Single-case metric sensitivity analysis applies several preselected, defensible metrics to one locked dataset and traces divergent results to reference, tie, distance, trend, and denominator rules. It is useful when one summary may be fragile. Record the analysis plan before calculation, report every selected result, and integrate convergence or divergence with raw visual analysis, client relevance, and design evidence.

Preselect the metric set

Name each metric and why its information target fits the review question. Include the formula version and tie rule before calculating.

Lock the shared data

Use identical phase values, improvement direction, exclusions, and timestamps across metrics. Document any metric that cannot use the same data.

Calculate each result transparently

Show reference values, numerators, denominators, pairs, ties, and rounding. Preserve software output and an independently checked example.

Trace divergence to rules

Identify whether the baseline median, extreme, pair weighting, distance, trend, or phase size explains different results.

Report sensitivity as a finding

State whether conclusions converge or depend on the metric. Do not average incompatible percentages into a synthetic score.

Build Ari's locked-dataset comparison grid

Start with a locked worklist for Ari. Record the review question, metric name, formula version, phase labels, improvement direction, planned and valid observations, missing or invalid states, tie handling, trend handling, source timestamps, calculation owner, and review date. The multi-metric sensitivity analysis result should be reproducible from this worklist without relying on an unlabeled dashboard value. Keep the data in their original unit so reviewers can connect the supplement to the graph and to the actual outcome. For the single-case metric sensitivity analysis question, Ari's reviewer also records the release criterion, expected evidence, unresolved limitations, and exact action that the result may inform. This small decision log prevents a calculation from becoming an open-ended label and lets a later reviewer distinguish the observed value from the judgment made with it.

Work the calculation for Ari

Ari's baseline values are 2, 4, and 6; intervention values are 3, 5, 6, and 8; higher is preferred. PEM is 3/4, or 75%; PND is 1/4, or 25%; NAP is 8.5/12, or 70.8%; uncorrected Tau is (8 minus 3)/12, or 41.7%. The results differ by design, not arithmetic error. Show every intermediate count or statistic at enough precision to reproduce the displayed result. Keep the raw phase values in chronological order beside any sorted, paired, or transformed table. The fictional arithmetic illustrates the method; it does not create a clinical threshold, minimum phase size, or promised treatment effect.

Audit Ari's denominator and formula

Ari's grid uses the same seven raw observations, direction, phase labels, and validity states. It records median 4, maximum 6, eight improved pairs, three deteriorated pairs, one tie, and the exact formula version for each result. The audit also checks duplicate timestamps, silent imputation, phase-boundary drift, direction reversal, premature rounding, spreadsheet ranges, software version, and correction history. Any unresolved source discrepancy stays on hold with an owner and due date instead of being converted into a convenient zero or exclusion.

Watch for the main failure mode in Ari's review

Calculating many metrics and publishing only the largest is metric shopping. Adding a new metric after seeing an inconvenient value has the same risk. Ari's review locks the selected set and purpose before the calculations are generated. Reviewers should be able to see how one point, tie, extreme, missing observation, or phase-length decision affects the result. A sensitivity example is labeled as hypothetical and never mixed with observed evidence.

Integrate the metric with visual and design analysis for Ari

The team explains that PND is controlled by one baseline extreme, PEM by one median, NAP by pair ordering with half-credit ties, and Tau by the signed pair balance. It then evaluates level, trend, variability, immediacy, overlap, phase replications, and whether the change matters to Ari. The WWC Version 5.0 handbook is a research evidence standard, so its design criteria are not a universal clinical protocol. In care, the responsible clinician also considers current assessment evidence, professional scope, the treatment plan, ordinary supports, risk, and the person's choices.

Protect client relevance and access for Ari

During Ari's how conclusions change across defensible metrics review, keep augmentative and alternative communication, interpreters, mobility, food, water, bathroom use, prescribed care, health support, rest, relationships, and emergency help available. Use accessible communication to ask whether the goal, direction, magnitude, burden, and observed change matter to Ari. A favorable coefficient cannot repair an unwanted target, inaccessible measurement process, unsafe plan, or missing consent and assent process.

Use current evidence within scope for Ari

For Ari's multi-metric sensitivity analysis review, the BACB ethics hub and CASP public summary provide professional context, while the BCBA Test Content Outline supplies examination scope for measurement, graphing, interpretation, experimental design, and data-based evaluation. A single-case methods review describes common overlap, Tau, and two-SD calculations. Decision-accuracy research shows why phase size and design structure matter and why overlap percentages do not measure distance. A reproducibility tutorial documents ambiguity among Tau-U implementations and baseline corrections. Nonoverlap and mean-difference methods explain their distinct information targets. ASHA supports continuous access to AAC tools or devices.

Close Ari's supplemental analysis review

Review the locked-dataset comparison grid with Ari and the responsible qualified clinician. Preserve source data, graph, phase definitions, calculation specification, coverage, ties, uncertainty, limitations, client input, selected action, owner, and next review date. Reopen the analysis when the measure, phase, context, access, health, goal, software, or design changes. The final note should make clear which conclusions are supported, which remain uncertain, and which require different evidence.

Related resources

Sources