Study at a glance
- Authors: Paulo A. Graziano, Karen Derefinko
- Year: 2013
- Journal: Biological Psychology 94(1): 22–37
- DOI: 10.1016/j.biopsycho.2013.04.011
- N: 44 studies, ~4996 children (publication years 1992–2012); effect sizes: externalizing k=25 N=3331, internalizing k=18 N=1697, social k=15 N=1969, cognitive/academic k=10 N=1025
- Design: systematic review + random-effects meta-analysis (Fisher’s z, Hedges & Vevea 1998)
Key findings
- Greater RSA withdrawal (RSA-W) related to fewer externalizing (r = −.16, 95% CI [−.19, −.12]), internalizing (r = −.16, [−.21, −.12]) and cognitive/academic (r = −.16, [−.22, −.10]) problems — three small, negative (adaptive) effects, each p < .001.
- No reliable association between RSA-W and social problems (r = −.06, 95% CI [−.10, −.01]; below Cohen’s small threshold, treated as null). The authors call this the study’s ‘most surprising and noteworthy’ result — counter to what Polyvagal Theory and the Social Engagement System predict, since social functioning is exactly the domain PVT stakes the ventral vagus on.
- Sample type reverses the social-problems effect: negative/adaptive in community/healthy samples (M = −.14) but positive/maladaptive in clinical/at-risk samples (M = +.16), F(1,2003)=… t=−2.67, p<.05. For clinically/at-risk children, more RSA withdrawal tracks MORE social problems.
- Ethnicity moderated externalizing: the adaptive (negative) RSA-W↔externalizing association was stronger among studies with a higher % Caucasian children (β = −.62, p < .01). No moderation by respiration measurement.
- Co-varying baseline RSA strengthened the RSA-W↔outcome associations (co-varied M = −.19 vs not co-varied M = −.09, F(1,21)=6.16, p=.02). Measuring vs estimating respiration did NOT change effect magnitudes (p=.70); type of challenging task was only marginal overall (F(1,21)=2.62, p=.078), with negative/stress tasks trending to larger effects than cognitive tasks (p=.041 on probing).
- Baseline RSA and RSA-W were positively correlated (r=.427) and both rose with age. Clinical/at-risk children showed lower baseline RSA (weighted M=6.09) than community children (M=6.26), lower RSA-W, and a tighter baseline↔withdrawal coupling (r=.36 vs .20) — i.e. less flexible physiological range.
- File-drawer robustness was uneven: externalizing (113 null studies needed) and internalizing (65) are stable; social (30) and cognitive/academic (30) are more fragile.
Limitations
- Only linear RSA-W↔outcome relations examined; the authors repeatedly note a quadratic (optimal-mid-level) function is plausible for internalizing and cognitive/academic domains (citing Marcovitch et al. 2010) and cannot be ruled out.
- Heterogeneity was moderate-to-high (I² = 54% externalizing, 54% internalizing, 73% social, 39% cognitive), and the social/cognitive cells rest on few studies (k=15, k=10).
- RSA-W collapses vagal withdrawal and vagal augmentation into one continuous score; the paper argues augmentation may be a distinct (maladaptive/hypervigilant) endophenotype that a linear index conceals — the same withdrawal-vs-augmentation confound Wesarg (2022) later flags.
- Very few studies used positive-mood-induction challenge tasks (k=2), so the adaptive value of RSA-W in rewarding contexts is essentially untested; minority samples are sparse, weakening the ethnicity moderator.
- Correlational and cross-sectional throughout — direction of the RSA-W↔functioning link is not established.
Connections
- Methods: Respiratory sinus arrhythmia (RSA), Heart rate variability (HRV)
- Concepts: Respiratory sinus arrhythmia (RSA), Polyvagal theory, Autonomic specificity of emotion, Early adverse life events
- Debates: Is Polyvagal Theory scientifically valid?
- Related studies: Wesarg et al. (2022), Zahn et al. (2016), Porges (2023), Grossman (2023), Kerdabadi et al. (2024)
The second of the three RSA meta-analyses queued in raw/papers/Polyvagal/ to reach the wiki (after Wesarg et al. 2022), and the one that puts Polyvagal Theory’s flagship claim — that the vagus underwrites social engagement — to a direct predictive test. Pooling 44 studies and ~4996 children, Graziano & Derefinko ask whether RSA withdrawal (RSA-W: the drop in respiratory sinus arrhythmia from rest to a challenge, the canonical “vagal brake” response) predicts children’s adaptive functioning across four domains. It does — small and adaptively, for behaviour and cognition — but not for social functioning, the one domain the theory most insists on.
What was tested
RSA-W is the reactivity measure of the vagal literature: baseline RSA indexes resting parasympathetic tone, and its withdrawal under challenge is read (per Porges) as the vagal brake disengaging to mobilize metabolic resources for coping. Greater withdrawal has been treated as a marker of good emotion regulation and flexible engagement. The meta-analysis regresses RSA-W (computed as baseline minus challenge RSA) against four adaptive-functioning outcomes:
- Externalizing problems (aggression, hyperactivity, inattention, conduct)
- Internalizing problems (anxiety, depression, worry)
- Social problems (peer relations, social competence, social preference)
- Cognitive/academic problems (executive function, school readiness, achievement)
The a-priori expectation, drawn explicitly from Polyvagal Theory and Porges’s Social Engagement System, was that the strongest adaptive association would be in the social domain. The result inverts that ranking.
The central results
| outcome | weighted r | 95% CI | k | N | I² | file-drawer |
|---|---|---|---|---|---|---|
| Externalizing | −.16 | [−.19, −.12] | 25 | 3331 | 54% | 113 |
| Internalizing | −.16 | [−.21, −.12] | 18 | 1697 | 54% | 65 |
| Cognitive/academic | −.16 | [−.22, −.10] | 10 | 1025 | 39% | 30 |
| Social | −.06 | [−.10, −.01] | 15 | 1969 | 73% | 30 |
Three small, robust negative (adaptive) effects — greater RSA withdrawal → fewer problems — and one near-zero, below Cohen’s small-effect threshold (.10) and treated as a null. The paper’s own summary: the “most surprising and noteworthy aspect” is the absence of an overall RSA-W↔social-functioning link, “counter to what one would expect based on Polyvagal theory and the Social Engagement System.”
The moderators that matter
- Sample type flips the social effect. In community/healthy samples RSA-W↔social problems is negative/adaptive (M=−.14); in clinical/at-risk samples it is positive/maladaptive (M=+.16). So for at-risk children, more vagal withdrawal accompanies worse social functioning — the opposite of the theoretical prediction. Whatever the vagal brake buys socially, it is not uniform across children.
- Ethnicity moderates externalizing. The adaptive RSA-W↔externalizing association is stronger among studies with a higher % Caucasian children (β=−.62). The authors read this cautiously (small minority-sample sizes, power to detect), and flag it as a call for diverse samples — Latino children being the largest under-studied U.S. group.
- Co-varying baseline RSA strengthens everything (M=−.19 co-varied vs −.09 not), consistent with the law of initial values and with baseline RSA and RSA-W being positively coupled (r=.427). Notably, whether respiration was actually measured vs estimated from norms did not change the effects (p=.70) — a measurement caveat the field worries about (see Respiratory sinus arrhythmia (RSA)) that here made no difference to outcome prediction. Type of challenging task was only marginal.
- Clinical children start and move less. Clinical/at-risk samples had lower baseline RSA and lower RSA-W than community samples, and a tighter baseline↔withdrawal coupling (r=.36 vs .20) — less independent physiological range, “a disadvantage in terms of their overall ability to react and respond to environmental demands.”
Why it matters here
For the polyvagal debate. This lands on Is Polyvagal Theory scientifically valid?’s evidential-robustness axis alongside Wesarg, but it is arguably the sharper of the two for the theory specifically: Wesarg finds RSA weakly tied to a developmental risk factor (childhood adversity); Graziano & Derefinko find RSA-W essentially unrelated to social functioning, which is the exact outcome the Social Engagement System is built to explain. The meta-analysis does not test PVT’s dorsal/ventral brainstem anatomy (the mechanism axis Grossman and Porges fight over), so it is a null within the vagal-reactivity paradigm, not a critique of it — but it removes an empirical prop the theory has leaned on. As the discussion puts it, “the phylogenetic contributions to social behavior outlined by Polyvagal Theory have little empirical support for individuals with other significant clinical impairment.” The wiki records it as evidence about predictive robustness, not mechanism.
Against a simple “more withdrawal is better” reading. The paper repeatedly gestures at a quadratic (optimal-mid-level) function for internalizing and cognitive/academic outcomes, and the sample-type reversal for social problems shows the linear-adaptive story is context-dependent. This is a physiological cousin of the recurring wiki lesson that body-based individual-difference measures carry less clean signal than the theories built on them assume — the same deflation Wesarg reports for adversity and that the reliability/domain-generality literatures reach for interoceptive-accuracy measures.
Reliability of the finding, and its brakes
Strengths: 25-year comprehensive search, dual-coder reliability (all ICC > .90, κ = 1.00), random-effects model, moderator and file-drawer analyses. Brakes, mostly the authors’: only linear relations were modelled (a quadratic optimum is live and untested); heterogeneity was high in the social cell (I²=73%) where the surprising null sits; the social and cognitive/academic cells rest on few studies (k=15, k=10) with modest file-drawer robustness (30 each); and RSA-W conflates withdrawal and augmentation into one score, so a genuine but bidirectional reactivity effect (some children over-withdrawing, others augmenting) would wash toward the observed values — the same confound Wesarg (2022) independently flags and recommends a pattern-based approach for. Correlational throughout, so the RSA-W↔functioning direction is unestablished.