An ABA phase pattern comparison calculator can make a dense graph easier to discuss without pretending that a handful of summaries can read the graph for you. The worksheet below keeps ordered observations visible, calculates level, linear trend, a declared variability summary and a boundary-window difference, and then puts measurement quality, context, client priorities and the complete single-case design back at the center.

Clinicians & ABA Professionals / Data, Outcomes and Clinical Decision-Making.

A calculator for description, not a verdict

Two reviewers can look at the same graph and emphasize different features. One may notice a shift in typical values, while another is concerned about a rising baseline or a wide spread in the comparison phase. A transparent calculation sheet helps them locate the disagreement. It does not settle whether an intervention produced an effect.

This tool reports eight items for each phase: the number of valid observations, mean, median, minimum, maximum, range, least-squares slope and mean absolute deviation from the phase median. It also reports the difference between phase medians, the difference between phase means and, when each phase has at least three valid observations next to the boundary, the difference between the last three values in the first phase and the first three values in the second.

Those outputs are descriptive. They do not establish that a phase is stable, that a change is immediate, that the comparison is representative or that the underlying design demonstrates experimental control. A qualified reviewer must inspect the chronological graph and the surrounding evidence.

The record that belongs beside the numbers

Begin with a single outcome measured in compatible units under clearly labeled conditions. Preserve the observation order. If one point is a percentage based on five opportunities and another is based on twenty, keep those denominators beside the values rather than treating the percentages as interchangeable by default.

The calculation record should name:

  • the client-selected outcome and why it matters;
  • the operational definition, unit and exposure or opportunity base;
  • phase labels, phase boundaries and the intended comparison;
  • the ordered raw value for every observation;
  • ordinary communication, mobility, sensory and other supports in place;
  • the observer, implementer, setting and relevant treatment-integrity information;
  • missing, invalid, late, corrected and disputed states;
  • every intervention, measurement, staffing, health or contextual change relevant to interpretation;
  • the prespecified preferred direction, if a directional summary will be discussed;
  • the calculation version, reviewer, date and unresolved questions.

Do not recode a missing observation as zero. Do not compress two measures with different units into one phase. If observation intervals are uneven, the slope calculated by sequence number is not a slope per minute, day or session. Retain the actual time variable and use an appropriate analysis when timing is part of the question.

Copyable ordered-input sheet

Use one row per scheduled observation. Add rows instead of replacing an original entry.

OrderPhaseDate or sequenceRaw valueUnit and denominatorStatusContext or change note1Avalid / missing / invalid / late / disputed2A3A4A5A6B7B8B9B10B

The team's declared eligibility rule determines which rows enter the displayed calculation. A value should enter the valid series only after that documented rule is applied. Keep excluded values in the audit record with the reason, author and date.

Formula panel

Let a phase contain valid values y1 through yn in chronological order, with sequence positions x = 1, 2, ..., n.

OutputTransparent ruleInterpretation limitValid nCount of observations eligible under the declared ruleA larger n does not prove representative samplingMeanSum of valid values divided by nSensitive to extreme values and incompatible denominatorsMedianMiddle ordered value, or mean of the two middle valuesDoes not show chronology or spreadMinimum and maximumLowest and highest valid valuesEach may depend on one observationRangeMaximum minus minimumIgnores how other values are distributedLeast-squares slopeSum of (x - mean x)(y - mean y) divided by sum of (x - mean x)^2Summarizes one linear direction; it can hide curvature and is per sequence position unless actual time is modeledMean absolute deviation from phase medianSum of absolute value(y - phase median) divided by nA declared spread summary, not a universal stability testMedian differencePhase B median minus Phase A medianSign depends on order and does not establish importance or causeMean differencePhase B mean minus Phase A meanSensitive to outliers and phase compositionBoundary-window differenceMean of first three valid B values adjacent to the boundary minus mean of last three valid A values adjacent to the boundaryA compact edge summary, not a universal definition or proof of immediacy

The tool should display not calculated when a formula lacks enough valid inputs. An undefined result is not zero. If a scheduled observation is missing near the boundary, show the scheduled and valid sequences explicitly. Do not call nonadjacent valid values a three-observation boundary window unless the declared method permits that choice; otherwise, leave the result blank.

Blank phase-summary output

SummaryPhase APhase BB minus A or noteValid observationsMeanMedianMinimumMaximumRangeLeast-squares slopeMean absolute deviation from phase medianThree-point boundary-window mean

Under the table, retain the graph, all eligible and ineligible points, calculation precision, the observation-time basis and a narrative review of level, trend, variability, immediacy, overlap and consistency across similar phases.

Mina's fictional phase comparison

Mina is fictional. Her team is reviewing a client-selected communication outcome: independent communication responses during matched 20-minute observations with her usual communication system available. Higher values are preferred for this example. Phase A contains 2, 3, 2, 4 and 3. Phase B contains 5, 6, 7, 6 and 8. There are no missing values in the example.

SummaryPhase APhase BB minus AValid observations550Mean2.86.43.6Median363Minimum253Maximum484Range231Least-squares slope per observation sequence0.30.60.3Mean absolute deviation from phase median0.60.80.2Three-point boundary-window mean3.06.03.0

The Phase A mean is 14 / 5 = 2.8; the Phase B mean is 32 / 5 = 6.4. Their medians are 3 and 6. The ranges are 4 - 2 = 2 and 8 - 5 = 3.

For Phase A, the absolute distances from median 3 are 1, 0, 1, 1 and 0. Their mean is 3 / 5 = 0.6. For Phase B, the distances from median 6 are 1, 0, 1, 0 and 2, so the result is 4 / 5 = 0.8.

With sequence positions 1 through 5, the least-squares slope is 0.3 for Phase A and 0.6 for Phase B. These are responses per observation sequence, not responses per day, because the example has not supplied elapsed time. The final three Phase A values, 2, 4 and 3, average 3.0. The first three Phase B values, 5, 6 and 7, average 6.0. The declared boundary-window difference is 3.0 responses.

The numbers describe a higher typical value and a positive boundary difference in this fictional contrast. The values alone cannot explain why the change occurred. The rising summary slope in each phase, the greater Phase B range, the conditions around each observation and the rest of the design still matter. One A-to-B comparison does not demonstrate replication or a functional relation.

What disagreement among summaries can reveal

Mean and median can separate when one value is unusually high or low. A shallow slope can coexist with a large phase-level difference. A boundary window can look favorable while later points return toward the first phase. A narrow range can occur around an unwanted level. These are not calculation failures. They are reasons to inspect the graph and ask a sharper question.

When reviewers disagree, record the feature each person weighted, the observations that support the description and the additional evidence that could resolve the disagreement. Do not average incompatible judgments into a confidence score. Do not let the calculator label a phase as stable or unstable without a separately defined, justified review rule.

Corrections, missingness and sensitivity

Every corrected value can change several outputs. Preserve the original value, corrected value, reason, author, timestamp and approval path. Recalculate from the amended series and version the result rather than overwriting the earlier analysis.

If a point is missing because a difficult session ended early, missingness may carry more clinical meaning than a random blank. Show scheduled observations, valid observations and reasons by condition. The ABA phase pattern comparison calculator should never infer what happened during unobserved time.

A sensitivity display can show how a disputed point changes a mean, median, slope or spread summary. It should present both declared states and the validity question. It must not become permission to discard a value because one version tells a preferred story.

Visual analysis and causal limits

The What Works Clearinghouse Version 5.0 handbook describes single-case visual review using level, trend, variability, immediacy, overlap and consistency, within a design capable of repeated demonstrations. This worksheet calculates selected descriptions for one adjacent contrast. It neither conducts the complete WWC review nor applies WWC research ratings to clinical care.

Wolfe, Barton and Meadan developed systematic visual-analysis protocols and discussed the possibility of disagreement among analysts. Their weighted protocol is not reproduced here. The calculator returns auditable descriptive values and leaves synthesis to qualified reviewers.

The SCRIBE 2016 Statement supports clear, complete and transparent reporting of single-case work. A simulation study by Manolov and colleagues found that data features affect quantitative techniques and emphasized visual inspection as an initial stage. Neither source turns a summary statistic into a clinical decision rule.

Client relevance and professional authority

The graph can move in a preferred numeric direction while the outcome is unimportant, the procedure is unacceptable or the conditions are not sustainable. Review the measure, visible change, burdens, adverse effects and tradeoffs with the person through accessible communication. Preserve assent or dissent when applicable and the person's requested changes.

Keep ordinary communication and access supports available during observation and review. The BACB ethics materials provide the current professional source for certificants. They do not certify this tool. Case-specific assessment, interpretation and treatment decisions remain with qualified professionals working within competence, licensure, supervision, payer and setting requirements.

Privacy, security and an accountable decision record

Use the least identifying data that still support the legitimate purpose. Store the worksheet only in approved systems, limit access, preserve correction history and follow the organization's retention and incident processes. The HHS Privacy Rule summary and HHS Security Rule summary describe federal requirements for regulated entities and protected information. They are not complete compliance guides, and this page cannot determine whether HIPAA applies to a particular workflow.

Close the review by recording the question, raw graph version, calculations, disagreements, client input, interpretation, qualified decision owner, unresolved evidence and next review date. Reopen the record when a source value, phase boundary, measure, support, condition or goal changes.

Related resources

Sources