An ABA goal attainment scaling calculator can reproduce the traditional Kiresuk-Sherman transformation only after the underlying goal scales are complete and defensible. This working tool keeps every attained level visible, shows the weighting and expected-correlation assumptions, and calculates two prespecified views side by side. Locally written ordinal anchors remain local records; the transformation does not turn them into a norm-referenced test or decide whether a result is important, reliable, or caused by treatment.

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

A summary of fixed goals, not a verdict

Goal Attainment Scaling can retain several individualized outcomes in their own five-level records. The traditional transformation summarizes those attained levels on a familiar 50/10 scale. Its narrow question is arithmetic: under the declared weights and expected inter-goal correlation, what value does the formula produce for this fixed set of goals?

It does not answer whether the scales were written well, whether the ratings are reproducible, whether the goal set sampled everything that matters, or whether care produced the outcomes. Those questions require the underlying records, direct data, client interpretation, design evidence, and qualified review.

The original Kiresuk-Sherman paper proposed individualized scales and a standardized transformation. Later applications have varied considerably. This page therefore treats the output as a transparent model-based summary, not as an automatically standardized clinical score.

The traditional formula with its assumptions exposed

For goals indexed by i, the traditional transformation is:

T = 50 + 10 x sum(wi xi) / sqrt((1 - rho) x sum(wi^2) + rho x (sum(wi))^2)

where:

  • x_i is the attained level for goal i, restricted to one of -2, -1, 0, +1, or +2 from a frozen scale;
  • w_i is a positive weight assigned before outcomes are known; and
  • rho is the documented expected correlation among goal attainment scores, entered from 0 through 1 for this calculator.

The arithmetic also preserves sum(wi xi), sum(wi^2), sum(wi), the squared weight sum, and the denominator. That trail makes a weighting or transcription mistake easier to find. The familiar constants 50 and 10 are part of the transformation; they do not establish a local normative mean or standard deviation.

The systematic review of GAS measurement properties reproduces the formula and describes the expected-correlation assumption. It also reports limited evidence for several measurement properties in the reviewed contexts. Correct arithmetic therefore leaves validity and intended-use questions open.

Cases that should stay uncalculated

Leave the output blank when any required input is missing or when:

  • an attained level was assigned without a complete prewritten five-level scale;
  • the scale was edited after reviewers saw the outcome;
  • a value falls outside the discrete set from -2 through +2;
  • a weight is zero, negative, undocumented, or chosen after the attained level was known;
  • rho is missing, outside 0 through 1, or entered only because it produces a preferred result;
  • the goal set differs between the two views being compared;
  • one goal is duplicated so that the same outcome is counted twice;
  • an invalid observation condition was silently scored as nonattainment; or
  • the intended decision requires a validated instrument, inferential analysis, payer rule, or research protocol beyond this worksheet.

Equal weights may fit a stated purpose when that choice was made in advance. They still encode an assumption rather than a neutral default that erases differences in priority. When a publication or protocol supplies rho = 0.3, retain that source and explain why it belongs in this calculation; 0.3 is not a natural constant.

A calculator record that can be audited

Use this ABA goal attainment scaling calculator only after completing the provenance block and freezing the underlying scales:

FieldPrespecified recordRecord ID, goal-set version, and calculator versionClient or authorized representative involvedPurpose of aggregationReview date and eligible evidence windowConfirmation that every scale was frozen before outcome reviewWeighting rule and date selectedExpected-correlation source and rationaleSensitivity view planned before calculationMissing, invalid, or ambiguous rating ruleReviewers and approval date

Enter one row per distinct goal:

Goal record and versionPlain-language priorityAttained level x_iWeight view A w_iWeight view B w_iRating evidence locationGoal 1Goal 2Goal 3Add rows only when prespecified

Record the assumptions and outputs without hiding intermediate values:

CalculationView AView BExpected correlation rhosum(wi xi)sum(w_i^2)sum(w_i)(sum(w_i))^2(1 - rho) x sum(w_i^2)rho x (sum(w_i))^2DenominatorTraditional transformed scoreDifference between viewsInterpretation that remains supportableQuestions that remain unresolved

Walking through three fictional goal records

This example uses invented records and no real clinical data. The attained levels are +1, 0, and -1. The goal weights 3, 2, and 1 were declared before rating, and the main view uses a documented rho of 0.30.

Goalx_iw_iwi xiw_i^2Participation in a chosen community routine+1339Independent access to an agreed communication support0204Tolerable effort for a selected home routine-11-11Totals6214

The main-view denominator is:

sqrt((1 - 0.30) x 14 + 0.30 x 6^2)

= sqrt(0.70 x 14 + 0.30 x 36)

= sqrt(9.8 + 10.8)

= sqrt(20.6)

= 4.5387222872

The transformed value is:

T = 50 + (10 x 2 / 4.5387222872) = 54.4065264924

Report full precision in the audit trail and use a declared display rule, such as 54.41, only after the calculation. The individual +1, 0, and -1 levels stay beside the aggregate. A score above 50 does not prove that all goals were achieved, because the positive weighted result on one goal can offset another goal below its expected level.

Choose sensitivity views before seeing the score

Suppose the team prespecified a second view using the same goals, levels, and weights with rho = 0. The denominator becomes sqrt(14) = 3.7416573868, and the result is 55.3452248382. The difference from the main view is about 0.94 transformed-score points.

The spread shows dependence on the correlation assumption without identifying a true value. Trying several values after the fact, hiding most of them, and keeping the most convenient result would turn a sensitivity check into selective reporting.

Weighting can change the numerator as well as the denominator. If all three fictional goals receive weight 1 while their attained levels remain +1, 0, and -1, the denominator is 2.1908902300, sum(wi xi) = 0, and the transformed value is 50 under rho = 0.30. The equal-weight result does not invalidate the priority-weighted result. It shows that the summary depends on the declared weighting model.

The 2022 scoping review of GAS in randomized trials notes variation in goal setting, weighting, training, and reporting. Its discussion supports making those choices visible. It does not supply an ABA-specific default weight or expected correlation.

Why 50 and 10 do not make a norm

The transformation's constants resemble a T-score metric, but a norm-referenced interpretation requires separate evidence that has not been established here. The output is also not a confidence interval, standard error of measurement, reliable change index, minimum clinically important difference, social-validity score, or effect size.

The 2023 GAS educational review highlights scale formulation, baseline placement, level adequacy, content validity, context, training, and rating reliability. Those concerns sit upstream of the arithmetic. A correct denominator cannot repair a multidimensional or ambiguous scale.

Research using GAS in an autism psychosocial-intervention trial illustrates a structured protocol with predefined levels and later T-score conversion. That study is evidence about one research implementation, not proof that a locally built ABA calculator yields comparable scores across clients, programs, goals, or settings.

The current BACB Test Content Outline describes competencies in measurement, validity, reliability, representative data, client-informed goals, and data-based decisions. The BACB ethics-code page directs certificants to current ethics requirements. Neither source designates this formula as a required treatment or payer decision rule.

Keep nearby tools in their own lanes

The five-level design worksheet answers whether one GAS record is clear enough to score. This page answers how a fixed set of attained levels transforms under declared assumptions. The existing percent-of-goal-obtained tool uses phase means and a quantitative baseline-to-goal distance. The standard error of measurement tool uses reliability and score-scale variability. These outputs answer different questions and should not be converted into one another.

If the practical question is whether direct ABA data changed after a phase transition, retain the graph, design logic, replication, treatment-integrity evidence, contextual events, and the applicable single-case analysis. GAS aggregation does not supply experimental control.

Retain the inputs, not only the display number

Calculation copies should use coded record IDs. Keep goal text, rating evidence, access controls, and the retention schedule inside an approved system. The HHS Privacy Rule summary and Security Rule summary explain federal protections at a high level and state that summaries do not replace the rules.

Preserve the calculator version, formula, input rows, weights, rho, full-precision result, display rounding, reviewer, and calculation date. If a goal scale changes, create a new goal-set version. Do not overwrite the values used for an earlier aggregate.

The Standards for Educational and Psychological Testing provide broad guidance about intended uses, validity, reliability, fairness, and score interpretation. They are used here to bound claims, not to certify the GAS transformation or this implementation.

Related resources

Sources