intermediate

Multiple baseline design

Definition

A multiple baseline design is a single-case experimental design in which an intervention is introduced at different times across participants, behaviors or settings while outcomes are measured repeatedly. The staggered starts provide repeated opportunities to examine whether changes occur when the intervention begins, while other series remain in baseline. The design can be useful when withdrawing an intervention is impractical or inappropriate. Its evidential strength depends on the planned timing, repeated measurement, separation among tiers and pattern of results, not simply on having several before-and-after comparisons.

What it means

Each tier provides a baseline phase and an intervention phase. If the outcome changes only after intervention begins in each tier, at different times, that pattern can weaken explanations based on a single external event affecting everyone simultaneously. The tiers must be sufficiently independent for the comparison to be informative. If intervention in one behavior changes another untreated behavior, the remaining baseline is no longer unaffected. A concurrent design observes tiers over the same time period; nonconcurrent variants require particular care because the same calendar-time comparisons are unavailable.

How it works

Define the intervention, outcomes, measurement schedule and tiers before implementation. Collect repeated baseline observations long enough to understand their level, trend and variability. Introduce the intervention at planned staggered points, using randomization of eligible start times when the design calls for it. Continue measurement in both treated and untreated tiers. Examine changes in level, trend, variability, immediacy, overlap and consistency across replications. Use statistical analysis suited to the design and serial dependence where appropriate. Document fidelity, missing observations, concurrent events and any departures from the planned schedule.

Three practical examples

Skills research: introduce a teaching procedure at different times for several independently assessed skills.
Workflow evaluation: stagger a new operational procedure across separate teams while tracking a consistently defined outcome.
Setting comparison: evaluate an intervention across different locations when withdrawal would not be appropriate. These are possible applications, but spillover and shared external changes must be considered before treating the tiers as independent replications.

Think of it this way

Think of switching on three lamps at different scheduled times while watching the room's measured brightness near each lamp. A local change that repeatedly follows its own switch is more informative than all lamps changing together. The analogy fails if one lamp illuminates every measurement location, illustrating why spillover between tiers can undermine the design.

Worked example

Consider a deliberately simplified 14-session illustration with three independent tiers. The intervention starts at sessions 5, 8 and 11, giving baseline lengths of 4, 7 and 10 sessions. Suppose each tier's outcome is 8 units during baseline and 3 units after its own intervention begins. At session 5, the values are 3, 8 and 8; at session 8 they are 3, 3 and 8; at session 11 they are 3, 3 and 3. Each tier shows a five-unit reduction, or 62.5% relative to its baseline level. This original example illustrates staggered replication, not a claim that these phase lengths meet any particular review standard. Real data would vary, and causal interpretation would require checking measurement, fidelity, independence and alternative explanations rather than relying on phase averages alone.

Illustrative case study

In a fictional case, Birch Learning introduces a feedback procedure across three separate practice settings. The team schedules different start points and keeps measuring the settings still in baseline. One untreated setting improves immediately after the first begins, prompting investigation of shared staff practices. Researchers document the possible spillover and revise their interpretation rather than counting that tier as a clean independent replication. The report presents the timing and individual series so reviewers can assess what the design actually supports.

Common misconceptions and mistakes

Several ordinary pre-post studies do not automatically form a strong multiple baseline experiment. Simultaneous intervention starts remove the planned staggered contrast. Stable-looking averages can conceal trends or high variability, so repeated observations should remain visible. A change occurring before intervention weakens the intended demonstration. Nonwithdrawal does not eliminate all design limitations, and each tier's observations are not automatically independent over time. Generalization beyond the studied participants, behaviors or settings requires additional evidence even when the within-study pattern is persuasive.

Business use

Businesses should choose outcomes that can be measured reliably and often enough to reveal meaningful changes. Plan implementation timing, protect untreated tiers from unintended exposure where feasible and record operational events that could explain results. Assess whether delaying implementation is acceptable and whether the intervention can be delivered consistently. Report tier-specific patterns, not only a combined average. The design can support careful small-scale evaluation, but its value comes from replicated timing evidence and transparent implementation rather than a small sample being treated as automatically experimental.

Check your understanding

An untreated tier changes as soon as another tier receives the intervention. Why does this matter?

Answer: It may indicate spillover or a shared external cause, weakening that tier's role as an unaffected baseline comparison.

Related terms

Replication (statistics)
Spillover (experiment)
Single-case design
baseline phase
staggered intervention
replication
experimental control
intervention fidelity
serial dependence
spillover

Learn next

Study visual analysis of repeated observations and the distinction between concurrent and nonconcurrent baselines. Practice mapping intervention timing against alternative explanations before summarizing an effect.

Sources

What Works Clearinghouse, Single-Case Design Research in WWC Intervention Reports: https://ies.ed.gov/ncee/wwc/Docs/referenceresources/SCD-Webinar-Slides.pdf What Works Clearinghouse, Single-Case Design Technical Documentation: https://ies.ed.gov/ncee/wwc/docs/referenceresources/wwc_scd.pdf

Previous: Multioutput regression Next: Multiple comparison techniques