Between-condition analysis

Start inside each condition

Between-condition interpretation begins with a clear picture of each condition. Confirm the dependent variable, unit, observation schedule, graph scale, condition label, phase boundary, and any missing or excluded data. Then describe each condition's level, direction of trend, variability, and stability.

The current BACB BCBA Test Content Outline includes graphing quantitative relations and interpreting graphed data in Domain C. Domain D covers single-case design features, prediction, verification, replication, threats to internal validity, design types, and critique of design data. It identifies entry-level examination content rather than prescribing one analysis worksheet.

A mean can summarize level while hiding an accelerating trend or a late shift. A range can summarize spread while hiding an alternating pattern. Keep the data path visible and write the within-condition description before calculating or comparing summaries.

Compare six visual dimensions

Use the expected direction of change and the design logic when answering each question.

The What Works Clearinghouse single-case technical documentation defines level as the phase mean, trend as the slope of a best-fitting straight line, and variability as fluctuation around the mean. Its evidence review also considers immediacy, overlap, and consistency across similar phases. Other visual-analysis traditions may operationalize or weight these dimensions differently, so state the chosen rule.

Immediacy and overlap need explicit windows

Immediacy compares data close to a condition change. The WWC procedure examines the last three points in one phase and the first three in the next when enough points exist. A clinician should name the window, expected direction, and reason for any alternative. Gradual skill acquisition may produce a different expected pattern from a rapidly acting environmental change.

Overlap also needs a declared method. One simple descriptive approach asks how many intervention points fall within the baseline range. Other indices define overlap differently and can yield different values from the same graph. Report the raw points and calculation instead of writing “little overlap” without a rule.

The authors of Systematic Protocols for the Visual Analysis of Single-Case Research Data developed structured protocols around level, trend, variability, immediacy, overlap, and consistency. Their procedures show how the expected direction changes the overlap comparison. The protocols support disciplined judgment; they do not turn one percentage into a functional-relation test.

Similar conditions test the prediction

Adjacent A-to-B comparison shows what changed at one boundary. A single change can coincide with history, maturation, measurement drift, a schedule change, or another event. Repeated conditions let the analyst compare observed data with earlier predictions.

In an ABAB design, similar A phases and similar B phases should show reasonably consistent patterns, while predicted changes occur as the independent variable changes. Multiple-baseline, multielement, and changing-criterion designs use different replication logic. A design should fit the behavior and question; withdrawing an effective or safety-critical support may be inappropriate.

WWC describes three demonstrations of an effect at three points in time as a design norm for its evidence standards. Its documentation also says a basic two-phase AB design has limited causal strength because it lacks replication. Clinical monitoring can still learn from an AB comparison, while its conclusion should stay proportional to the design.

A worked ABAB comparison

Suppose a fictional training graph shows independent help requests per session. The already-authorized support is B. The data are:

Level rises in B1, falls in A2, and rises again in B2. The last-three to first-three means are 1.67 to 4.00 at A1-to-B1, 4.67 to 1.67 at B1-to-A2, and 1.67 to 4.67 at A2-to-B2. The adjacent ranges have no overlap. A1 and A2 are similar; both B phases are higher, though B1 trends upward more clearly than B2.

This synthetic pattern contains three directionally consistent boundary changes. A defensible experimental conclusion would still require valid and reliable measurement, adequate condition implementation, appropriate phase length and timing, absence of disqualifying concurrent changes, and a design suitable for the question. The values illustrate analysis steps rather than evidence about a real intervention.

Use a repeatable review sequence

1. Verify the graph, condition definitions, measurement, missing data, and procedural-integrity record.
2. Describe level, trend, and variability within each condition.
3. Compare adjacent conditions for direction, magnitude, immediacy, and overlap.
4. Compare repeated or simultaneous conditions for consistency and predicted replication.
5. Mark contradictory, delayed, transient, or noneffect patterns.
6. Integrate the full design, contextual events, social validity, benefit, risk, and uncertainty.

A review of single-case design analysis likewise separates within-phase examination from between-phase comparison and discusses the role of replication. Statistical summaries or effect sizes can supplement the graph when suited to the question. They should retain the condition structure, time order, and raw data.

Common comparison errors

Common errors include comparing means before inspecting trend, calling the first post-change point an effect, using an unnamed overlap formula, ignoring variable baseline, pooling unlike conditions, and treating a condition label as proof of procedural integrity. Another error is claiming causation from a two-phase clinical graph whose change coincided with several other events.

Document who interpreted the graph, the data cutoff, rules used, disagreements, and the decision that followed. If decisions affect risk, treatment, or service continuation, obtain the review required by the case and setting.