Interoceptive Accuracy and Panic

Study at a glance

• Authors: Lori A. Zoellner, Michelle G. Craske
• Year: 1999
• Journal: Behaviour Research and Therapy
• DOI: 10.1016/S0005-7967(98)00202-2
• N: 58
• Design: Analogue between–within experiment: infrequent panickers vs nonanxious controls, crossed with safety/no-safety information, tracked across nine 10-min trials with a caffeine (600 mg) arousal induction; heartbeat perception by the Schandry/Ehlers–Breuer mental-tracking task with concurrent ECG

Key findings

• Independent replication of the founding group difference, in an analogue sample. Infrequent panickers counted their heartbeats more accurately than nonanxious controls at baseline (44.7% vs 61.7% error, t(54)=2.16, p<.05; n=58) and across all nine trials (35.6% vs 53.0%, main effect Group F(1,35)=6.88, p<.05; n=40). The difference survived covarying self-rated certainty, perceived speed and perceived pounding — so it is not explained by the groups differing in how confident or how cardically-aroused they felt. This is the wiki’s first first-hand replication of Ehlers & Breuer (1992), from an outside lab, in people who are not panic-disorder patients.
• The arousal manipulation raised everything except the heart — and accuracy tracked the heart. 600 mg of caffeine reliably raised a composite autonomic-arousal index over trials (F(8,144)=23.29, p<.001; trials 3–9 > baseline) via skin conductance, EMG, expired CO₂ and skin temperature — but produced no main effect on heart rate (footnote 1: no Trial effect for ECG). Interoceptive accuracy did not change over trials. The authors conclude arousal is unrelated to accuracy; read against the wiki’s cardiodynamic account the honest reading is narrower — the cardiac signal never moved, so this is not the test of the arousal→accuracy link it was designed to be (see below).
• Within-subject state anxiety improved the score, with no inverted-U. Stratifying each participant’s own trials into their lowest/medium/highest anxiety, error was greatest at the lowest anxiety level (54.7%) and lower at medium (51.6%) and highest (45.8%) — main effect Anxiety Level F(2,81)=10.04, p<.001. The hypothesized curvilinear detriment at extreme anxiety did not appear; anxiety only ever helped. Zoellner & Craske read this as state anxiety magnifying attention to bodily cues; the wiki’s validity debate reads the same pattern as the count-inflation artefact (below), and the study cannot distinguish them.
• Undercounting is near-universal — replicated. 89.7% of infrequent panickers and 88.9% of controls underestimated their heartbeats; only ~7–8% overcounted. General knowledge of one’s heart rate (r=−.14, ns) and time-estimation ability (r=.18, ns) did not correlate with error, replicating Ehlers’s ruling-out of the estimation-strategy artefact. The one self-report that tracked error was perceived heartbeat speed (r=−.32).
• Groups were discriminated by self-report, not physiology. Infrequent panickers reported more anxiety (2.2 vs 1.1) and far more panic symptoms (7.8 vs 2.2, both p<.01) but did not differ from controls on heart rate or any other physiological index — the report/physiology discordance the wiki tracks under Schema-guided symptom perception and Anxiety sensitivity. Panic-symptom reports correlated negatively with error (r=−.30).
• Safety information moved the body but not the score. Participants given no-safety information about the substance showed higher physiological arousal (esp. skin conductance) than those reassured — but the manipulation changed neither self-rated anxiety nor interoceptive accuracy, so the cognitive manipulation of autonomic arousal again failed to touch accuracy.

Limitations

• Analogue sample, not panic disorder. Infrequent panickers selected for ≥1 DSM-III-R panic attack in the past year plus ASI ≥ 23 (group mean 30.4 vs 12.9 in controls). The authors justify the analogue on the grounds that infrequent panickers are more likely to develop panic disorder (Ehlers 1995) and attend to threat like patients — but the founding finding it replicates is about patients, and the extrapolation is theirs.
• The caffeine induction did not raise heart rate, so it cannot test the cardiodynamic confound. The wiki’s cardiodynamic account is specifically about cardiac signal amplitude (stroke volume / HR). Caffeine here raised electrodermal, muscular and metabolic arousal while leaving HR flat, so ‘accuracy is unrelated to arousal’ is established only for non-cardiac arousal — which the cardiodynamic account never predicted a relationship with. The relevant experiment (raise the cardiac signal, watch accuracy) is the Antony et al. (1995) exercise result, not this one.
• The state-anxiety effect cannot separate perception from artefact. Error is absolute (|actual − perceived|/actual), and because almost everyone undercounts, anything that makes a participant count more lowers error without improving perception. If anxious trials raise the expected/counted rate (as the Van der Does artefact predicts), the within-subject anxiety→accuracy effect is manufactured, not perceptual. Zoellner & Craske report unsigned error and never check the count direction, so their attention-magnification reading and the artefact reading are observationally identical here.
• Single methodology, by the authors’ own closing caution. ‘The present findings must be interpreted with caution due to dependence on a specific methodology’ — the effect appears on Schandry mental tracking and, as the authors note, other paradigms (Ehlers et al. 1988a,b tone-pip matching) fail to find it. Cross-sectional/correlational for the trait difference; only longitudinal work could say whether accuracy precedes or follows panic.
• Underpowered and dose-inconsistent in the caffeine phase. Only 40 of 58 participants entered the information/caffeine trials (10 per cell), and the first 18 received a lower caffeine dose than the final 40; the initial 18 were pooled into the baseline comparison to buy power. Several ANOVA cells are small, and the anxiety-stratification is post-hoc.

Connections

• Methods: Heartbeat detection task
• Concepts: Anxiety sensitivity, Cognitive model of panic, Attentional bias to threat, Schema-guided symptom perception, Interoceptive sensitivity, Interoception
• Debates: Is the heartbeat-counting task a valid measure of cardiac perception?, Is more interoceptive awareness better?
• Related studies: Ehlers (1993), Does et al. (2000), Pollatos et al. (2005), Wiens et al. (2000)

Lori Zoellner and Michelle Craske’s 1999 analogue experiment: infrequent panickers vs nonanxious controls on the Schandry mental-tracking task, with a caffeine arousal induction and a safety-information manipulation layered on top. It is the wiki’s first first-hand replication of the founding group difference — Ehlers & Breuer (1992) — and the only source here that set out to manipulate arousal within an anxious sample and measure what it did to the count.

Why it matters to the wiki

Two things. First, it corroborates, from an independent American lab and in a non-patient sample, the result the whole panic-interoception literature rests on: people prone to panic count their heartbeats more accurately. Everything the wiki has on that finding until now traces to Ehlers’s own Marburg programme (read first-hand in Ehlers (1993)) or to the reanalysis she co-authored. Zoellner & Craske are outside that lineage and get the same group difference.

Second, and more usefully, it runs the two experiments the wiki’s open debates most wanted run — does raising arousal raise the score? and does anxiety raise the score? — and the answers land squarely on the validity question, though not the way the authors intended.

The replication, and how clean it is

Infrequent panickers were more accurate at baseline (44.7% vs 61.7% error) and across all nine trials (35.6% vs 53.0%), and the group difference survived covarying self-rated certainty, perceived heartbeat speed, and perceived pounding. That last point is worth holding: the difference is not explained by panickers merely feeling more confident or more cardically aroused during the task. It is a difference in the count itself.

The authors also did the validity housekeeping Ehlers pioneered and found the same things: near-universal undercounting (≈89% in both groups), and no correlation of error with either general heart-rate knowledge (r=−.14) or time-estimation ability (r=.18), which rules out the two obvious non-perceptual strategies. So on the pro-validity side’s own terms this is a solid replication.

The caffeine experiment is a near-miss, and the miss is instructive

Zoellner & Craske’s stated aim was to test whether physiological arousal drives accuracy — the hypothesis the wiki’s cardiodynamic confound answers yes to (a louder heart is an easier signal). They induced arousal with 600 mg of caffeine, confirmed it raised a composite autonomic index across trials, and found accuracy flat. They conclude “no direct relationship between interoceptive accuracy and arousal level.”

But the composite rose through skin conductance, EMG, expired CO₂ and skin temperature — not heart rate, which showed no trial effect at all (their footnote 1). And the cardiodynamic account is not about arousal in general; it is about the amplitude of the cardiac signal specifically. A manipulation that raises electrodermal and muscular arousal while leaving the heart unchanged is orthogonal to it. Their null is real and worth having — accuracy does not ride on general sympathetic/metabolic arousal — but it is not evidence against the cardiodynamic confound, and read carefully it points mildly the other way: the one channel that did not move (cardiac) is the one accuracy is supposed to track, and accuracy did not move either. The decisive experiment (raise the cardiac signal, watch the score follow) is Antony et al.’s (1995) exercise result, where heart rate really did rise and accuracy really did follow it. This study is that experiment attempted on the wrong variable.

The safety-information manipulation reinforces the same point from a second angle: no-safety information raised autonomic arousal (skin conductance especially) but changed neither felt anxiety nor accuracy. Cognitive manipulation of the body left the score alone.

The state-anxiety effect is the study’s sharpest datum, and it is double-edged

Within each participant, error fell as state anxiety rose — greatest at their lowest-anxiety trials (54.7%), least at their highest (45.8%) — with no sign of the hypothesized breakdown at extreme anxiety. Zoellner & Craske read this as state anxiety magnifying attention to bodily cues, and connect it to the attentional-bias tradition: anxiety sharpens the deployment of attention to the body, so the anxious perceive their hearts better. On that reading the study is a within-subject demonstration of Ehlers’s “habit of attending to bodily cues” turned into a state manipulation, and a proposed resolution of why some studies find the panic group difference and some do not (uncontrolled state anxiety as a moderator).

The wiki has to record the competing reading, because the study cannot exclude it. Error here is unsigned (|actual − perceived|/actual), and in a task where nearly everyone undercounts, any process that makes a participant count more moves them toward the true value and lowers their error without improving perception. The Van der Does artefact is exactly such a process: an anxious person who expects a racing heart counts faster and scores as more accurate. Zoellner & Craske’s finding — higher anxiety, lower error — is precisely what that artefact predicts, and they never report whether the anxious trials’ counts shifted up (which would betray the artefact) or whether perception genuinely sharpened. Their attention account and the artefact account make the same prediction and the data do not separate them. See Is the heartbeat-counting task a valid measure of cardiac perception?.

That the same effect reads as “attention sharpens perception” or “anxiety inflates the count toward truth” is not a flaw the authors could have designed away without a signed-error analysis — it is the counting task’s structural ambiguity showing through a clean within-subject manipulation. It is one of the better illustrations in the wiki of why the validity debate is not a caveat but a rival measurement model.

Where it sits

Not a duplicate: no prior Zoellner or Craske page exists, and this is a distinct study from Ehlers (1993) (different lab, different sample, an original experiment rather than a review) though it replicates Ehlers & Breuer (1992). Not a hard contradiction of anything here — it corroborates the pro-validity group difference and its novel results (the non-cardiac arousal null, the within-subject anxiety effect) enrich the open validity and dose debates without overturning existing content. Its lasting value is the pair of near-experiments: the arousal manipulation that missed the cardiac channel, and the anxiety manipulation that improved the score in a way the wiki’s two camps read in exactly opposite directions.