Method at a glance
- Modality: respiratory (with visual feedback)
- Measures: Voluntary respiratory control rather than respiratory perception: how accurately a person can drive their breathing to match a prescribed waveform (veridical MAE), and how much accuracy costs when the visual feedback mapping is distorted (perturbation ratio). The first instrument in this wiki aimed at the regulation slot of the interoceptive taxonomies rather than the perception slots.
- Procedure: The participant wears a wireless force-sensing respiration belt (Vernier GDX-RB, 0-50 N, ~10 Hz over BLE) and watches a scrolling real-time trace of their own breathing on screen, with a target dot traversing a sinusoidal path (typically 0.1 Hz / 6 breaths per minute). A 15 s range calibration establishes their comfortable breathing range (5th-95th percentile clipping, scaled by 0.80 to set target amplitude). Each trial then runs baseline (10 s natural breathing, sets a trial-specific centre), countdown (3 s, target blends in from the current respiratory position), and tracking (n cycles; force, target and signed error logged every sample). Dot colour gives continuous HSV error feedback (green through yellow to red). On perturbed trials the displayed trace is scaled about the breathing centre by a gain g (f_display = c + g(f_actual - c)), leaving the target where it is, so the task goal is preserved and only the sensorimotor mapping changes. Implemented as an open-source Python 3.10 / PsychoPy package (MIT licence).
- Also called: respiromotor tracking, respyra, breathing tracking task, visuomotor gain perturbation of breathing, closed-loop respiratory biofeedback
Strengths
- Measures a performance against a known physical target, continuously and objectively — the participant never reports anything, so the score cannot be a report of a belief about the body rather than the body. This is the exact failure mode the cardiac tasks cannot rule out.
- Good trial-level reliability in the one validation dataset: split-half r = .759, Spearman-Brown .857.
- The perturbation ratio normalizes remapping cost against baseline ability, so two dissociable individual-differences measures come out of one session — and in the validation data they behaved dissociably (fatigue hit the ratio, spared the baseline).
- Gain is a graded, parametric manipulation: g = 1.0 / 1.5 / 2.0 are one config line apart, so dose-response designs are cheap.
- The RMSE/MAE ratio gives error shape (heavy-tailed vs steady) alongside error size, and in validation recovered a breath-phase effect the mean missed.
- Cheap and unspecialized: a commodity educational-supplier belt and a laptop, no biomedical hardware, no MRI, no physiological expertise. Deployable in teaching labs, which matters for a field whose instruments mostly are not.
Limitations
- It cannot show that interoception is in the loop. The feedback is visual; a participant could in principle track well on exteroceptive information with the interoceptive contribution doing nothing. No condition in the paradigm as published separates respiratory control informed by afference from respiratory control informed by the screen.
- Validated on N = 1 (the developer), 48 trials, one sitting. Naive-participant performance, test-retest across days, and any clinical sensitivity are entirely unknown.
- Calibration sets task difficulty and is fragile: an over-deep calibration breath produces an unsustainable target amplitude, and a tightened belt compressed the usable range and produced air hunger in one validation session. The apparatus is partly setting the difficulty.
- Under perturbation the reported error is visual MAE, not physical tracking error, so scores across gains are not comparisons of what the chest did.
- 10 Hz sampling suits ~0.1 Hz breathing but misses sub-second error dynamics; single-band chest force is coarser than dual-band respiratory inductance plethysmography (RIP).
- Only works on a channel that can be voluntarily driven, i.e. respiration. There is no gastric or cardiac version of this task, which is a limit on the construct as much as on the instrument.
- Instructing a breathing pattern changes the physiological state being measured (0.1 Hz paced breathing is itself an autonomic intervention — see device-guided breathing (DGB)), so the task is never a neutral observation of the respiratory system.
Connections
- Key studies: Allen (2026)
- Related methods: device-guided breathing (DGB), Heartbeat detection task, Interoceptive exposure, Ecological Momentary Assessment (EMA), Skin conductance response (SCR)
Introduced with the Allen (2026) ingest. Structurally the odd one out among this wiki’s methods, and worth stating why plainly.
What kind of measure this is
Every other measurement method here asks the participant to detect, discriminate, rate or report. Count your heartbeats (Heartbeat detection task). Say how much you notice your body (Multidimensional Assessment of Interoceptive Awareness (MAIA)). Colour where you feel it (emBODY (topographical bodily sensation mapping)). Rate how strongly a property comes to mind (Property association task).
This one asks them to do something to their body and scores how well they did it. The dependent variable is a physical error signal against a known target, sampled ten times a second, with no report anywhere in the causal chain. That is a genuinely different measurement relation, and it is the source of both the method’s advantage and its central unresolved problem — see Interoceptive control.
The two scores
| Score | What it is | Reads as |
|---|---|---|
| Veridical MAE (g = 1.0) | mean absolute deviation between target and breathing trace | baseline respiratory control ability |
| Perturbation ratio (perturbed MAE / veridical MAE) | cost of the distorted mapping, normalized | sensorimotor flexibility / remapping cost |
The pairing is the design’s point. Allen’s worked example: low veridical MAE with a low ratio means good control and good flexibility; high veridical MAE with a low ratio means poor control but intact adaptability. Raw performance alone cannot distinguish these, and in the validation data the two came apart — accumulating fatigue degraded the ratio across sessions while veridical tracking stayed flat.
RMSE/MAE ratio is the third number and the least obvious one. Above √(π/2) ≈ 1.253 means heavy-tailed error: intermittent large deviations rather than uniformly larger error. It distinguishes “harder” from “unstable,” and it is what caught the inspiration/expiration asymmetry.
The exteroceptive problem
The honest limitation, stated once here and once on the study page because it governs how far anything measured with this task can be pushed:
The participant is watching a screen. The screen shows their breathing. They correct toward the target. Nothing in that loop requires them to feel their breathing at all — the visual channel is sufficient in principle, and vision is fast and precise where interoception is slow and diffuse. A tracking score is therefore an interoceptive measure by substrate (it is the respiratory system being controlled) and not necessarily by information (the control signal may be entirely visual).
Designs that would bear on this, none of them run:
- Remove or degrade the feedback. Track the target with the trace hidden after countdown. Performance under no-feedback is closed-loop on interoception alone, and the veridical-minus-blind difference is the visual contribution.
- Correlate against respiratory perceptual sensitivity (RRST, Nikolova et al. 2022; Filter Detection Task, Harrison et al. 2021). This is Allen’s own proposal. It establishes covariation between perceiving and controlling the breath, not that perception is doing work in the task.
- Perturb the afference rather than the display — inspiratory resistive loading during tracking — so the mapping distortion is interoceptive rather than visual.
Until at least the first of these exists, the wiki should describe this as a task measuring respiratory motor control with interoceptive involvement assumed, not demonstrated.
Where it would earn its keep
- Panic and hyperventilation. Cognitive model of panic and Anxiety sensitivity are entirely about the reading of respiratory sensation. Whether panic patients also control the breath abnormally — worse baseline tracking, or a higher remapping cost — is a question no instrument in this wiki could previously ask.
- Intervention fidelity. device-guided breathing (DGB) is prescribed at ~5-6 breaths/min and the wiki has no way to verify anyone complied. This measures compliance continuously, which turns a fixed-dose assumption into a measured covariate.
- Contemplative training. contemplative training increase objective interoceptive accuracy? has repeatedly come out null for cardiac accuracy while trainees insist something changed. Control is an untested alternative dependent variable, and one that breath-focused practice plausibly trains directly.
- The regulation slot. Interoceptive taxonomy (Farb et al. Box 1)‘s four competing nomenclatures agree on almost nothing except that regulation belongs on the list. None of them measure it.