Experimental design Glossary

Name the variables and question

An independent variable is the condition or intervention deliberately changed for evaluation. A dependent variable is the measured outcome expected to change as a function of that manipulation. Define both in observable, replicable terms.

A confounding variable changes with the independent variable and offers another explanation for the result. Staff, setting, materials, time, instruction, measurement, or access can become confounds when they shift at the same boundary.

A functional relation is a replicable demonstration that changes in the independent variable reliably correspond with changes in the dependent variable under a design that addresses plausible alternatives. One before-and-after change is usually a hypothesis-strengthening observation rather than that demonstration.

The BACB BCBA Test Content Outline includes measurement, graphing, visual analysis, single-case designs, internal validity, and data-based decisions as examination content. It is not a case protocol.

Use baseline logic deliberately

Baseline logic uses prediction, verification, and replication to evaluate whether an intervention produced a change. Prediction states what would likely occur without the change. Verification tests that prediction. Replication repeats the effect across phases, behaviors, settings, people, or criteria.

An ABAB design introduces, withdraws, and reintroduces an independent variable. A reversal design arranges phase changes to show responding reverses with conditions. A withdrawal design removes an intervention, though the response may not return to baseline. These labels overlap in practice, so describe the actual sequence.

Do not withdraw an effective support when return to baseline could create harm, loss of an acquired skill, serious disruption, or an unacceptable violation of client preference. Select another design when reversal is unsafe, infeasible, or unlikely.

Replicate across tiers or criteria

A multiple-baseline design staggers intervention across people, behaviors, or settings. A multiple-probe design uses intermittent baseline measurement where continuous measurement would be impractical, reactive, or unnecessary. Both need planned tier logic and enough opportunities to evaluate change.

A changing-criterion design evaluates responding as successive performance criteria change. The criteria, step sizes, phase lengths, range, and reversals or bidirectional shifts affect interpretability. A person-centered goal should not become an endless demand for higher performance.

The What Works Clearinghouse single-case standards and handbooks provide evidence-review criteria within their program. They are useful methodological references, not universal clinical requirements.

Compare conditions with sequence controls

An alternating-treatments design, often called a multielement design, rapidly alternates conditions to compare differentiated response patterns. Distinctive cues, balanced scheduling, enough exposure, and stable measurement help interpretation.

A carryover effect occurs when one condition influences responding in a later condition. A sequence effect occurs when order affects the result. Randomization, counterbalancing, washout periods, condition cues, and design changes may reduce these risks, depending on the question.

Peer-reviewed discussions of alternating-treatments designs and single-case quality show why arrangement and reporting details matter. No design name repairs poor measurement or uncontrolled co-change.

Match the analysis to the decision

A comparative analysis asks which of two or more conditions produces the more useful result. A component analysis asks which elements of a treatment package contribute to its effect. A parametric analysis varies the value of one dimension, such as duration, magnitude, or schedule, to examine how outcomes change.

Each question needs adequate exposure, a feasible sequence, client involvement, and a decision rule. Efficiency should be considered with comfort, safety, social validity, burden, maintenance, and generalization.

Separate internal and external validity

Internal validity is the confidence that the independent variable produced the observed change in the studied case and conditions. External validity concerns how findings extend across people, behaviors, settings, implementers, materials, and time.

Strong internal validity in one context does not promise the same result elsewhere. Replication, transparent case description, implementation evidence, ordinary-support conditions, and direct client feedback help readers judge relevance.

Interpretability also depends on measurement and implementation. Define eligible opportunities, observation windows, missing-data rules, prompting, ordinary supports, intervention fidelity, and observer agreement before the phase begins. Mark a phase boundary when the independent variable actually changes, rather than when the team intended it to change. Report any simultaneous staff, setting, schedule, medication, health, or access change. Visual analysis should consider level, trend, variability, immediacy, overlap, and consistency where they fit the design, alongside the client's view of importance and burden.

A design-choice example

Ravi is a fictional sixteen-year-old who chooses a goal of starting a preferred community class with less waiting. Withdrawing the visual schedule would create unnecessary disruption, so the team avoids reversal. It considers a multiple baseline across three class routines, with staggered starts and the same defined opportunity window.

The design could strengthen causal confidence if change occurs only after each staggered start. It still needs fidelity, opportunity counts, access conditions, Ravi's view, and alternative-explanation review. The design choice itself establishes no effect.

Explore clinical roles at Finni practices. Confirm the practice's research literacy, measurement support, clinical governance, consent, assent, supervision, and safety review during diligence.