How Many Baseline Measurements Are Better Than One?

Is One Pre-Intervention Measurement Enough to Establish a Baseline?

You measure an outcome once before introducing an intervention. After implementation, you measure it again and find an improvement. But what if the original score was unusually low? What if the outcome was already improving before the intervention began?

A single baseline provides a snapshot. It cannot, by itself, show whether the outcome was stable, fluctuating, or following an existing trend. Collecting several pre-intervention measurements might reveal a different starting condition and lead to a different interpretation of subsequent change.

Yet repeated measurement also requires time and resources. For some outcomes, additional observations may offer little useful information. The methodological question is therefore how many baseline measurements are needed to characterize the outcome adequately and support the intended inference.

There Is No Universal Number of Baseline Measurements

How Repeated Baseline Measurements Improve the Interpretation of Change

A baseline is not necessarily a single observation. It can also be a period during which an outcome is measured repeatedly before an intervention begins.

These observations may help establish the outcome's typical level, variability, and direction of change. Their usefulness depends on whether they capture meaningful information rather than simply repeating the same measurement.

What Can One Baseline Measurement Tell You?

A single baseline establishes the observed value of an outcome at a particular point before the intervention.

For example, a student's research-literacy score of 62 before an instructional program provides a reference value for a later assessment. If the posttest score is 78, the observed difference is 16 points.

However, that single baseline does not reveal whether the student ordinarily scores around 62, whether the score was affected by temporary circumstances, or whether the student's knowledge had already been increasing.

A single measurement can be useful, especially in randomized trials where the primary analysis compares outcomes between assigned groups. Nevertheless, the adequacy of the measurement still depends on its reliability, validity, and timing.

What Do Additional Baseline Measurements Reveal?

Repeated baseline observations can provide information that a single pretest cannot.

They may reveal ordinary fluctuations, an upward or downward trend, recurring cycles, unusual observations, or changes occurring before formal intervention exposure.

For instance, weekly measurements of student engagement may show that participation was already increasing before a new learning platform was introduced. If engagement continues increasing afterward at approximately the same rate, a simple before-and-after comparison could exaggerate the platform's apparent contribution.

Repeated observations can also reveal whether the outcome is sufficiently stable to support a meaningful comparison with later measurements.

Baseline Level, Trend, and Variability Are Different Properties

When interpreting a baseline series, researchers should distinguish its typical level from its trend and variability.

A baseline can be stable in one sense but not another. For example, observations may follow a consistent upward trend with little random variation. Such a series is predictable, but its level is not constant.

Researchers should avoid treating the absence of large fluctuations as proof that the outcome would have remained unchanged without the intervention.

How Many Measurements Are Needed to Detect a Trend?

Two observations can establish a difference between two time points, but they provide little information about the shape, stability, or persistence of an underlying trend.

Three observations allow a simple pattern to be examined, but they remain highly sensitive to individual observations. More measurements generally provide better opportunities to distinguish systematic movement from random variation.

There is no universal threshold at which a trend suddenly becomes reliable. The required number depends on the strength and shape of the trend, the outcome's variability, temporal dependence between observations, and the analysis.

Importantly, observations collected close together may contain less independent information than their number suggests. A sequence of highly correlated daily measurements cannot necessarily be treated as equivalent to the same number of independent observations.

Why the Measurement Interval Matters as Much as the Number

Ten measurements collected within one hour do not necessarily provide the same information as ten measurements collected across ten weeks.

The first schedule may characterize short-term fluctuations. The second may reveal longer-term changes, depending on the outcome.

Researchers should choose the spacing between observations according to the expected rate of change and the purpose of the baseline period.

For rapidly fluctuating outcomes, frequent observations may be appropriate. For slower processes, a longer observation window may be necessary even when the total number of measurements is relatively small.

Repeated measurement should therefore be understood as a sampling strategy over time, not merely an exercise in accumulating data points.

When Does Averaging Several Baselines Help?

When an outcome fluctuates around a relatively stable level, averaging repeated measurements may provide a more precise estimate of that level than relying on one observation.

For example, suppose a participant's daily engagement scores are 48, 55, 51, 46, and 50. Their mean baseline score is 50.

The average summarizes the five observations and reduces the influence of individual fluctuations. However, the precision gain depends on measurement properties and the degree of correlation among observations.

If scores instead increase steadily from 40 to 60, the average may obscure the fact that the participant was already improving before intervention implementation.

Researchers should not automatically average all pre-intervention measurements. The appropriate summary depends on whether the outcome is stable, trending, or following a more complex pattern.

Can Repeated Baselines Reduce Regression to the Mean?

Repeated baseline observations can help identify whether an unusually high or low score is representative of the participant's usual performance.

Suppose students are selected for an intervention because they obtained exceptionally low scores on a single assessment. Additional pre-intervention observations may reveal that some of those initial scores were temporary departures from their typical level.

This information is useful because regression to the mean can produce apparent improvement after selection based on extreme observations.

Nevertheless, collecting additional baseline measurements does not automatically remove regression-to-the-mean bias. Selection rules, measurement error, the outcome's underlying distribution, and the comparison design still matter.

How Many Baseline Points Does an Interrupted Time-Series Study Need?

An interrupted time-series design uses repeated observations before and after an intervention to examine whether the outcome's level or trend changes around implementation.

Unlike a conventional two-time-point pretest-posttest study, it requires enough observations to characterize the pre-intervention trajectory and assess changes after the intervention.

Some methodological guidance has used a minimum of three observations before and after an interruption as an eligibility threshold. However, such a minimum should not be interpreted as generally sufficient for reliable estimation.

Methodological discussions commonly recommend substantially longer series, sometimes around 12 observations before and 12 after, for particular segmented-regression applications. These figures are planning heuristics rather than universal statistical requirements.

The necessary number depends on autocorrelation, outcome variability, seasonality, the anticipated effect, and whether researchers seek to estimate an immediate level change, a change in slope, or both.

For a more detailed design decision, researchers should consider when an interrupted time-series design offers stronger evidence than a simple before-and-after comparison.

How Many Baseline Points Are Needed in Single-Case Experimental Designs?

Single-case experimental designs repeatedly measure an outcome within individual participants, behaviors, or settings while systematically manipulating intervention conditions.

Baseline observations are particularly important because researchers examine patterns of level, trend, and variability across phases.

The What Works Clearinghouse has established design-specific standards for single-case research. Its technical guidance has distinguished phases containing at least five data points from phases containing three or four, with different implications for meeting standards.

These thresholds belong to particular single-case design standards and should not be applied indiscriminately to group-based pretest-posttest studies or interrupted time-series analyses.

Researchers should consult the current What Works Clearinghouse standards and the requirements of their intended review framework, since criteria may differ by design and version.

Repeated Baseline Measurements Are Not the Same as a Multiple-Baseline Design

Collecting five baseline measurements from one group does not, by itself, create a multiple-baseline experimental design.

The experimental design depends on staggered intervention introduction, systematic repeated measurement, and appropriate replication across tiers. The distinction matters because the two approaches support different levels of causal inference.

When Are Additional Baseline Measurements Unnecessary?

Repeated baselines are not automatically advantageous.

For some relatively stable characteristics, repeated measurement over a short period may contribute little new information. In a properly randomized trial, one high-quality pre-intervention outcome measurement may be sufficient for the intended baseline-adjusted analysis.

Additional testing can also create burden, fatigue, learning effects, or changes in participants' behavior. These effects may undermine the very baseline stability that repeated assessment is intended to establish.

The relevant decision is whether another observation would materially improve the characterization of the starting condition or the credibility and precision of the planned analysis.

Can Baseline Collection Continue Until the Data Look Stable?

In some single-case experimental applications, researchers examine baseline stability before introducing an intervention. This may be methodologically justified when the decision rule is defined in advance and consistent with the design.

However, repeatedly extending the baseline until the data display a favorable pattern can introduce researcher discretion and analytical bias.

Researchers should specify how stability will be assessed, what conditions permit intervention introduction, and whether any maximum baseline duration applies.

When the baseline remains unstable, researchers should consider whether the outcome, measurement schedule, or proposed intervention timing needs reconsideration rather than assuming that more observations will necessarily solve the problem.

How Multiple Baseline Measurements Change the Interpretation of Improvement

If researchers used only Week 1 as the baseline, they would report a 15-point increase by Week 6. Using Week 5 instead, they would report a 3-point increase.

Neither comparison fully describes the underlying pattern. The repeated baseline measurements reveal that the cohort was already improving by exactly 3 points per week before the intervention.

The post-intervention score therefore provides no visible departure from the previously observed linear pattern.

This does not prove that the intervention had no effect. A single post-intervention observation provides little information about a possible delayed effect or subsequent change in trajectory. The linear continuation is also a hypothetical projection rather than an observed counterfactual.

Nevertheless, the five baseline measurements fundamentally change the interpretation. What initially looked like a 15-point improvement following intervention implementation is revealed to be consistent with an established pre-intervention trend.

The researchers should continue collecting post-intervention observations and consider a design and analysis capable of examining changes in level or slope. If feasible, a suitable comparison series would provide additional evidence about broader changes affecting the cohort.

Common Mistakes When Planning Multiple Baseline Measurements

How to Determine the Appropriate Number of Baseline Measurements

Begin with the purpose of baseline measurement. Are you establishing a single starting value, estimating ordinary variability, identifying an existing trend, or developing a counterfactual trajectory?

Next, consider how the outcome behaves over time. Stable outcomes may need fewer observations than rapidly fluctuating or strongly seasonal outcomes. However, the length of the observation window is often as important as the number of data points.

Finally, match the measurement schedule to the intended analysis and research design.

When feasible, examine pilot data or existing records to estimate outcome variability and temporal dependence before finalizing the schedule. For formal time-series analyses, consider simulation-based planning or other design-appropriate methods rather than relying solely on a fixed number of observations.

The measurement plan should specify the baseline period, assessment frequency, instrument, handling of missing observations, and criteria for excluding invalid measurements. If the intervention start date can vary, the rules governing that decision should also be established in advance.

Repeated baseline measurements are most valuable when they answer a question that one observation cannot. The aim is not to maximize the number of pretests, but to collect enough relevant information to interpret subsequent change responsibly.

Before Choosing the Number of Baseline Measurements, Check:

Frequently Asked Questions About Repeated Baseline Measurements

Collect Enough Baseline Data to Understand What Was Already Happening

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.