Consistency across single-case replications asks whether planned demonstrations show a similar relation between condition changes and outcome patterns. Compare direction, level, trend, variability, immediacy, overlap, implementation, and context for every tier or repeated phase. Similarity does not require identical values. Preserve weak, delayed, or contradictory replications, because an average can hide the design evidence that needs the most careful explanation.

Create one row per replication

Keep each phase pair or tier visible with its raw data, predicted direction, and complete visual-analysis notes.

Compare pattern features

Review direction, size, immediacy, trend, variability, overlap, and persistence without requiring numerical identity.

Add implementation and context

Link fidelity, opportunity, access, health, partner, task, and setting information to the replication where it occurred.

Flag contradictions explicitly

A reversed direction, absent effect, or effect before treatment changes the overall interpretation and deserves a named status.

Avoid pooled concealment

Use aggregate summaries only after showing every replication and the denominator that entered the summary.

Build Jun's replication-consistency matrix

For the consistency across single-case replications question, create a versioned replication-consistency matrix. Preserve the target definition, unit, opportunity or observation time, raw series, graph, phase boundaries, design logic, predicted changes, planned demonstrations, actual implementation, measurement quality, client input, access and safety evidence, context, reviewer judgments, and claim status. The record should let another qualified reviewer reconstruct Jun's evidence without relying on a summary score.

Work through Jun's replication example

Jun's three classroom tiers rise from baseline medians 1, 2, and 1 to intervention medians 7, 8, and 3. The first two changes are immediate with little overlap. The third is smaller, delayed, and highly variable. A cross-tier summary of the reported medians looks favorable, yet the matrix identifies the third tier as inconsistent evidence requiring investigation. The example does not supply each tier's raw observations, so a true pooled median cannot be independently calculated and should not be reported from these six medians alone. Display every value, phase, denominator, and time sequence. This fictional three-classroom requesting study example illustrates one evidence pattern and does not establish a universal phase length, effect threshold, functional relation, treatment recommendation, or review outcome.

Audit Jun's design evidence

Jun's matrix keeps each tier's raw series, boundary, context, implementation, level, trend, variability, immediacy, overlap, and client report. It records that two tiers are consistent and one is ambiguous instead of assigning one global percentage to the whole design. The audit also checks graph axes, session spacing, invalid and missing states, implementation fidelity, condition discriminability, carryover, definition changes, concurrent events, software settings, and correction history. A missing field does not become a negative value or disappear from the denominator.

Address Jun's main inference risk

Equal effect sizes are rarely required, while opposite directions or missing temporal relations matter. Jun's review distinguishes acceptable contextual variation from a tier that fails to reproduce the predicted pattern. Reviewers describe the raw pattern first, then the narrower conclusion it supports. A clinical improvement can be real and valuable even when the design cannot attribute it confidently, and a strong experimental contrast can still lack personal importance.

Decide the next evidence step for Jun

The team reviews classroom opportunity, AAC access, task difficulty, staffing, measurement, and intervention delivery in the third tier. Any post hoc explanation remains labeled and the original inconsistency stays in the report. The responsible clinician and methodologist choose any added observation, replication, design change, or claim revision. The choice protects needed care, uses prospectively stated rules where possible, and preserves the original graph and decision history.

Protect Jun's access and participation

Keep Jun's augmentative and alternative communication, interpreters, mobility, food, water, bathroom use, prescribed care, health support, rest, relationships, and emergency help available throughout the three-classroom requesting study. Use accessible consent and assent processes when applicable and respond to withdrawal, dissent, or distress. A replication plan never supplies authority to withhold a necessary support or delay urgent action.

Apply current sources to Jun's review

Jun's sources support evaluating similarity of patterns across comparable phases while retaining the distinctive context of each replication. The BACB ethics hub and CASP public summary provide professional context, while the BCBA Test Content Outline identifies examination content on measurement and single-case design. The WWC Version 5.0 handbook supplies a current research-review framework. A single-case design review, systematic visual-analysis protocols, current analytic reflections, nonconcurrent multiple-baseline methods research, and a visual-analysis software tutorial describe methods and limits. ASHA supports continuous AAC access.

Rehearse Jun's review process

Before live use, run the replication-consistency matrix on fictional graphs with a strong effect, weak effect, delayed effect, opposite trend, carryover, missing agreement, one contradictory tier, and no replication. Confirm that reviewers preserve raw values, count planned opportunities separately from observed demonstrations, identify holds, and keep client relevance distinct from causal inference. Store test cases, expected decisions, software version, and correction log.

Close Jun's claim review

Review the replication-consistency matrix with Jun, the responsible clinician, and a methodologist familiar with the exact design. Preserve raw data, graph, design protocol, basic effects, demonstrations, fidelity, measurement checks, contextual events, client input, reviewer disagreements, source standard, final wording, and limits. Keep the page draft and noindex until all manifest-named reviews are complete.

Related resources

Sources