An ABA rolling average calculates the mean of a fixed number of consecutive observations and moves that window through the series. Declare the window, alignment, missing-data rule, and purpose before use. Plot raw points beside the smoothed line. Smoothing can clarify a broad pattern, while it also delays and softens sudden changes that may require immediate clinical or safety review.

Decide whether smoothing answers the question

Use an ABA rolling average when the clinical team needs a secondary view of short-term variation across an ordered series. It can help show whether several recent observations are generally rising, falling, or stable when single-session movement makes the raw line hard to read. It is a display aid, not a replacement data stream. Sudden safety events, access failures, possible adverse effects, measurement problems, or abrupt losses of performance still require review from the raw record.

Write the purpose in the worksheet before selecting the window. “Describe the recent direction of session-level rate during the current phase” is specific enough to audit. “Make the graph smoother” invites a window chosen for appearance. State the person, measure, phase, dates, setting, observation unit, graph version, calculation owner, and planned review date.

Verify the source series

Confirm that all points use the same operational definition, unit, opportunity or exposure rule, phase, and compatible measurement conditions. Trace each plotted value to its source record. Inspect late entries, corrections, duplicate sessions, changes in observer or device, and shifts in observation duration. If a percentage is used, retain its numerator and denominator because 80% across five opportunities carries different information from 80% across fifty.

Keep missing, zero, and unavailable distinct. Zero means the measured event or response did not occur under a valid observation. Missing means the required value was not obtained or cannot be used. An unavailable point may reflect a session that never occurred. Converting any of these states into another changes the rolling calculation and can create a false trend.

Choose and document the window

A three-point window gives recent points substantial influence and begins producing values quickly. A five-point window is smoother and lags more. The choice should match the observation frequency, expected pace of meaningful change, and review purpose. Record the fixed width and avoid changing it after seeing which line looks more favorable.

For a simple unweighted window, add the values in the window and divide by the number of valid values required by the declared rule. A three-point trailing average at observation 5 uses observations 3, 4, and 5. The team should define whether every point must be present. Silently dividing by two when a three-point window contains a missing observation produces a different statistic.

Label alignment and phase boundaries

A trailing average is plotted at the last position in its window and uses only current and earlier observations. A centered average is placed at the middle position and uses observations on both sides, so it is unsuitable for a real-time display at the newest point. State the alignment in the legend and methods note.

Do not carry a window across a phase, definition, setting, or exposure change unless the review question explicitly calls for that mixed window and the graph labels it. Starting a new treatment phase with two baseline observations inside the first average can blur the transition. A phase-specific rolling line usually begins only after the new phase contains a complete window.

Irregular spacing also matters. A three-session average may cover three days in one span and three weeks in another. Plot the true dates or clearly label session order, and avoid describing the result as a time rate when the x-axis is merely observation number.

Work the fictional calculation

Ari has five sequential values: 10, 4, 7, 13, and 8. With a three-point trailing rule, the first window is (10 + 4 + 7) / 3 = 7.0. The second is (4 + 7 + 13) / 3 = 8.0. The third is (7 + 13 + 8) / 3 = 9.3 after rounding 9.333 to one decimal place.

The rolling series has three values because five raw observations contain three complete three-point windows. Its first value aligns with raw observation 3, leaving positions 1 and 2 blank on the smoothed line. The value 13 contributes to the last two windows but appears less prominent than it does in the raw series. The worksheet records the rounding rule and preserves full precision for recalculation.

If observation 4 were missing, the current complete-window rule would yield no averages at positions 4 or 5. The graph would show the gap. A different predeclared missing-data method could answer another question, but the team must label it and explain its effect.

Graph and report both series

Plot raw points with their original markers and connect them only under the graphing convention used by the clinical team. Add the rolling line with a different line style, a legend such as “3-session trailing mean,” and blank leading positions. Mark phase changes, relevant context changes, and missing observations without covering the raw data.

A concise report could say: “Across five observations in the current phase, raw values were 10, 4, 7, 13, and 8. Three-session trailing means were 7.0, 8.0, and 9.3. The smoothed series rose across its three windows, while the raw graph retained substantial session-to-session movement and a value of 13 at observation 4.” This wording describes the record without claiming a durable trend or treatment effect.

Interpret with clinical and causal limits

A rising rolling line does not show why values changed, whether the measure is valid, whether treatment was delivered as planned, or whether the outcome is meaningful to the person. Window overlap also means adjacent averages share observations, so three smoothed points are not three independent summaries. The window can delay recognition of a reversal and make a large value look modest.

Review level, trend, variability, immediacy, overlap, consistency, data density, exposure, treatment integrity, health and setting context, and the person’s feedback through the qualified process. Single-case design guidance and visual-analysis protocols provide broader context; neither creates a universal rolling-average decision rule.

Keep access and authorship visible

Ask whether the selected outcome and display represent something useful to Ari and whether the graph can be reviewed through an accessible format. Keep AAC and other communication tools continuously available as described in ASHA’s AAC guidance. Smoothing must never delay access to food, water, bathroom use, mobility, prescribed health care, rest, relationships, or emergency help.

The BACB ethics hub and CASP public summary offer professional context. The BCBA Test Content Outline describes examination content, not a case protocol. NIST moving-average guidance explains smoothing concepts but does not supply a clinical cutoff.

Clinician review checklist and limits

Before using the rolling line in a decision, confirm:

  • the source values, operational definition, unit, exposure, phase, and dates are compatible;
  • zero, missing, canceled, and unavailable observations retain distinct states;
  • the window width, weighting, alignment, phase rule, and rounding were declared;
  • raw points, missing positions, context annotations, and the smoothed line remain visible;
  • the written interpretation describes level and variability as well as smoothed direction;
  • immediate safety, health, access, integrity, and client-feedback signals were reviewed from raw evidence; and
  • the worksheet records the decision, owner, next evidence, and review date.

A rolling average is descriptive and sensitive to the chosen width, alignment, endpoints, gaps, and data density. Five observations support only a small display, and the three overlapping values in Ari’s example should not be treated as a large new sample. Reopen the worksheet when the definition, unit, phase, exposure, software, or missing-data rule changes. Preserve the prior graph and calculation so later reviewers can reproduce what the team saw at the time.

Related resources

Sources