Emotional granularity
Barrett’s construct for the resolution of a person’s emotional experience. A high-granularity person reports feeling ashamed on one occasion and disappointed on another; a low-granularity person’s ratings of “ashamed,” “disappointed,” “sad” and “angry” rise and fall together, because what they are reporting is closer to undifferentiated unpleasantness. The claim is not that the second person has fewer emotions but that they carve fewer distinctions in the same affective material.
It sits directly downstream of core-affect. If experience begins as valence and arousal and is made specific by conceptualization, then granularity measures how finely the conceptual step cuts — which is why it is the constructionist tradition’s favourite individual-differences variable, and why the wiki meets it late (it is an emotion-concepts construct, not an interoceptive one).
How it is measured, and why the measure is worth reading carefully
Not by asking people how precisely they feel. The standard operationalization (Kalokerinos et al. 2019, followed by Ventura-Bort et al.) is:
- Collect many emotion ratings per person across many occasions — recalled episodes, sampled moments, or stimuli.
- Compute an intraclass correlation across the same-valence emotion terms for each person, treating participants and terms as random effects.
- Fisher-transform it, and reverse the sign, so that higher = more granular.
So granularity is a covariance index turned upside down. Three consequences the literature does not always foreground:
- It rewards inconsistency as much as precision. A person whose ratings are noisy for reasons unrelated to emotional discrimination — inattention, careless responding, an unfamiliar rating scale — scores as granular. Nothing in the index separates fine discrimination from low reliability.
- It requires a repeated-measures design, which is why it appears here alongside the day-reconstruction-method and experience-sampling rather than in single-session studies.
- Negative ICCs are uninterpretable and get dropped. In Ventura-Bort et al., 12 of 130 participants were excluded from the negative-DRM analysis on that basis — a nontrivial, non-random loss.
The two valences are not one construct
The strongest empirical thing the wiki currently holds about granularity, and it is inconvenient for the construct’s name.
In Ventura-Bort et al. (2021)‘s day-reconstruction data:
- Granularity for negative emotions was substantially higher than for positive ones (t(116) = 13.39, p < .001, d = 1.23).
- The two were uncorrelated (r = .13, p = .17) — against Barrett (1998), which the authors note and do not explain.
- Only negative granularity carried the study’s positive finding (its association with believed interoceptive/emotional accuracy, β = 0.27).
That last asymmetry is not idiosyncratic: the outside literature the paper cites — Barrett et al. (2001), Demiralp et al. (2012), Kashdan & Farmer (2014), Kalokerinos et al. (2019) — consistently finds negative granularity, not positive, predicting emotion-regulation efficacy, depression and social anxiety. “Emotional granularity” may be two abilities with one name, of which one does the predicting.
A structural echo worth recording: Tong & Jia (2017) found positive emotions overlapping enormously on pleasantness (η² = .94) and needing residual appraisal overlap to explain their clustering. If positive emotions are simply closer together in whatever space people rate them in, then a covariance-based index will find less differentiation among them for reasons that have nothing to do with the rater. The floor may be in the emotions, not the person.
What predicts it, on the wiki’s one source
Ventura-Bort et al. split self-reported interoceptive/emotional awareness into two orthogonal factors and found them pushing granularity in opposite directions:
| negative granularity | positive granularity | |
|---|---|---|
| Sensibility (believed accuracy about internal states) | + β = 0.27, p = .011 | ns |
| Monitoring (tendency to attend inward) | − β = −0.31, p = .004 | − β = −0.21, p = .049 |
The Sensibility direction is what constructionism predicts: better conceptual grip on internal states → finer categories (situated-conceptualization; Lindquist & Barrett 2014; Smith et al. 2019).
The Monitoring direction is the interesting one, and the authors’ explanation runs backwards from the usual causal story: people who differentiate poorly may have to attend to their internal state more, because a single coarse category (“bad”) does not specify what to do, so more evaluation is needed. They flag this as speculative — the design is correlational and cannot order the two.
Either direction matters for is-more-interoceptive-awareness-better: attending to the body more is associated with worse emotional resolution, in the one sample where both have been measured.
Its relation to alexithymia
alexithymia is defined as difficulty identifying and describing one’s own emotions; low granularity is the inability to tell them apart. They are close enough that the wiki should be explicit about the difference: alexithymia is measured by self-report about one’s emotional life (TAS-20), granularity is computed from the structure of the ratings themselves. One asks the person; the other watches what they do.
That distinction is doing real work in Ventura-Bort et al.: the TAS-20 loaded on the Sensibility factor, and the Sensibility factor predicted granularity — so the self-report and the behavioural index agree, weakly (β = 0.27, one valence only), in one sample. That is the wiki’s only evidence that the two constructs converge, and it is thin.
The open questions
- Is it a trait? Every source here measures it once, over one week or one session. Given state-vs-trait-interoception’s finding that half the variance in self-reported interoception is within-person, the same partition has not been attempted for granularity.
- Is it interoceptive at all? The wiki holds granularity because it is the outcome that interoceptive sensibility predicts. Nothing yet ties it to interoceptive accuracy, and the one study here measured no bodily signal of any kind.
- Does it cause anything? Every association in this literature is correlational. The outside citations run “granularity → better regulation,” Ventura-Bort et al. suggest “poor granularity → more monitoring,” and no design here can separate them.