Study at a glance
- Authors: Nicolas Legrand, Niia Nikolova, Camile Correa, Malthe Brændholt, Anna Stuckert, Nanna Kildahl, Melina Vejlø, Francesca Fardo, Micah Allen
- Year: 2022
- Journal: Biological Psychology, 168, 108239
- DOI: 10.1016/j.biopsycho.2021.108239
- N: 223 healthy adults (206 analysed at session 1, 191 at session 2; 179 with both)
- Design: Methods paper with test-retest: heart rate discrimination task (5 s heart listening with pulse oximetry, then five tones faster or slower than the heart, 2-IFC, trial-wise confidence), a matched tone-vs-tone exteroceptive control, and a revised heartbeat counting task, in two sessions about 47 days apart. Psi adaptive staircase; post hoc Bayesian psychometric fits; hierarchical meta-d’/M-ratio.
Key findings
- Heart rate is underestimated. Interoceptive threshold -6.97 BPM at session 1 and -8.50 at session 2 (d = -0.93, -1.03), against a near-zero exteroceptive threshold. Individual interoceptive thresholds ranged from about -38 to +34 BPM, far wider than exteroceptive ones (-10 to +16).
- Cardiac judgments are less precise and less sensitive than tone judgments: slope 15.3 vs 9.6 BPM, d’ 1.43 vs 2.05 (session 1), both replicated at session 2. The interoceptive slope improved more across sessions, suggesting practice effects.
- Cardiac metacognition is less efficient: M-ratio 0.81 vs 0.96, i.e. about 19% of the evidence is lost between cardiac decision and confidence, against 4% for tones; replicated at session 2. Mean confidence was also lower (51.5 vs 61.4).
- Confidence travels across modalities, efficiency does not. Mean confidence correlated rs = .60 between cardiac and tone tasks, while thresholds, slopes and M-ratio were largely uncorrelated across modalities (a modest threshold correlation, rs = .26, appeared at session 2).
- The HRDT threshold tracks heartbeat counting. Interoceptive (not exteroceptive) thresholds correlated with counting accuracy (rs = .29 at session 1, .19 at session 2) and with the number of beats counted, controlling for actual beats (semi-partial rs = .20): people who judge their heart as slower also count fewer beats.
- Threshold is reliable, slope is not. Test-retest r = .51 for the interoceptive threshold and r = .10 for the slope.
Limitations
- The authors state the task measures beliefs about heart rate (a mix of priors, context and afferent signal), not objective sensitivity to ascending cardiac signals, and recommend the method of constant stimuli for that question.
- The slope, the task’s claimed measure of interoceptive precision, had essentially no test-retest reliability (r = .10), partly because session 2 changed the prior range, staircase and response window. As a trait measure of precision it is currently unsupported.
- The exteroceptive reference tones were drawn uniformly from 40-100 BPM rather than matched to each participant’s heart rate, so the two conditions sample different stimulus ranges.
- The 2-IFC design adds a working-memory load (hold a heart-rate estimate, then compare), and a 5 s window sets a floor on precision because of beat-to-beat heart rate variability.
- Healthy, mostly young adults at rest; no stroke volume or other cardiodynamic measure, so the signal-amplitude confound is untouched.
Connections
- Methods: Heart Rate Discrimination Task (HRDT), Heartbeat detection task
- Concepts: Interoceptive precision, Metacognitive efficiency and metacognitive bias, Interoceptive sensitivity, Interoceptive inference
- Debates: Is the heartbeat-counting task a valid measure of cardiac perception?, Is interoception domain-general or organ-specific?
- Related studies: Banellis et al. (2026), Murphy et al. (2020), Desmedt et al. (2022)
The methods paper for the HRDT, from Allen’s Aarhus group. The wiki had described the task second-hand through Banellis et al. (2026); this is the source, with its validation data.
What the task is for
Heartbeat counting gives one error score that mixes perception, estimation and bias (A heartbeat counting score is not, by itself, evidence of interoceptive accuracy). The HRDT instead asks, trial by trial, whether a tone train is faster or slower than your heart, and fits a psychometric function. The threshold is your bias in BPM (where you think your heart is), the slope is how consistent you are, and trial-wise confidence gives metacognition. A tone-vs-tone condition in the same units subtracts what is not cardiac. The subtitle is careful: it measures the accuracy and precision of interoceptive beliefs.
What it found
People think their heart is slower than it is, by 7 to 8.5 BPM on average and by up to 38 BPM individually. Cardiac judgments are less precise, less sensitive and less metacognitively efficient than matched tone judgments. And the underestimation shows up in counting too: the more someone underestimates on the HRDT, the fewer beats they count, with actual beats controlled.
That last result is the one this wiki most needs. The debate over heartbeat counting turns on whether undercounting is missed beats or a belief in a slower heart. Here the same underestimation appears on a task with no counting at all, which favours the belief reading (Van der Does’s side). It also means the HRDT and the counting task partly measure the same thing, a belief about one’s heart rate, which is convergent validity for that construct and not for perception.
The pattern it shares with Banellis
Confidence correlated strongly between heart and tones (rs = .60) while thresholds, slopes and metacognitive efficiency did not. This is the result Metacognitive efficiency and metacognitive bias and Is interoception domain-general or organ-specific? build on from Banellis et al. (2026), found first here with a tone control rather than a respiratory task: confidence bias is a trait of the person, efficiency is specific to the channel.
The caveat the wiki had missed
The slope was the reason the wiki treated the HRDT as the first empirical handle on Interoceptive precision. Its test-retest reliability was r = .10. The threshold was acceptably stable (r = .51). The authors attribute part of the slope’s instability to changes in the staircase between sessions, and they add their own warning that psychophysical “precision” is not the precision-weighting term of predictive processing. Until a reliability study with an unchanged protocol shows otherwise, slope-based individual differences should be read as unestablished.
Placement
Held first-hand on Heart Rate Discrimination Task (HRDT) (now with reliability figures), Interoceptive precision (the slope caveat), Metacognitive efficiency and metacognitive bias, A heartbeat counting score is not, by itself, evidence of interoceptive accuracy and Is the heartbeat-counting task a valid measure of cardiac perception?. No contradiction; one correction of emphasis (slope reliability).