03 · What You Need to Know
How to Evaluate Baseline Stability Before Intervention Introduction
What is a baseline in single-case experimental research?
A baseline is a phase during which researchers repeatedly measure a target outcome under a defined condition before introducing the experimental intervention, or before changing to another planned condition.
Its purpose is to characterize how the outcome behaves without the intervention being evaluated.
For example, researchers may record the percentage of task steps a student completes independently during ordinary instructional sessions before introducing digital prompts.
Baseline observations provide information about existing performance, temporal patterns, and natural fluctuations.
However, baseline is not necessarily a no-treatment condition. Participants may continue receiving ordinary instruction, necessary accommodations, or other services.
The researcher should define precisely which conditions remain in place and which intervention component is absent.
What does baseline stability actually mean?
Baseline stability generally refers to the degree to which observations exhibit a sufficiently consistent and interpretable pattern before intervention introduction.
Researchers commonly examine three related characteristics: level, trend, and variability.
A baseline may have relatively consistent values around a typical level, show a gradual increase or decrease, or fluctuate substantially across sessions.
These characteristics affect how confidently researchers can distinguish intervention-related changes from patterns already present.
However, stability is not an absolute property established by one universal numerical threshold.
Its interpretation depends on the outcome, measurement conditions, intervention mechanism, and experimental design.
What is baseline level?
Baseline level concerns the typical magnitude of the measured outcome during the baseline phase.
For example, task completion percentages of 38%, 41%, 39%, 43%, and 42% suggest a baseline centered near 41%.
The mean or median may summarize the level, but researchers should inspect the individual observations.
A mean can conceal an increasing trend or large fluctuations that make later intervention effects difficult to interpret.
Level is therefore one component of baseline assessment rather than a complete measure of stability.
What is baseline trend?
Trend describes the direction and pattern of change over time.
A baseline may be increasing, decreasing, approximately horizontal, or characterized by changing directions.
For example, performance scores of 25%, 32%, 40%, 47%, and 55% indicate a pronounced upward pattern.
If an intervention is intended to improve performance, this existing improvement complicates interpretation because later gains may partly reflect the pre-intervention trajectory.
Conversely, a baseline moving in the opposite direction from the expected treatment effect may create a different interpretive situation.
Researchers should not automatically assume that an opposite-direction trend proves a later intervention effect. Regression toward typical performance, measurement changes, and other influences may still matter.
The direction and magnitude of baseline trend should be considered in relation to the hypothesized intervention response.
What is baseline variability?
Variability describes the extent to which observations fluctuate within the baseline phase.
Consider two hypothetical baseline sequences:
| Baseline |
Session observations |
Interpretive concern |
| Relatively consistent |
38%, 41%, 39%, 43%, 42% |
Limited fluctuation around a similar level. |
| Highly variable |
20%, 65%, 30%, 70%, 25% |
Large fluctuations may obscure intervention-related change. |
| Increasing trend |
25%, 32%, 40%, 47%, 55% |
Improvement is occurring before intervention introduction. |
The first sequence may provide a clearer reference for detecting a sustained improvement.
The second sequence requires greater attention to natural fluctuation and possible measurement inconsistencies.
The third sequence raises questions about whether improvement following intervention would exceed the existing trend.
None of these patterns can be evaluated adequately using the mean alone.
Does a stable baseline have to be horizontal?
No. Although an approximately horizontal baseline may make some intervention effects easier to interpret, a perfectly flat series is not a universal requirement.
Many behavioral and educational outcomes fluctuate naturally.
Some outcomes also exhibit predictable trends because of development, ongoing instruction, or environmental changes.
The relevant question is whether the baseline pattern permits a credible assessment of what changes when the intervention begins.
For example, an intervention expected to reduce a behavior may still produce an interpretable contrast when the baseline has a modest upward trend.
However, strong trends, especially those moving in the predicted treatment direction, can substantially weaken interpretation.
Researchers should explain how baseline characteristics influence their conclusions rather than applying a universal flatness requirement.
How many baseline observations are required?
There is no single number that is sufficient for every single-case experiment.
The appropriate phase length depends on the design, outcome variability, anticipated treatment response, and methodological standards used to evaluate the study.
For example, the What Works Clearinghouse single-case design standards distinguish levels of design adequacy according to phase observations and opportunities to demonstrate effects.
In commonly applied versions of these standards, five data points per phase are associated with meeting certain phase-length requirements without reservations, while three or four may satisfy some requirements with reservations. These numbers must be interpreted within the applicable design and standards version, not as universal rules for every phase.
Meeting a minimum observation count does not guarantee that the baseline is interpretable.
Five observations showing a strong upward trend may be less useful for a particular contrast than a longer series that clarifies the underlying pattern.
Researchers should consult the relevant standards and evaluate the actual observations before determining phase readiness.
Should researchers use a fixed stability percentage?
Some applied research traditions use operational criteria based on the proportion of observations falling within a specified range around a phase mean or median.
Such rules may help standardize decisions within a particular protocol.
However, there is no universally accepted percentage that establishes baseline stability for all single-case experiments.
A numerical criterion may conceal important trends, be sensitive to the measurement scale, or perform poorly with very few observations.
For example, observations may fall within a relatively narrow range while showing a consistent increase.
Researchers should therefore examine level, trend, and variability together.
If an operational stability rule is used, its rationale and consequences should be specified before data collection rather than selected retrospectively to justify intervention introduction.
Why does the expected direction of intervention effects matter?
Baseline patterns should be evaluated in relation to what the intervention is expected to change.
Suppose an intervention is intended to increase independent task completion.
An already increasing baseline makes it difficult to distinguish the intervention's contribution from continued improvement.
If the intervention is intended to reduce disruptive behavior, a baseline already decreasing may create a similar difficulty.
Researchers should therefore consider whether the baseline trend moves in the therapeutically or educationally desirable direction.
However, the absence of a favorable baseline trend does not by itself establish causal validity. Experimental replication and appropriate manipulation remain essential.
What should researchers do when baseline is highly variable?
First, determine whether the variability reflects the underlying phenomenon or a problem with measurement.
For example, student performance may genuinely fluctuate according to task difficulty, fatigue, or instructional context.
Alternatively, observers may apply scoring rules inconsistently or assess substantially different tasks across sessions.
Researchers should verify operational definitions, measurement reliability, and the comparability of observation conditions.
Additional baseline observations may help characterize natural variation, but they should not be collected indefinitely without a clear purpose.
If the outcome remains highly variable, the researcher may need a different measurement strategy, a more appropriate experimental design, or an analysis that explicitly addresses the variability.
Can baseline be extended until the pattern becomes stable?
Extending baseline may be appropriate when additional observations are likely to clarify an uncertain pattern.
For example, an unexpectedly high observation may warrant further measurement to determine whether it reflects ordinary fluctuation.
However, extending baseline until the data happen to look favorable can introduce researcher discretion that affects the credibility of the experiment.
Researchers should establish phase-change criteria in advance or use a defensible, documented decision process.
Repeatedly extending baseline also has practical and ethical consequences.
Participants may remain without a potentially beneficial intervention, and prolonged measurement may increase fatigue or missingness.
Additional observations should serve a defined scientific purpose rather than an arbitrary pursuit of visual perfection.
How does baseline stability matter in multiple-baseline designs?
In a multiple-baseline design, intervention introduction is staggered across participants, behaviors, or settings.
Each tier provides information about its outcome before treatment begins.
The researcher examines whether outcomes change after their respective intervention introductions while untreated tiers remain consistent with their existing patterns.
If several tiers are already improving before intervention, the staggered comparison may become difficult to interpret.
However, baseline stability should not be evaluated independently of intervention timing and replication.
A strong multiple-baseline design requires meaningful demonstrations of effect at different points in time, not merely several individually stable baseline series.
See when multiple-baseline designs are preferable for the broader experimental logic.
Can intervention timing be randomized?
Yes. Some single-case experiments randomly select intervention start points from a prespecified set of permissible occasions.
Randomization can reduce researcher discretion and support randomization-based inference under an appropriate assignment mechanism.
However, eligible start points must be scientifically and ethically acceptable.
For example, the researcher should not randomize a student to an excessively long baseline when delaying support would create unacceptable harm.
Randomization also does not eliminate the need to examine baseline patterns and intervention fidelity.
Researchers should document the permissible start points and follow the planned assignment procedure.
What if the baseline shows a ceiling or floor effect?
A ceiling effect occurs when baseline performance is already near the maximum measurable score, leaving little room for improvement.
A floor effect occurs when observations cluster near the minimum measurable score.
For example, a student completing 98% of task steps independently during baseline may have little measurable opportunity to improve on a percentage-completion outcome.
Conversely, an outcome recorded as zero across many sessions may provide limited information about the frequency or severity of a rare behavior.
Researchers should determine whether the measurement instrument is sufficiently sensitive to the anticipated intervention effect.
Changing the outcome after observing results can create interpretive problems, so measurement suitability should be considered during design planning.
How can measurement reactivity affect baseline?
Repeated assessment may influence the outcome even before the experimental intervention begins.
For example, students who repeatedly complete similar tasks may improve through practice.
Observers may also influence behavior through their presence, feedback, or changes in classroom routines.
Such effects can create trends that complicate later intervention comparisons.
Researchers should consider whether measurement itself functions as an intervention and whether equivalent assessment tasks or standardized procedures can reduce unnecessary reactivity.
These considerations are central to repeated measurement in single-case experimental research.
Does a stable baseline establish causation?
No. A stable baseline can make intervention-related change easier to identify, but it does not by itself demonstrate that the intervention caused the change.
For example, a stable baseline followed by a sudden improvement may coincide with an unrelated change in instruction or environment.
Single-case causal inference depends on systematic intervention manipulation and credible experimental contrasts.
Reversal designs may provide repeated introduction and withdrawal contrasts when effects are reversible.
Multiple-baseline designs may provide replicated changes following staggered intervention introduction.
The appropriate structure depends on the intervention and the ethical feasibility of the manipulation.
A stable baseline supports interpretation, but experimental control requires more than stability.
What if the baseline never becomes sufficiently interpretable?
Researchers should reconsider whether the original experimental arrangement can answer the question.
Possible responses include improving measurement reliability, selecting a more sensitive outcome, adjusting observation conditions, or considering another design.
However, changes to measurement procedures can compromise comparability with earlier observations and may require restarting baseline collection.
If substantial variability is intrinsic to the outcome, researchers may need to account for it explicitly rather than attempting to eliminate it.
When no ethically acceptable baseline duration can provide meaningful information, the proposed single-case experiment may be unsuitable.
Researchers should document the limitation rather than treating an unstable baseline as stable merely to proceed.
How should baseline decisions be reported?
Reports should describe the number and timing of baseline observations, operational outcome definitions, and procedures used to assess measurement reliability.
Researchers should present the baseline data graphically and discuss level, trend, and variability.
If phase-change criteria were used, they should be reported along with any departures from the planned procedure.
Unexpected baseline patterns should not be concealed through selective omission of observations.
The final interpretation should explain how baseline characteristics affect the credibility of the intervention contrast.
Transparent reporting allows readers to distinguish a sufficiently informative baseline from one that merely appears stable after selective presentation.
