Filter Detection Task (FDT)
Introduced by Harrison, Garfinkel, Marlow, Finnegan, Marino, Köchli, Allen, Finnemann, Keur-Huizinga, Harrison & Pattinson (2021), The filter detection task for measurement of breathing-related interoception and metacognition, Biological Psychology 165:108185. That methods paper is not in raw/; the wiki meets the task first-hand through Harrison et al. (2021, Neuron), where it is one of three measurement layers.
The task the wiki had been naming for three ingests without a page. It appears on respiratory-resistance-sensitivity-task as a row in the instrument table, on allen-2026-respyra and respiratory-tracking-task as the perceptual measure Allen proposes correlating a control measure against, and on is-interoception-domain-general as half of the phrase “comparing a counting score to a filter-detection score.”
What it is for
Add a small resistance to the inhaled breath and ask whether the person noticed. Titrate the resistance until they notice it about three-quarters of the time. Then ask them, sixty times, whether they noticed and how sure they are.
From that you get four numbers that the taxonomies insist are different constructs and most instruments cannot separate:
| quantity | what it is | taxonomy slot |
|---|---|---|
| perceptual threshold (filter number) | how big a load is needed | Farb’s sensitivity |
| decision bias (c) | tendency to say “yes, there was a load” | none of the taxonomies has a slot for it |
| metacognitive bias (mean confidence) | disposition to feel sure | Garfinkel’s sensibility, in part |
| metacognitive performance (M-Ratio) | how well confidence tracks correctness | Garfinkel’s awareness / Farb’s coherence / Khalsa’s insight |
The last two are the efficiency distinction that page argues the interoception literature has repeatedly conflated. The FDT is one of the few interoceptive instruments that delivers both, and in Harrison et al. they behaved differently: anxiety lowered confidence and left efficiency untouched.
Where it sits among the respiratory instruments
Three respiratory tasks now have pages here, and they divide the channel cleanly:
| task | what the participant does | what it yields |
|---|---|---|
| FDT | judges whether a filter load was added | threshold, bias, metacognition — no slope |
| RRST | judges which of two inhalations was harder | threshold and slope, plus metacognition |
| respiratory-tracking-task | actively drives the breath to a target | control, not perception |
The RRST was built as the FDT’s successor and its papers say so: the motor-driven continuum exists because discrete filters cannot support a psychometric fit. So the FDT’s central limitation is documented by the instrument that replaced it, and the wiki should read the pair as a methodological upgrade rather than as rivals.
What the FDT retains is reach. It needs filters and a mouthpiece; the RRST needs a bespoke motorised circuit, which is why the FDT is the one that has been carried into asthma cohorts (Harrison et al. 2021c) and anxiety samples.
The M-Ratio point that the cardiac literature cannot match
metacognitive-efficiency records a structural problem: M-Ratio misbehaves when first-order performance is near chance, and near chance is where cardiac interoception lives — Banellis et al. had to exclude 96 of 241 participants for implausible cardiac M-Ratio against 10 respiratory.
The FDT’s staircase is a partial answer to that. By construction it holds accuracy between 60% and 85%, which is exactly the band where the metacognitive model is well-behaved. You cannot run a heartbeat-counting task at a criterion accuracy — there is no stimulus knob to turn. Here there is.
The cost is what the staircase buys the criterion with: first-order sensitivity is no longer free to vary, so the threshold and the metacognitive parameters are estimated from different parts of the session, and individual differences in sensitivity are expressed as the filter number rather than as task performance.
What it has been used to show
From the Neuron paper, in 60 healthy adults split by trait anxiety:
- Moderate-anxiety participants needed more filters — they were less respiratory-sensitive (3.0 vs 4.0, Z = −2.4, p = 0.01), replicating Garfinkel et al. (2016a) and Tiller et al. (1987).
- They were less confident (metacognitive bias 6.7 vs 6.2, Z = 2.0, p = 0.02), matching the exteroceptive metacognition literature where anxiety and depression lower confidence (Rouault et al. 2018).
- Their metacognitive efficiency was equal (M-Ratio 0.8 vs 0.8, ns).
- Decision bias did not differ — notable because the hypothesis was that anxious participants would over-report load presence, and they did not.
That pattern is worth keeping: anxiety made people worse at the task and less sure, without making their insight worse or their criterion looser. It also runs opposite in sign to the cardiac finding that anxiety and panic go with better heartbeat detection, which is discussed on the study page and on is-interoception-domain-general.
One further FDT result with no counterpart elsewhere in the wiki: metacognitive performance correlated with peak anterior insula activity for resistance-related prediction errors (r = 0.42) in the same participants — the only strong cross-task link in the paper’s whole 16-measure matrix, and evidence for Stephan et al.’s (2016) proposal that prediction error is the route by which bodily deviation reaches metacognition.
Open
- No slope, so nothing about precision. Anyone wanting both metacognition and a psychometric function needs the RRST.
- Never run alongside a cardiac task in an anxious sample, which is the study that would separate the two readings of the anxiety/sensitivity sign reversal.
- The methods paper (Harrison et al. 2021, Biol Psychol) is not in
raw/, so the wiki’s account of the staircase and of the task’s reliability is secondhand from the Neuron paper’s methods section.