Heartbeat detection and the experience of emotions
Wiens, Mezzacappa & Katkin (2000), Cognition and Emotion 14(3):417-427. A citation the wiki had been carrying secondhand — it appears in Pollatos et al.’s (2005) roll-call of the pre-Dunn consensus (Schandry 1981, Wiens et al. 2000, Critchley et al. 2004, Ferguson & Katkin 1996, Montoya et al. 1993) — now read first-hand. (The PDF’s “2010” is a Taylor & Francis online-repost date; the paper is 2000, filed by journal volume, the same convention used for nummenmaa-2014-bodily-maps and van-der-does-2000-heartbeat-perception-reanalysis.)
It is a small, clean study that turns out to matter more than its size, for three reasons the wiki cares about: it demonstrates the intensity-not-valence dissociation first (five years before Pollatos, ten before Dunn); it does so with the Katkin discrimination task, not Schandry counting, which bears directly on the wiki’s central measurement dispute; and it is the one study in this cluster that measured bodily arousal and controlled for it, which is the omission the wiki holds against everyone else in the consensus.
The result
Fifty-two undergraduates did two things in one session (order counterbalanced): a heartbeat discrimination task, and rated six film clips (two each for amusement, anger, fear; Gross & Levenson 1995 stimuli) on 9-point intensity and pleasantness scales. Nine qualified as good detectors, 43 as poor.
| measure | good vs poor detectors | statistic |
|---|---|---|
| felt intensity (across all three valences) | good > poor, no film-type interaction | F(1, 50) = 7.61, p < .01 |
| felt pleasantness (valence) | no difference | F < 1 |
| intensity, as a continuous predictor | detection ↔ intensity survives controls | ΔR² = .17, p < .02 |
| heart rate (rest / task) | no group difference | ns |
| skin-conductance responses (rest / films) | no group difference | ns |
The dissociation is the headline: perceiving your heartbeat better raises how intensely you feel emotion, and does nothing to how pleasant it is — and the intensity effect is the same size for a comedy clip, a murder scene, and a stalking scene, so it is not about any particular emotion. This is the arousal-vs-valence split, demonstrated in 2000 on a magnitude-vs-hedonic self-report.
Why this is the sturdiest leg of the consensus, not just the earliest
The wiki’s standing complaint about the pre-Dunn tradition — made sharpest on Pollatos — is that the body is never measured: the chain “better perception → more perceived bodily change → more felt emotion” is asserted with its middle term missing, and a rival account (“these people’s hearts are louder, and a louder heart means both an easier detection score and higher tonic arousal”) predicts the whole result with no perceiving in it.
Wiens et al. close part of that gap that Pollatos leaves open. They recorded skin conductance throughout and found:
- good and poor detectors did not differ in electrodermal activity, at rest or across films;
- the intensity effect was unchanged when electrodermal activity was entered as a covariate;
- the continuous relationship survived controlling for sympathetic arousal explicitly (arousal alone: R² = .09, ns).
So the “louder-heart → more-arousal → more-intense-emotion” pathway is ruled out: good detectors were not more sympathetically aroused, yet felt more. The authors call this the “decoupling” of self-report from arousal level and read it as evidence for genuine perceptual mediation — you feel more because you perceive more, not because there is more to perceive. That is a real design advantage over every other member of the consensus the wiki has read.
But the closure is partial, and the wiki should say exactly how far it goes. Skin conductance indexes sympathetic outflow; the cardiodynamic confound that troubles the whole heartbeat-detection-task literature is about stroke volume — the mechanical loudness of the beat as a signal to be detected — which Wiens et al. did not measure. It remains possible that good detectors had more forceful hearts (easier to detect) without more sympathetic arousal (SCR), and the study cannot exclude that. What it excludes is the arousal confound, not the signal-amplitude one. The distinction matters because it is the signal-amplitude version, not the arousal version, that the exercise manipulation later turned from a correlation into a demonstration.
The task is the point: this is discrimination, not counting
The single most consequential fact about this paper for the wiki is which task it used. Not Schandry mental counting — the Eichler-Katkin preferred-interval discrimination task: tones are played at a fixed delay of either 200 ms or 500 ms after each R-wave, and the participant judges, forced-choice, whether the tones were simultaneous with or delayed from their heartbeat. Because the tones are triggered by the participant’s own ECG they carry the true rate and rhythm and differ only in timing, so — unlike counting — the task cannot be passed by guessing from a belief about one’s resting heart rate (the “educated guess” confound of counting; Ring & Brener 1996). Wiens et al. use the instrument built to fix counting’s central flaw.
This lands on is-the-heartbeat-counting-task-valid in two directions at once:
- For the sceptics, on prevalence. Only 17.3% of unselected undergraduates met the discrimination criterion. That is almost exactly the Van der Does prevalence for counting (7.9% of controls, 17.1% of panic patients under a <10%-error rule). The “most people cannot do it” finding therefore generalizes across task family — it is not an artefact of the counting format. Whatever cardiac accuracy is, it is genuinely rare, on the discrimination task too.
- For the consensus, on the effect. Yet in that same discrimination task the emotion-intensity relationship is present — which the wiki should hold against its own frontmatter caution (carried from Van der Does) that discrimination/tone-pip variants “typically find poor (chance-level) performance and no differences among groups.” That caution is about the Brener-Kluvitse tone-pip matching method, where tones interfere with the count; the preferred-interval discrimination method here is a different design, and it finds both a rare accurate minority and a real emotion effect in them. The two discrimination variants should not be lumped.
A nuance to keep straight, so this paper is not misread as contradicting Ehlers. Ehlers (1993) reports that the panic-patient group difference is absent on Katkin discrimination (present only on Schandry counting). Wiens et al. are not measuring a panic group difference — they are measuring an individual-differences correlation between discrimination ability and emotion intensity in healthy undergraduates. Different question, different population; no conflict. If anything the pair is informative: the discrimination task may not separate panic patients from controls while still tracking how intensely healthy people feel — which is what you would expect if the panic effect is an attentional/belief artefact (Ehlers’s own deflationary reading) rather than a perceptual one.
Where it sits in the arousal/valence record
With this ingest the wiki’s valence dissociation gains a fourth and earliest leg:
| source | year | task | arousal/intensity side | valence side |
|---|---|---|---|---|
| Wiens et al. | 2000 | discrimination | good > poor (F = 7.61), arousal-controlled | nothing (F < 1) |
| pollatos-2005-interoceptive-awareness-erp | 2005 | counting | good > poor (F = 5.90) | nothing (F = 0.14) |
| dunn-2010-listening-to-your-heart | 2010 | counting | null as main effect; interaction only | nothing (ΔR² = .00) |
| ferre-2024-emotion-prototypicality | 2024 | lexical norms | — | nothing (r = −.043) |
On the valence null Wiens is a clean fourth confirmation, and the earliest. On the arousal/intensity main effect — the point where Pollatos and Dunn actually disagree — Wiens sides with Pollatos: there is a main effect of accuracy on felt intensity. That shifts the tally on the disagreement to 2–1 (Wiens + Pollatos find the main effect; Dunn does not), and it does so with the strongest design of the three, since Wiens is the only one that measured and controlled bodily arousal. It does not settle it — Wiens measured intensity (a magnitude rating), not arousal per se, and like Pollatos never measured a cardiac response to the stimuli, so it cannot test Dunn’s gain-term model, only its margin. But the wiki should stop treating Dunn’s null as the field’s verdict: two of the three designs that looked found the effect.
Honest weaknesses, recorded
- Nine good detectors. The between-groups effect is an extreme cell; the continuous regression (ΔR² = .17) is the more trustworthy analysis and it agrees.
- Correlational, and the authors are the ones who insist on it — they close by proposing feedback training to manipulate detection ability and watch emotion move, exactly the causal test the wiki keeps noting is missing everywhere in this literature.
- Stroke volume unmeasured (see above) — the arousal confound is closed, the signal-amplitude confound is not.
- Two single-item scales. Intensity and pleasantness only; no discrete-emotion resolution, no non-self-report emotion measure. The intensity/valence dissociation is therefore a dissociation between two self-report dimensions, which is what makes it comparable to Pollatos’s SAM ratings but also what limits it to the same self-report level.
Bottom line
The consensus the wiki had been meeting only through Dunn’s rebuttal of it has a first member on the record, and it is a better study than its reputation as a citation suggested: it demonstrated the intensity-not-valence dissociation first, used the task designed to beat counting’s confounds, found the same rare-accuracy prevalence Van der Does later found for counting, and — alone in the consensus — measured arousal and showed the emotion effect survives it. What it cannot do is what none of them can: rule out that a louder heart drives both the score and the feeling, and prove that perceiving, rather than having, the bodily signal is what does the work.