An ABA reliable change index calculator can compare the difference between two comparable scores with a classical estimate of measurement error. Its usefulness depends on the score pairing, standard deviation, reliability coefficient, and direction rule being defensible before the result is visible. This worksheet preserves the raw sign, shows every intermediate value, and places two prequalified source sets side by side only when an assumption-sensitivity review has a documented reason.

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

The calculation has a deliberately narrow job

For two scores on the same interpretable scale, this worksheet uses the classical Jacobson-Truax sequence:

SEM = SD x sqrt(1 - r)

SEdiff = sqrt(2 x SEM^2)

RCI = (score2 - score1) / SEdiff

The raw change is always score2 - score1. If higher scores were declared as the preferred direction, the oriented change and oriented RCI retain the raw sign. If lower scores were declared as preferred before the data were inspected, multiply the raw change and raw RCI by -1 in separate fields. Never replace or hide the raw sign.

This calculator implements one classical equal-error form, so it treats the same SEM construction as applicable to both occasions. The calculation does not adjust for practice effects, regression to the mean, unequal occasion-specific errors, conditional SEM, item-response-theory information, serial dependence, multiple testing, or other reliable-change variants. It reports no diagnosis, recovery classification, clinical significance, functional relation, treatment effect, or causal attribution.

The original Jacobson and Truax paper distinguishes statistically reliable change from a broader clinical-significance framework. The formula reproduced here is an arithmetic aid, not a claim that the complete framework applies to an ABA case or that a distributional boundary establishes client-valued importance.

Comparable scores come first

Record the assessment name and edition, exact score, scale, administration conditions, language, mode, accommodations, and scoring rules at both occasions. The score must retain the same interpretation. A change in edition, form, norm group, item set, scoring transformation, informant, administration method, or accommodation may make subtraction misleading even when both values are numeric.

Document elapsed time, the planned retest interval, developmental or contextual changes, and any plausible practice or memory effect. The calculator leaves those features uncorrected and visible. A qualified reviewer may decide that another method is needed or that no reliable-change calculation is interpretable.

The standard deviation and reliability coefficient must match the score and intended replication conditions. A coefficient from a different subscale, population, interval, or method is not interchangeable. The Standards for Educational and Psychological Testing call for reliability and precision evidence that supports the proposed score interpretation and use. Arithmetic completeness does not establish that match.

A comparability record reviewers can challenge

Complete this record before reviewing the magnitude or direction of change.

FieldLocked entryResolved?Evidence or reviewer noteAssessment and editionyes / noExact reported score and scaleOccasion 1 date and scoreOccasion 2 date and scoreSame score meaning at both occasionsAdministration mode, language, and accommodationInformant or rater consistencyElapsed interval and planned retest intervalPractice, memory, maturation, or context concernsHigher or lower scores declared preferredAny comparison boundary and rationaleMultiple scores or repeated looks plannedCalculation version, analyst, and reviewer

For each source set, preserve the score-scale SD, reliability coefficient, reliability type, sample, subgroup, interval, conditions, source page or table, publication date, and check date.

Source fieldSet ASet BScore-scale SDReliability, rReliability methodReference sample and subgroupInterval and replication conditionsSource page or tableConditional or score-specific error available?Why this set is independently defensible

A second row needs a reason that exists before the output. A sensitivity comparison is appropriate only when both sets are source-qualified before the results are inspected and the unresolved assumption matters to the review. Producing a favorable RCI is never a valid reason to add or retain a source.

Reproducible calculation ledger

An ABA reliable change index calculator should leave the following inputs and intermediate results open to review.

Input or outputSource set ASource set BOccasion 1 scoreOccasion 2 scoreRaw change, score2 - score1Direction multiplier, 1 or -1Oriented changeSDReliability, rSEM = SD x sqrt(1 - r)SEdiff = sqrt(2 x SEM^2)Raw RCIOriented RCISource-qualification statusInterpretation status

Use full precision until the display step. Save the formula version and independent spot check. If SEdiff is zero, the RCI is undefined; do not divide by zero, substitute a small denominator, or report infinity as evidence of change.

Devon's fictional higher-better example

Devon is a fictional BCBA reviewing two synthetic scores. Occasion 1 is 52, occasion 2 is 61, and higher scores were declared as the preferred direction before calculation. Raw change and oriented change are both 9.

Source set A provides SD = 10 and r = 0.84:

SEM_A = 10 x sqrt(0.16) = 4

SEdiff_A = sqrt(2 x 4^2) = 5.6568542495

raw RCI_A = 9 / 5.6568542495 = 1.5909902577

The oriented RCI is also 1.5909902577 because higher is the declared preferred direction.

Source set B provides SD = 8 and r = 0.90:

SEM_B = 8 x sqrt(0.10) = 2.5298221281

SEdiff_B = sqrt(2 x 2.5298221281^2) = 3.5777087640

raw RCI_B = 9 / 3.5777087640 = 2.5155764747

ResultSource set ASource set BRaw change99Oriented change99SEM42.5298221281SEdiff5.65685424953.5777087640Raw RCI1.59099025772.5155764747Oriented RCI1.59099025772.5155764747

For this demonstration, the team had predeclared 1.96 as a comparison boundary. Set A falls below it, while set B exceeds it. That conflict is the point of the sensitivity table. It must not be resolved by choosing set B because it produces the desired label. Preserve both results, investigate which source supports the intended interpretation, and describe the remaining uncertainty. A boundary does not turn either result into clinical significance or treatment attribution.

Lin's fictional lower-better direction check

Lin is a fictional reviewer checking whether the implementation preserves direction. The synthetic scores are 20 and 14, lower scores were declared preferred, SD = 5, and r = 0.75.

SEM = 5 x sqrt(0.25) = 2.5

SEdiff = sqrt(2 x 2.5^2) = 3.5355339059

The raw change is 14 - 20 = -6, so the raw RCI is -1.6970562748. Applying the separately recorded lower-better multiplier gives an oriented change of 6 and an oriented RCI of 1.6970562748. Both signs stay in the record. This check prevents a dashboard from silently presenting every preferred-direction result as though the underlying score increased.

Invalid, undefined, and redirected states

Reject an SD that is zero, negative, missing, or nonnumeric. Reject reliability below 0 or above 1. A reliability of 1 produces SEM = 0 and SEdiff = 0, so RCI is undefined. That boundary is not proof that a real assessment has no measurement error. A reliability of 0 makes SEM equal the supplied SD, but the source still requires qualification.

Do not calculate when the score meaning changed across occasions, the observed scores use different scales, or the source values do not fit the score. Redirect when the manual provides conditional SEM, occasion-specific error estimates, practice-effect adjustments, or a method designed for the assessment. Also stop when repeated testing, multiple outcomes, selective reporting, floor or ceiling effects, changed informants, or missing administrations make the simple result misleading.

The current methods paper on reliable-change calculations shows that different standard-error constructions and score-specific information can change reliable-change results. The Journal of Statistical Software article and implementation documents multiple methods and keeps reliable change distinct from a complete clinical-significance judgment. These sources support transparency about the selected form; they do not authorize mixing formulas after results are known.

The RCI returns to the complete clinical record

An RCI describes the score change relative to an assumed classical error distribution. The number does not reveal why the score changed. It cannot separate intervention effects from maturation, history, measurement drift, practice, expectancy, rater changes, regression, concurrent services, medication, or ordinary variation. No functional relation follows from this calculation.

Review the complete direct-data graph, assessment report, administration notes, treatment-integrity data, setting events, adverse effects, generalization, maintenance, feasibility, and client-valued outcomes. Ask whether the measured construct and declared direction matter to the person. Preserve accessible client and caregiver input, assent and consent processes, dissent, and the burdens as well as benefits of any next step.

The BACB Ethics Codes page provides the current official professional source, and the BCBA Test Content Outline includes validity, reliability, representative measurement, and assessment interpretation. Neither source certifies this calculator, supplies competence, or converts an RCI into an ethical, clinical, payer, diagnostic, or legal conclusion. Use qualified psychometric and clinical review within scope, supervision, licensure, and organizational requirements.

Protect the data and preserve the analysis trail

Use fictional or appropriately de-identified values for demonstrations. When identifiable information is necessary in an authorized clinical workflow, use the least identifying data needed, approved systems, role-based access, and a documented retention and versioning process. The HHS Privacy Rule summary and Security Rule summary describe federal obligations for regulated entities. They do not decide applicability, certify this worksheet, or replace legal review and risk analysis.

Close the ledger with both observed scores, comparability record, source sets, exact formulas, raw and oriented values, any boundary and its provenance, full-precision results, display rounding, alternative-method review, client and caregiver input, disagreements, analyst, qualified reviewers, decision owner, and next review date. Append changes as new versions rather than silently replacing prior evidence.

Related resources

Sources