Baseline logic

Prediction starts with a measured data path

A baseline is a defined condition measured repeatedly before or apart from the independent variable of interest. Its level, trend, variability, context, and measurement process support a prediction about future responding if that condition continues.

A baseline phase differs from a baseline value. One initial score provides a reference point; it rarely shows whether responding is stable, trending, cyclical, or highly variable. Repeated observations make the prediction visible and expose features that can complicate later comparison.

The BACB BCBA Test Content Outline, 6th edition lists repeated measures, individuals serving as their own controls, prediction, verification, and replication among defining single-case design features. It also covers measurement validity and reliability, graphing, procedural-integrity data, internal validity threats, interpretation, and application. The outline describes examination content rather than a complete design protocol.

Verification tests whether the prediction remains credible

Verification asks whether evidence supports the baseline prediction after the independent variable is manipulated. Its form depends on the design.

• In a reversal design, a return to the comparison condition tests whether responding moves toward the earlier baseline pattern.
• In a concurrent multiple-baseline design, untreated tiers can remain under baseline while the intervention begins in another tier.
• In a changing-criterion design, repeated criterion shifts test whether responding tracks each planned value.
• In an alternating-treatments design, rapid repeated contrasts can test differentiation even without a separate initial baseline phase.

The mapping is therefore conceptual rather than a universal sequence of steps. A methodological analysis of multiple-baseline design variations discusses limits of the traditional verification account and proposes prediction, contradiction, and replication as a broader framing. The debate is useful: labels cannot replace a design-specific analysis of plausible alternative explanations.

Replication repeats the effect test

Replication means the predicted contrast appears again at another point in time, tier, condition, person, setting, or experiment, as the design specifies. Within-case replication can strengthen a causal inference for that case. Replication across people or settings addresses a different question about generality.

The WWC Single-Case Design Technical Documentation explains that single-case designs use structured repetition to support inference. The current WWC handbook page identifies Version 5.0 as its current procedures and standards. Its single-case research ratings require opportunities to demonstrate an intervention effect at different times, along with design-specific observation rules.

Those research-rating thresholds should not become an automatic clinical phase-change rule. Decisions also depend on the data pattern, client priorities, risk, benefit, feasibility, the design question, and the predeclared decision rule.

A predictable baseline is contextual

“Stable” need not mean a perfectly flat line. A consistent trend can support a prediction when the design and analysis can distinguish it from the expected intervention effect. A therapeutic baseline trend, meaning improvement already occurring in the desired direction, can make an intervention contrast ambiguous.

A study of waiting for baseline stability used simulation to examine fixed, response-guided, and random baseline lengths. Its results challenge the assumption that waiting longer always improves inference. The study is methodological evidence, not a rule to ignore trend or variability. Predefine phase-change criteria, examine the actual graph, and explain departures from the plan.

Avoid stopping baseline as soon as a convenient point appears. Selective phase changes can make random fluctuation resemble an effect. At the same time, gathering observations solely to reach a preferred count can waste time or prolong a condition that no longer fits.

Measurement and implementation anchor the logic

Define the dependent variable, unit, opportunity, observation window, prompts, exclusions, and missing-data rule. Graph observations in time order. Inspect level, trend, variability, immediacy, overlap, and consistency across similar phases or tiers.

Measure the independent variable separately. A claimed treatment effect is hard to interpret when implementation changed gradually, appeared during baseline, or varied with the observer, partner, setting, or time of day. Record procedural integrity, concurrent events, medication or health changes, access barriers, and other plausible influences.

Train and calibrate observers when measurement drives a causal conclusion. Sample interobserver agreement across relevant conditions and report it separately. High agreement cannot repair an outcome measure that misses what matters to the person.

Baseline is a condition, not deprivation

A baseline should retain ordinary communication, AAC, health, safety, mobility, sensory, and relationship supports. It should never require withholding food, water, bathroom access, prescribed care, pain care, emergency help, or an effective safety protection. Define what remains available and what variable changes.

The current BACB Ethics Code applies to BCBA and BCaBA certificants and people who completed an application. It addresses competence, client involvement, informed consent and assent when applicable, assessment, intervention, risk, data, and evaluation. Withdrawal, distress, benefit, or new risk can require a design change even when that weakens the planned demonstration.

A fictional baseline-logic example

Three fictional adult trainees voluntarily complete the same de-identified chart-audit simulation. A checklist is introduced at staggered calendar times. Before checklist access, Dara completes 1 of 5, Eli 1 of 7, and Mina 2 of 9 audits accurately. These are separate participant denominators.

When Dara receives the checklist, Dara completes 5 of 6 audits accurately. During the same period, Eli and Mina each remain accurate on 1 of 3 baseline audits. Eli then receives the checklist and completes 5 of 6, while Mina remains at 1 of 3 in baseline. After the checklist reaches Mina, Mina completes 5 of 6.

The unchanged concurrent tiers help evaluate the prediction while staggered improvement supplies replications. The pattern supports a functional relation in this simulation if measurement, checklist fidelity, tier independence, and concurrent-event evidence are credible. Shared practice, unequal task difficulty, history, or communication between trainees could still weaken the inference. Results establish neither effects on client care nor general benefit outside the tested task.