Explaining an assessment score and its uncertainty to a client

START HERE

The essential idea

Before explaining a number, identify the measure, version, construct, time frame, score direction, and comparison group. A symptom score is not an intelligence score, a percentile is not a percentage of questions answered correctly, and a high score is not always worse. Some scales measure positive functioning, others distress. The same number on two scales can therefore have very different meanings. Tell the client what the measure asked about and how the result relates to the purpose of this assessment, rather than beginning with a technical label that sounds authoritative but answers no practical question.

BUILD THE REASONING · 01

Begin with what the score can answer

Before explaining a number, identify the measure, version, construct, time frame, score direction, and comparison group. A symptom score is not an intelligence score, a percentile is not a percentage of questions answered correctly, and a high score is not always worse. Some scales measure positive functioning, others distress. The same number on two scales can therefore have very different meanings. Tell the client what the measure asked about and how the result relates to the purpose of this assessment, rather than beginning with a technical label that sounds authoritative but answers no practical question.

An observed score is produced under particular administration conditions. Fatigue, misunderstanding, missing responses, assistive access, a different language version, or a changed setting may affect interpretation. Review these before calculating a precise interval around a questionable score. Reliability concerns consistency or precision under specified conditions; validity evidence concerns support for the proposed interpretation and use. A precise measure of the wrong construct still does not answer the clinical question. A population norm can describe relative standing but does not establish diagnosis, need for care, or the person's goals by itself.

Ask what decision the client thinks the score will make. One person may fear it means permanent illness; another may believe a small improvement proves treatment is complete. State the intended use: adding structured information to an interview, tracking change, or identifying a need for further assessment. Explain limits in terms of that use. 'We will compare this with your sleep, work, and account of the last two weeks' is more useful than a generic warning that tests are imperfect. The remainder of this lesson uses invented numbers to make the uncertainty visible; they are not cutoffs for a real instrument.

BUILD THE REASONING · 02

Compute an interval and state the assumptions

For an illustrative classical-test-theory calculation, suppose a fictional distress scale has a relevant reference-group standard deviation of 10 points and an appropriate reliability estimate of .84. The standard error of measurement is SEM = SD × square root of (1 − reliability), so 10 × square root of .16 = 4 points. A score of 60 has an approximate 95% measurement interval of 60 ± 1.96 × 4, or about 52.2 to 67.8, under the model's assumptions. This is a teaching approximation; use the actual manual's score-specific precision and interval method when available.

The interval expresses uncertainty associated with the measurement process, not the full uncertainty about the person's diagnosis or future. In frequentist terms, the interval procedure has the stated coverage over repeated comparable measurements under its assumptions; avoid presenting the fixed unknown true score as a random event with a newly calculated probability. For client feedback, a simpler defensible explanation is that the observed value should be treated as an estimate with a meaningful margin, not an exact location. Explain the direction and size of the margin without claiming the person will necessarily score anywhere in that range next week.

Suppose an organization uses an illustrative threshold of 65 for a follow-up interview. The interval spans that threshold, so a categorical description such as 'definitely below concern' would overstate precision. This does not tell you to ignore the threshold or automatically reclassify the score. Follow the measure's validated decision rules and integrate other evidence, especially direct safety or functional concerns. The choice to gather more information depends on the consequence of an error as well as the score. A measurement interval cannot compensate for inappropriate norms, invalid administration, or an unsupported threshold.

BUILD THE REASONING · 03

A lower score is not automatically reliable recovery

Tracking change adds a second measurement and therefore a second source of error. In a simplified example with the same 4-point SEM at both occasions and independent errors, the standard error of the difference is square root of (4 squared + 4 squared), about 5.66. If the score falls from 60 to 52, the change is −8 points and the reliable change index is −8 / 5.66, about −1.41. Its magnitude does not exceed the illustrative 1.96 threshold. The observed decrease may still matter, but this calculation does not establish change beyond the expected measurement error at that criterion.

If the score instead falls from 60 to 45, the difference is −15 and the index is about −2.65, exceeding the illustrative threshold. That supports a reliable decrease under these assumptions. It does not prove recovery, identify the cause, or tell whether the person's priorities have improved. A reliable change criterion and a clinically meaningful functional criterion answer different questions. Someone may improve reliably while still experiencing substantial difficulty; someone may achieve an important personal goal with a numerical change that is not statistically reliable on this measure.

Use a method appropriate to the instrument and design rather than applying the example mechanically. Precision can vary across score levels; errors may not be independent; the comparison population and reliability estimate may not fit; repeated testing and regression toward the mean can complicate interpretation. Changes in language, format, or support can also change comparability. The Jacobson–Truax paper gives an important framework, not a universal stamp of treatment success. Combine the score trajectory with the client's account, function, adverse effects, and relevant contextual changes, and report what the available evidence cannot separate.

BUILD THE REASONING · 04

Explain the result and check the understanding

A useful feedback sequence has four parts: describe what was measured; give the result with its uncertainty; connect it to other information; agree the implication and next step. For the first example: 'This questionnaire summarizes the distress you reported for the stated period. Your score was 60. It is an estimate with a margin, not an exact measurement of you. It fits your report that sleep and work have been difficult. We will use it with the interview to decide what support and follow-up make sense.' Invite the client to say what fits and what seems missing.

When discussing change, separate encouragement from statistical overstatement. 'You reported less distress and returned to two valued activities. The score decreased, although the change is not large enough in this example to clearly separate it from measurement variation. Both the improvement you describe and the remaining uncertainty matter.' This is more respectful than dismissing lived improvement because a threshold was not crossed. It is also more accurate than calling a small decrease proof of recovery. Ask whether the measure captured the changes the client cares about and what other outcomes should be followed.

Use teach-back about the decision, not a mathematics exam. Ask the client how they understand what the result does and does not tell them, and what the next step will be. Correct a misconception gently: a percentile is relative standing, a threshold is a decision aid, and a score does not determine personal worth. Document the measure/version, relevant administration conditions, result, interpretation limits, other evidence, and agreed plan. If a third party receives the report, avoid categorical wording that becomes more certain than the actual analysis when separated from the conversation.

FROM CONCEPT TO DECISION

An eight-point decrease that matters but does not settle the question

1. 1

   Check what is comparable

   Confirm the same version, time frame, understood language, and comparable administration. If these differ materially, the calculation needs reconsideration before interpretation.

2. 2

   Calculate precision and change

   SEM = 4; SE difference ≈ 5.66; reliable change index = −8/5.66 ≈ −1.41. The decrease does not exceed the example's 1.96 criterion.

3. 3

   Integrate function

   Returning to class is a meaningful improvement; missed work is an ongoing concern. Neither the non-significant index nor the class attendance should erase the other evidence.

4. 4

   Explain and plan

   Describe the observed improvement, measurement uncertainty, and remaining functional goal. Agree continued assessment/support and the next review rather than assigning 'cured' from one number.

   The score and your return to class suggest progress, but the score alone cannot establish recovery. Let's look at what still makes work difficult and what you want to improve next.

MAKE THE REASONING YOURS

Work through a new situation

READ WITH A PURPOSE

Best references

These readings deepen the specific models, evidence, and decisions in this lesson.

1. 01
   AERA, APA, and NCME — Standards for Educational and Psychological Testing (2014), open access

   Read the validity, reliability/precision, fairness, and score-interpretation chapters. Evidence supports a proposed interpretation and use, not a universally valid test label.

2. 02
   HealthMeasures — Interpret scores

   Inspect reference populations, score direction, standard errors, and score-interval examples; use the actual measure's scoring guidance rather than transplanting this lesson's fictional numbers.

3. 03
   Jacobson & Truax (1991) — Clinical significance and reliable change

   Read the original distinction between change beyond measurement error and movement into a meaningful functional range; the statistical result does not prove treatment caused change.