Method

The order follows two rules: subtests from the same domain never sit next to each other, so that fatigue does not fall disproportionately on one domain; and the evaluation opens with figural material, because a verbal task at the start unsettles people who are unsure of their language more than it does anyone else.

1. 01

   Ability estimation per subtest

   The six untimed subtests are scored through a three-parameter logistic item response model. Every item has parameters for difficulty, discrimination and guessing; the guessing parameter equals the reciprocal of the number of options. What is estimated is the expected value of the posterior distribution (EAP) over a quadrature grid of 81 nodes with a standard normal prior. This has two advantages over maximum-likelihood estimation: the estimate stays finite when every answer is right or every answer is wrong, and the spread of the posterior yields the standard error of measurement directly, instead of having to derive it from an assumed reliability.

2. 02

   Raw-score transformation for spans and timed subtests

   For the two memory spans and the two timed subtests, difficulty is not a property of the individual item but of the sequence length or the time allowed. These four subtests are therefore brought onto the same metric through a declared linear transformation of the raw score. In Symbol Search, wrong answers are subtracted from right ones, because on a yes/no task fast guessing would otherwise be rewarded.

3. 03

   Scaled scores

   Each ability estimate is converted into a scaled score with a mean of 10 and a standard deviation of 3, bounded to the range 1 to 19 — the metric the Wechsler instruments use.

4. 04

   Index and overall scores

   An index is the sum of the scaled scores of its two subtests, standardised against the modelled distribution of that sum. The spread of that distribution follows from a declared correlation matrix; the reliability of the sum follows from Mosier's formula for composite scores. Index and overall scores have a mean of 100 and a standard deviation of 15. Every score is reported with its 95 per cent confidence interval, computed from the standard error of measurement.

• Rapid guessing

  Answers given in under two seconds on untimed items. An item cannot be taken in within that time; such answers are guesses, and they lower the result.

• Interruptions

  Switching to another window during the session, with the duration recorded. On the timed subtests this distorts the result directly.

• Omissions

  Items left unanswered. They are scored as wrong.

• Response times

  The median across all untimed items, as an indication of how carefully the test was worked through.

None of these findings makes a result invalid by itself. They determine how confidently it may be read, and they are reported in a section of their own rather than disappearing into the small print.

• Not clinical assessment. Giftedness, intellectual disability, learning disorders, ADHD and dementia processes can neither be established nor ruled out with this.
• No standardisation sample. See section 04. The absolute level of the scores is to be read with reservation.
• No age or education norms. Clinical instruments norm by age group, because fluid abilities decline slowly from roughly the third decade of life while crystallised abilities keep rising into old age. This instrument uses a single reference norm and therefore understates older adults and tends to overstate younger ones.
• Dependence on language. Vocabulary and verbal analogies rely on vocabulary in the test language and systematically disadvantage people who did not grow up speaking it.
• Unsupervised administration. Whether aids were used, whether someone helped, or whether the person was concentrating cannot be established with certainty — only partly, through the soundness checks.
• Practice effect. A repeat sitting usually comes out better, because the item types are familiar. That is not a gain in ability.
• No measure of personality, creativity or practical judgement. What is measured is a selected part of cognitive ability, not a person's capability as a whole.