How this is scored

The structure

Five facets, each measured by eight self-report items, plus a separate scored situational section of 12 items and 2 attention checks that are never counted toward a facet. That is 54 screens and about twelve minutes.

The four-quadrant arrangement, mapped in plain terms here, follows the domains popularised by Daniel Goleman and the ability model of Salovey and Mayer. Those are frameworks, which are ideas rather than property. The facet names and every item are ours.

Item writing and keying

Stems are frequency-anchored ("how often is this true of you") rather than agreement-anchored. Frequency anchors tend to be less self-flattering than asking whether someone agrees they are a good listener.

Five of the eight items in each facet are keyed positively and three are reversed, using semantic opposites rather than negations. Balanced keying cancels the tendency to agree with whatever is put in front of you, but it reliably creates a wording artefact of its own, which is why the split is five to three rather than an even one.

Items are adapted from the public-domain International Personality Item Pool (Goldberg, 1999; Barchard, 2001), which places its items in the public domain for any purpose. There is no emotion-regulation scale in that pool, so every Composure and Repair item is written from scratch. This is not the EQ-i 2.0, the MSCEIT, the TEIQue or any other published instrument, and we are not affiliated with their publishers.

The situational section

Two kinds of item. The understanding items are keyed from appraisal theory: each situation is built from a configuration that entails one emotion, so the answer follows from the situation rather than from a vote. The management items are scored for effectiveness with partial credit, so choosing the second-best response is reported as exactly that rather than as wrong.

Options are shuffled per person, so the effective answer is not in a fixed position.

The honest limitation. Our effectiveness ratings are currently authored rather than set by an independent expert panel. A documented panel is planned and this page will say so, with the panel size and the agreement between raters, once it exists. Until then, treat the situational band as the softer of the two results you are given.

Scoring, and why there are two tracks

Published research estimates that self-report emotional intelligence and performance-based emotional intelligence correlate at roughly .14. That is a literature estimate, not a correlation observed in Emotiscale data. They are not the same thing measured two ways, so we never average them into a single headline figure. You get a facet profile and a separate situational band.

Facet scores are reported as a percentage of the maximum and labelled as such. There is deliberately no mean-100 score with a standard deviation of 15: that is the IQ scale, and borrowing it would imply a comparability that does not exist.

Every facet figure is published with a margin of about 9 points either way, and the total with about 8. Those margins describe one reading. Comparing two sittings compounds measurement error, so the reliable-change threshold is 13 points rather than 9. Both values currently rest on the reliability assumptions below.

Reliability, and what these numbers are not

We currently model facet reliability at 0.9 and total reliability at 0.92. These are projections from item count and typical item intercorrelations, not measurements from our own data. They are what the margins of error are derived from, so they are stated here rather than buried.

They will be replaced with observed values, and this page updated, once enough sittings have accumulated. If the observed figures come out lower, this page will say so and the margins will get wider. That is the point of publishing the assumption.

Quality gates

Two instructed-response checks are woven into the questionnaire, and a longstring rule flags a run of 12 or more identical answers.

If either trips, we tell you which one and offer a free retake. We do not silently discard the sitting and we do not silently score it. A carelessly answered sitting that is quietly scored is a result someone might act on, and it also quietly poisons the norm sample.

A per-item response-time floor is not currently applied. Calibrating one needs our own timing distribution, and an uncalibrated floor would fail honest fast readers.

What this instrument cannot do

• It is a self-report questionnaire. It measures how you see yourself, which is useful and is not the same as how you come across.
• It is not clinical, not diagnostic, and cannot identify any health condition.
• It should not be used to decide anyone’s job, place or care.
• Emotional skill is trainable, but the published effects are moderate rather than dramatic. Treat anyone promising a fast jump with suspicion.
• We do not do face or voice emotion reading, and we will not. The validated image sets are not licensed for commercial use, and reading a person’s own face or voice is a prohibited practice in workplace and education settings under the EU AI Act, outside narrow medical and safety uses.

Version

Norm version v0-preview. Every result and every certificate is stamped with the version in force when it was produced, so a verified certificate never silently changes meaning.