Method at a glance
- Modality: respiratory
- Measures: Respiratory interoceptive sensitivity (the resistive load needed for reliable detection) and precision (slope), plus metacognitive bias and efficiency — an automated psychophysical replacement for manual filter-based breathing-load tasks.
- Procedure: Two successive inhalations through a computer-controlled respiratory circuit; a precision step motor compresses the tubing on one randomly chosen breath to apply a resistive load, the other is unloaded. The participant judges which breath was more difficult and rates confidence on a Guess-to-Certain VAS. A Psi adaptive staircase sets load magnitude across ~120 trials (~35 min, with breaks every 20 trials to limit hyper-/hypoventilation); a cumulative Gumbel psychometric function is fitted for threshold, slope and lapse rate. Headphones play continuous noise to mask the motor.
- Also called: RRST, respiratory resistance sensitivity, Nikolova task
Try it on RADlab
- Ebb & Flow: a related respiratory detection task with the same psychophysics (adaptive staircase, confidence ratings for metacognition), but the change to detect is in breathing pace, not resistive load.
Strengths
- Automated and continuously graded. Replaces the manual addition of physical filters — the older respiratory method’s low stimulus granularity and unreliable performance estimates — with a motor-driven load an adaptive algorithm can place anywhere on a continuum.
- Yields precision, not just detection. Filter-based tasks give no slope. The RRST does, which is why it can speak to Interoceptive precision.
- Unimodal. Unlike the HRDT, the judgment is made entirely within the respiratory channel with no exteroceptive comparison stimulus — a cleaner measurement, though it creates the task-asymmetry problem when the two are compared.
- Better-behaved metacognition than the cardiac task. In Banellis et al. (2026), 10 of 241 participants were excluded for implausible M-Ratio against 96 for the HRDT — respiratory metacognitive estimates are markedly more stable.
- Two-alternative forced choice with a physically specified stimulus, so — unlike heartbeat counting — there is no way to score well from a belief about one’s own physiology.
Limitations
- Loaded inhalation is a perceptually loud stimulus. Detecting added airway resistance is closer to a somatosensory-effort judgment than to the faint-visceral-signal detection the cardiac tasks attempt; whether the two tasks are sampling comparable perceptual difficulty is unaddressed, and the observation that respiratory sensitivity and precision correlate within the task (r = 0.337) while nothing correlates within the cardiac task suggests they may not be.
- Measures one respiratory feature. Resistance is not breath duration, frequency, depth or air hunger; the authors note a duration-discrimination respiratory task (Bodart et al. 2024) also fails to correlate with cardiac interoception, which corroborates the null but leaves the channel under-sampled.
- Long and effortful. ~35 minutes and 120 loaded inhalations invites fatigue; the enforced breaks address hyper/hypoventilation, not attentional decline.
- Requires a bespoke motorised circuit — not a method a lab can adopt without hardware.
- Substantial exclusion for low performance / equipment malfunction (48 of 315 in Banellis et al.), so the analysed sample is selected on task compliance.
- Voluntary control of the breath is available to the participant throughout and is neither measured nor constrained; inhaling harder or slower is a strategy the score cannot distinguish from perceiving better. Compare Interoceptive control, where the same drivability is the point rather than a nuisance.
Connections
Introduced by Nikolova et al. (2022), The respiratory resistance sensitivity task: An automated method for quantifying respiratory interoception and metacognition — the respiratory half of the Aarhus psychophysical battery. Met first-hand through Banellis et al. (2026); the methods paper is not in raw/.
Its place in the wiki’s respiratory story
Respiratory interoception has argued for several ingests that the field’s cardiac bottleneck has an under-used respiratory answer, and that the breath’s exceptional property — immediate voluntary drivability — makes it both the most tractable channel and the least representative one. Three respiratory instruments now sit on the wiki, and they carve that property differently:
| task | what the participant does | what drivability means here |
|---|---|---|
| RRST | passively judges which of two inhalations was harder | a nuisance — breathing strategy is uncontrolled variance |
| Filter Detection Task (Harrison et al. 2021) | detects added filter resistance | same, with coarser stimulus steps |
| Respiratory tracking task with visuomotor perturbation (respyra) | actively drives the breath to a target | the entire construct (Interoceptive control) |
The RRST is the perceptual member of that set: it treats the breath as a signal to be read, which is what makes it the right partner for the HRDT in a cross-modal comparison, and what makes the comparison a comparison of perception only.
What it bought
Two things the older respiratory literature could not supply.
A slope. Filter-based tasks add discrete physical filters, so the stimulus space is coarse and no reliable psychometric function can be fitted. Without a function there is no precision parameter — the quantity every computational account of interoception is actually about. The RRST’s motor-driven continuum makes it estimable.
(What the FDT gives up in slope it partly buys back in metacognitive estimability: its staircase pins first-order accuracy between 60% and 85%, which is the band where M-Ratio is well-behaved. Now that the FDT has its own page, read the pair as a trade rather than a straight upgrade.)
A matched null. Because RRST and HRDT parameters come out of the same hierarchical Bayesian pipeline, a cross-modal correlation between them is not confounded with instrument type. That is the whole evidential force of the domain-generality null: earlier cross-modal work compared incommensurable scores, and this pairing does not.
One result worth keeping in view
Within the respiratory task, sensitivity and precision correlated at r(256) = 0.337 — participants needing a larger load to detect it were also more consistent around their threshold. Within the cardiac task, nothing correlated with anything (BF01 = 8.55 for the same pair).
The wiki should not over-read this, but it is a hint that the two tasks are not psychometrically equivalent instruments pointed at different organs. A coherent internal structure in one and none in the other is at least as consistent with “the cardiac task is noisy” as with “the cardiac system is disorganised” — and if the cardiac measurement is the weaker one, that is an alternative reading of the cross-modal null that Banellis et al. do not raise.