Read a measure
Begin with the interpretation you want to make. A familiar instrument name does not answer whether its scores support that interpretation.
Name the construct
A construct is the attribute of interest; an operation is the observation used to study it. Perceived stress, stressful events and a physiological stress response are different targets. The PSS-10 profile shows why a questionnaire cannot stand in for all three.
A question to ask
What would count as a plausible alternative explanation for a high score? Write that down before seeing the results.
Identify the version
Record the exact form, language, age range, informant, recall window and scoring rule. A short version is a different selection of observations. A new edition can change both tasks and reference data, as the WAIS-5 profile illustrates.
A survey, a performance task and an informant report provide different perspectives. Agreement is useful evidence, but disagreement can reveal setting-specific behavior rather than a defective respondent.
Ask which reliability
- Internal consistency
- How responses to items relate within a particular administration. It does not by itself establish stability or a single underlying cause.
- Test–retest reliability
- How consistently scores distinguish people across occasions. The interval and expected true change matter.
- Interrater agreement
- Whether observers assign similar codes or ratings to the same material. It does not show that the code predicts an outside outcome.
A task may reliably slow people in a conflict condition while doing poorly at ranking individuals by that slowing. Hedge, Powell and Sumner’s reliability study investigates this distinction; compare the Stroop and flanker profiles. A coefficient belongs to a sample, score and protocol.
Inspect the interpretation
Ask whether the observations support the proposed meaning and use. Item coverage, relationships with other measures, prediction of a relevant criterion and alternative explanations answer different parts of that question. Correlation with another self-report measure is useful but can share method bias.
For a screener, identify the independent reference assessment and the errors a threshold makes. PHQ-9 and GAD-7 illustrate study-specific accuracy. The synthetic screening example shows why sensitivity alone does not tell you the meaning of a positive screen.
Check the population
Who supplied the evidence? A student sample, a national survey and a referred clinical group are different populations. A percentile describes position in a reference distribution; it is not percent correct. Reduced verbal demands, as in a matrix task, do not demonstrate cultural equivalence.
Translation requires more than changing words. Ask whether the construct, response process and score relationships support the intended comparison. A form listed on a website has not necessarily been validated in every setting.
Interpret change
Two different scores can reflect true change, practice, different conditions, changed instructions and measurement error. A treatment interpretation also needs a design that addresses alternative explanations. Keep the score observation separate from the causal conclusion.
A question to ask
Were the same version, reference period and administration conditions used? What evidence establishes that a change of this size is meaningful for this purpose?
Check access and use
A public webpage may describe a copyrighted test. Research use, commercial use, translation, online administration and scoring software can have different terms. Training and qualification requirements are separate from the cost of a form. Each profile links to its source and reports the permission scope actually inspected.
This guide explains measurement decisions. It does not supply protected assessments, professional certification or personal diagnostic results. Read the source notes before treating a profile as an evidence review.