Study at a glance
- Authors: Eszter Ferentzi, Tamás Bogdány, Zsuzsanna Szabolcs, Barbara Csala, Áron Horváth, Ferenc Köteles
- Year: 2018
- Journal: Frontiers in Human Neuroscience, 12:223
- DOI: 10.3389/fnhum.2018.00223
- N: 118 analysed (142 tested; 24 excluded for missing data on more than two tasks)
- Design: Cross-sectional, healthy undergraduates, six channels over four sessions: heartbeat counting (Schandry), water load test (fixed-volume fullness and discomfort ratings), ischemic pain (tourniquet threshold and tolerance), bitter taste (intensity and unpleasantness ratings), elbow joint position matching (one and two arms), one-leg balance with eyes closed. Spearman correlations (Bonferroni, p < .001) and exploratory PCA and ML factor analysis with oblimin rotation.
Key findings
- Of 45 pairwise correlations, only three were significant, and each paired two measures of the same channel: pain threshold with tolerance (rho = .51), bitter intensity with unpleasantness (.80), gastric fullness with discomfort (.48). Most between-channel correlations were between 0 and .1 in absolute size.
- Three-factor solutions (PCA and ML) recovered one factor each for bitter taste, gastric sensation and pain, with factor intercorrelations of .2 or less. No general interoceptive factor appeared.
- Heartbeat counting related to nothing, with low communality; in the PCA it loaded negatively (-.33) on the gastric factor. The positive cardiac-gastric association reported by Whitehead & Drescher (1980) and Herbert et al. (2012) was not replicated.
- The authors recommend writing ‘heartbeat perception accuracy’ rather than ‘interoceptive accuracy’ whenever only the heart was measured.
Limitations
- Most channels were measured as intensity ratings of a delivered stimulus (gastric, bitter, pain), while heartbeat counting and proprioception were accuracy tasks with a correct answer and balance was a motor performance. The null is therefore partly between incommensurable quantities; the authors concede this ‘might result in a decrease in the estimated strengths of associations’.
- The factor analysis rests on weak foundations: KMO = .389, below the usual .5 adequacy threshold, and large missing data (29 of 118 lacked the water load test) handled by EM-imputed correlation matrices.
- Heartbeat counting carries every objection on Is the heartbeat-counting task a valid measure of cardiac perception?, so its failure to correlate with other channels could reflect the task as much as the heart.
- Healthy students at rest only; no respiratory channel, which is the gap Banellis et al. (2026) later filled.
- Two unexplained cross-channel loadings (one-arm proprioception on the bitter factor) suggest some structure is noise or task-specific.
Connections
- Methods: Heartbeat detection task
- Concepts: Interoceptive taxonomy (Farb et al. Box 1), Interoceptive sensitivity, Nociception, Proprioception, Interoception
- Debates: Is interoception domain-general or organ-specific?, Is the heartbeat-counting task a valid measure of cardiac perception?, contemplative training increase objective interoceptive accuracy?
- Related studies: Banellis et al. (2026), Desmedt et al. (2023), Haruki & Ogawa (2023), Wilzok et al. (2023)
The most-cited paper the wiki had not read. It was the backbone of the organ-specific side of Is interoception domain-general or organ-specific? for 35 mentions before this ingest, and reading it first-hand mostly confirms how the wiki used it, with three corrections of emphasis.
What it shows
Six channels in 118 students, ten measures, and no general factor. Heartbeat counting in particular correlated with nothing. Every significant correlation paired two measures of one channel.
What the second-hand account missed
The within-channel correlations are the strongest part. The standing objection to every null on the domain-generality page is that unreliable instruments manufacture nulls. Here, measures were reliable enough to correlate strongly within a channel (bitter taste rho = .80, pain .51, gastric .48) while failing between channels. That does not rule out unreliability for the two channels with no partner (heartbeat counting, balance), but for pain, taste and gastric sensation the null is not a noise artefact.
It is a replication failure for the unitary side, not just a null. The “one interoceptive sense” position rests on cardiac-gastric convergence (Herbert et al. 2012; Whitehead & Drescher 1980). This study measured both and found the relation, if anything, negative. The gastric method differed (fixed water volume rated for fullness, rather than drinking to fullness), so it is a failure to replicate the construct rather than the procedure, and the authors say so.
“Sensitivity” means different things across its channels. Only heartbeat counting and proprioception have a right answer. The gastric, pain and taste measures are how intense or unpleasant a standard stimulus felt, which is closer to sensibility or amplification than to accuracy. So the precise claim is: cardiac counting accuracy does not predict how strongly people rate stomach, pain or taste stimuli, and those ratings do not predict each other. That is still a strong result against a general factor, but it is not the matched-instrument test; Banellis et al. supply that for two channels.
Placement
Now held first-hand on Is interoception domain-general or organ-specific? (the Ferentzi position), Interoceptive taxonomy (Farb et al. Box 1) and Heartbeat detection task. The authors’ naming recommendation is the same rule as A heartbeat counting score is not, by itself, evidence of interoceptive accuracy. Its aside that meditators show no heartbeat counting advantage (citing Khalsa et al. 2008 among others) is second-hand and adds nothing new to contemplative training increase objective interoceptive accuracy?. No contradiction with existing pages.