Consistency across similar phases

Repetition adds a test of the pattern

In an A-B-A-B design, condition A occurs twice and condition B occurs twice. If both A phases show one pattern and both B phases show another, the repeated result is replication within the case. That replication can make a condition-linked explanation more credible than a single A-to-B change.

Consistency concerns the overall pattern, not identical values. Human behavior varies. A second intervention phase may begin at a different absolute level because learning carried over, the environment changed, or reversal was incomplete. The analyst asks whether the direction, magnitude, timing, and variability still fit the design’s prediction.

First confirm that phases are comparable

Two phases with the same label may differ in important ways. Before comparing them, verify:

• operational definitions and measurement units
• observation length and opportunity rules
• participants, implementers, setting, and schedule
• procedure version, dosage, prompts, and reinforcement
• fidelity and observer agreement
• medication, health, staffing, school, or family changes
• missing observations and eligibility rules

A repeated label cannot repair a material change in procedure. Rename or annotate the phase when the actual condition changed.

Read consistency alongside other dimensions

Visual analysis usually considers several dimensions together. Level describes the typical magnitude within a phase. Trend describes direction. Variability describes fluctuation. Overlap shows how much values from adjacent conditions share a range. Immediacy asks how quickly the pattern changes after the condition changes.

The repeated-phase question asks whether those features recur when the condition returns. A stable treatment effect across two B phases carries different meaning from two B phases with opposing trends. Neither observation should be reduced to one average.

A fictional A-B-A-B example

Maya is practicing a two-step classroom routine. During the first baseline phase, independent completion is 1, 2, 1, and 2 across four eligible opportunities. During a visual-prompt phase, values are 5, 6, 5, and 6. Baseline returns with 2, 1, 2, and 2. The prompt returns with 5, 5, 6, and 6.

The two baseline medians are 1.5 and 2. The two prompt-phase medians are 5.5 and 5.5. Each prompt phase begins above every baseline value, and variability stays narrow. This pattern shows consistency across the repeated conditions.

The example supports a condition-linked interpretation within those observations. It says nothing by itself about maintenance after prompts change, performance in another setting, social importance, or Maya’s experience of the routine. Those questions need separate evidence.

Differences can be clinically informative

Inconsistency deserves investigation. A weaker second treatment phase may reflect reduced fidelity, changed motivation, illness, different materials, a ceiling effect, or a procedure that works only under limited conditions. A stronger second phase may reflect practice, carryover, or a changed starting point.

Document the difference before labeling it failure or success. Review raw data, protocol versions, attendance, exposures, and contemporaneous events. If a design assumption no longer holds, report that limit clearly.

Compare exposure as well as session count. Four observations collected after equal practice opportunities differ from four observations collected after 2, 20, 5, and 18 opportunities. Session length, time since the prior session, and contact with the procedure can change what “similar phase” means. A compact phase table can list observations, total exposure, fidelity samples, and material events beside the graph. This gives reviewers enough context to judge whether an apparent replication rests on comparable experience.

When phases have unequal lengths, examine early and later portions rather than comparing only full-phase summaries. A late trend in a long phase may have no counterpart in a short phase.

Avoid mechanical agreement scores

There is no universal percentage that makes phases “consistent.” A rule such as “medians within 10%” can ignore trend, timing, range, and clinical context. Predefined decision rules may aid a specific program, but they should retain raw values and never replace visual analysis.

Useful audit measures can track repeated phases with complete measurement and fidelity records divided by repeated phases due for review. That is a documentation measure, not proof of experimental control.

Training sources support a broader analysis

The BACB BCBA Test Content Outline, sixth edition includes graphing, visual analysis, experimental design, and data-based decision making. It is examination content and does not prescribe a consistency formula.

A precision-teaching synthesis describes within- and between-condition analysis, charting, and repeated measurement. Its charting context reinforces a practical point: interpretation depends on comparable measurement and clearly identified conditions.