Study at a glance
- Authors: John Eusebio, Bo Forbes, Charlotte Sahyoun, David R. Vago, Sara W. Lazar, Norman Farb
- Year: 2022
- Journal: Mental Health and Physical Activity
- DOI: 10.1016/j.mhpa.2022.100483
- N: 58
- Design: RCT, two active arms (Interoception-Focused vs Movement-Focused yoga), 10 weeks x 2 sessions/week, three assessments (baseline, post, 1-month follow-up), multilevel models
Key findings
- The differential interoception hypothesis FAILED: MAIA rose over time in both arms (beta = 9.78, p < .001) with no group x time interaction (beta = -2.49, p = .443). Interoception-Focused yoga did not raise self-reported interoception more than Movement-Focused yoga.
- The differential symptom hypothesis also FAILED: PHQ-SADS symptom burden fell over time (beta = -0.57, p = .003), equally in both arms (group x time p = .311), with no subscale differentiation for depression, anxiety or somatic symptoms.
- The behavioural hypothesis SUCCEEDED: Interoception-Focused yoga improved SART accuracy more than Movement-Focused yoga (group x time beta = -0.056, SE = 0.020, p = .006), surviving Holm-Bonferroni correction, with no baseline group difference.
- An aptitude-treatment interaction, and the paper’s most interesting result: concurrent MAIA x group x time predicted SART accuracy (beta = -0.004, p = .007) and NO other questionnaire did (all ps > .15) — attentional gain in the interoceptive arm scaled with interoceptive sensibility specifically.
- The mirror image in the other arm: baseline MAIA x group x time (beta = -0.001, p = .019) — Movement-Focused participants with the LOWEST baseline MAIA improved most, while high-MAIA participants did not improve at all. High interoceptive sensibility was a contraindication for movement-focused training.
- No transfer to lapse frequency: tau (the exponential tail of the RT distribution, the standard mind-wandering index) showed no group x time effect (ps > .110), and post-error slowing did not differ by group (ps > .24). Whatever improved accuracy, it was not a reduction in attentional lapses.
Limitations
- No inactive control arm. Both groups improved on interoception and symptoms, and the authors state plainly that this improvement cannot be attributed to yoga.
- Analysed N = 34 (16 vs 18) against a power analysis specifying 36 for the between-arm interaction — underpowered for the marginal differences between two active treatments, which is the design’s whole question.
- 83% female; the authors name the absence of non-female genders as a generalizability constraint.
- MAIA was used as a TOTAL score throughout. See Multidimensional Assessment of Interoceptive Awareness (MAIA): composite MAIA sums two orthogonal factors with opposite relations to well-being, so both moderation findings are moderations by an unknown item mix.
- No prescribed home practice, no manualized protocol, no independent fidelity assessment — research assistants attended a subset of classes and attendance was recorded, and the authors call that insufficient.
- The training rooms were cold (a hospital conference room), which the authors flag as potentially disadvantaging the slower Interoception-Focused arm specifically.
- Dropout was predicted by baseline symptom burden (PHQ-SADS t = -2.21, p = .036), so the analysed sample is the healthier tail of a sample recruited for symptoms.
- Bo Forbes, who co-designed both protocols, runs Embodied Awareness LLC, an integrative yoga programme with a financial interest in the interoceptive arm’s success. Declared; the authors note she did not access data or set conclusions.
Connections
- Methods: Sustained Attention to Response Task (SART), Multidimensional Assessment of Interoceptive Awareness (MAIA), Ex-Gaussian decomposition of reaction times
- Concepts: Interoceptive taxonomy (Farb et al. Box 1), Interoceptive sensitivity, Decentering, Perceptual inference (interoceptive)
- Debates: contemplative training increase objective interoceptive accuracy?, Is more interoceptive awareness better?
- Related studies: Farb et al. (2015), Farb et al. (2010), Farb et al. (2022), Weng et al. (2021), Ventura-Bort et al. (2021)
The wiki’s first randomized trial that manipulates interoceptive emphasis while holding the intervention otherwise constant. Two yoga curricula were built to be isomorphic — same duration, same instructors’ training, same room, same 20 hours — and to differ in one thing: whether the instruction directs attention to moment-to-moment bodily sensation, or to the correctness and pace of postures. That is as close as this wiki gets to a manipulation of interoceptive training with an active comparator that controls for exercise, group contact, expectancy and instructor attention.
The result is a three-part answer that is more interesting than the hypotheses were.
What the two arms actually were
| Interoception-Focused | Movement-Focused | |
|---|---|---|
| pace | slower, less physically demanding | faster, more postures, higher transition frequency |
| instruction | frequent direction of awareness to fluctuating bodily sensation | correctness of posture, how long it is held |
| language | ”mindful language”, anthropomorphizing the body (“when this pose feels complete to your body…“) | mindful language deliberately avoided |
| stated goal | interoceptive attention tendency, sensibility, awareness, metacognitive awareness | energy and overall fitness |
This is the Interoceptive taxonomy (Farb et al. Box 1) operationalized as a curriculum: the interoceptive arm targets attention tendency and regulation, not accuracy, and nothing in the trial measures accuracy at all.
Three results, in increasing order of interest
1. The self-report hypotheses failed, and the failure is informative. Both arms improved on MAIA and both improved on symptoms, with no differential. The authors’ own reading is the honest one: both interoception and symptom burden are self-report, so expectancy could be driving improvement in both arms and masking any marginal difference. But there is a second reading this wiki should hold alongside it — a curriculum explicitly built to raise interoceptive awareness did not raise a validated interoceptive-awareness questionnaire more than a curriculum built to avoid mentioning the body. That is a direct hit on the assumption underneath most of the wiki’s applied pages. See contemplative training increase objective interoceptive accuracy?.
2. The behavioural difference appeared where nobody trained it. SART accuracy improved more in the interoceptive arm (group × time p = .006, Holm-Bonferroni survived). The authors make the right point about this: there is no overlap between the training and the task beyond the general demand to sustain awareness of the present moment. Watching a stream of digits and withholding on a rare target has nothing to do with yoga. So this is a transfer effect, which is the thing laboratory attention-training studies famously fail to produce (Lintern & Boot 2021).
3. And the moderation runs in opposite directions in the two arms. This is the finding the wiki should carry forward:
| arm | who improved on sustained attention |
|---|---|
| Interoception-Focused | improvement scaled with concurrent MAIA; per the discussion, baseline MAIA did not gate it |
| Movement-Focused | only those with the lowest baseline MAIA improved; high scorers did not improve at all |
Read together: high interoceptive sensibility is an asset under interoceptive instruction and a liability under movement instruction. The authors’ phrasing is that “a strong interoceptive mindset may have led to poorer uptake of yoga benefits in the Movement-Focused group.” That is a matching claim rather than a dose claim, and it is the first one in this wiki with an experimental design behind it — see Is more interoceptive awareness better?.
Crucially, this moderation was specific to the MAIA: the same three-way interaction was run for every other questionnaire in the battery (PSS, Ryff, AAQ-II, Brief-COPE, FFMQ) and none approached significance (ps > .15). With eight questionnaires collected, a selective hit on the one that names the construct under manipulation is worth more than the p-value alone.
What did not transfer, and why it matters
The mechanistic hypotheses about how attention improved all failed. Tau — the exponential tail of the reaction-time distribution, the field’s standard index of lapses and mind-wandering (see Ex-Gaussian decomposition of reaction times) — showed no group × time effect. Post-error slowing did not differ between arms. Attentional lapses were not reduced.
So accuracy rose without lapse frequency falling. The authors’ cautious reading is that the meaning of long RTs may have changed — the interoceptive arm possibly pausing to control errors rather than pausing because of intrusive self-critical thought — supported by the significant accuracy × group × time interaction on tau (β = 85.05, p = .006). That is a reinterpretation of a variable rather than a measurement of a mechanism, and the paper says so.
Two things wrong with the paper, recorded because they bear on how it is cited
The concluding sentence misstates the study’s own results. Section 14 opens: “While both yoga training interventions reduced depressive symptoms and interoceptive accuracy…“. No accuracy measure was administered anywhere in the trial, and the interoception measure that was administered (MAIA) went up, not down. Anyone citing this paper’s conclusion at face value will get both the construct and the direction wrong.
The abstract and the discussion disagree about which MAIA moderated what. The abstract says improvements “were greatest for those endorsing high levels of interoception” in the Interoception-Focused group — that is the concurrent MAIA result (Fig. 5). The discussion says of the same group that “this improvement was unaffected by baseline MAIA scores” (Fig. 6). Both can be true, because they are different predictors, but the abstract does not say which, and the natural reading of “endorsing high levels of interoception” is a baseline trait. The finding that survives careful reading is the Movement-arm one: low baseline MAIA predicted benefit there, and high baseline MAIA predicted none.
Where it sits
This is Farb’s group testing, in an intervention, the framework laid out in Farb et al. (2015) — and the pattern echoes Farb et al. (2022) closely enough to be worth naming. Both are two-active-arm trials where the interoceptive/mindfulness arm fails to beat its comparator on the outcome the theory nominates, while a mechanistic measure moves. The wiki’s standing caveat on Farb’s clinical work — that mindfulness has not been shown to be the privileged route — is not weakened here, but it is qualified: the privileged-route question may have the wrong shape if the effect is an interaction with the trainee rather than a main effect of the training. See Yoga as interoceptive training, contemplative interoceptive training for well-being.