Heart rate variability and self-control—A meta-analysis

Study at a glance

• Authors: Daniela Zahn, Johanna Adams, Jeanette Krohn, Mario Wenzel, Caroline G. Mann, Lara K. Gomille, Vera Jacobi-Scherbening, Thomas Kubiak
• Year: 2016
• Journal: Biological Psychology 115: 9–26
• DOI: 10.1016/j.biopsycho.2015.12.007
• N: 24 articles / 26 studies / 132 effect sizes; N=2317 (mean age 22.44, range 18.4–57.8; mean 50.22% male)
• Design: systematic review + random-effects meta-analysis with robust variance estimation (RVE)

Key findings

• Small positive pooled association between resting HRV and self-control-task performance: r = 0.15, 95% CI [0.088, 0.221], p < .001, with moderate heterogeneity (I² = 56.10%, Q = 57.26, df = 25.13). Higher trait HRV goes with better laboratory self-control, but the effect is small.
• Publication bias reverses the verdict. Egger’s test was significant (t = 3.003, df = 24, p = .006); trim-and-fill imputed 10 missing studies on the left; the bias-adjusted average effect is r = 0.04, 95% CI [−0.038, 0.110], p = .334 — nonsignificant. Adjusting for the observed bias ‘drastically’ reduced the association to near zero.
• No moderator reached significance in meta-regression: self-control domains (emotion regulation / attention control / cognitive control), sets of executive functions (shifting / updating / inhibiting), recording device, HRV measure (HF / RMSSD / SDNN), duration of assessment, age, sex, and study quality all failed to moderate the effect magnitude.
• Averaged (subgroup) effects were significant only for some categories: emotion regulation (r = .171) and attention control (r = .136) but not cognitive control (r = −.006); among executive functions only inhibiting (r = .158); HF (r = .171) and RMSSD (r = .166) but SDNN null (k=3).
• Study quality trended negative — higher-quality studies showed smaller effects (b = −0.178, p = .248, ns overall) — and two single quality components were significant: adjusting for confounders (b = −0.168, p < .01) and using objective rather than self-report self-control measures (b = −0.178, t = −2.48) both drove effects down. Only 6 of 26 studies used self-report self-control measures.
• One influential outlier (Kimhy et al. 2013: largest N=817, oldest mean age 57.1, respiration-adjusted, confounder-adjusted, effect ≈ zero). Excluding it barely changed the pooled effect (r = 0.16) but removed the publication-bias signal entirely (Egger’s t = 0.227, p = .823; trim-and-fill nonsignificant) — so the bias diagnosis rests partly on that one careful study sitting where the funnel is empty.

Limitations

• The meta-analysis was restricted to Task-Force HRV metrics (time/frequency domain) and EXCLUDED RSA estimated with the Porges–Bohrer or peak-to-trough methods, so — as the authors state — the results ‘may only partially be generalizable to Polyvagal Theory’. Only 1 of 327 screened articles was excluded on this ground, so they judge the impact small, but it is a deliberate scope limit against PVT’s preferred quantification.
• Restricted to trait (resting) HRV and to laboratory self-control tasks; Polyvagal Theory and the somatic-marker/reactivity literatures emphasise phasic/state cardiac change, which is outside scope and might yield larger effects.
• Medium-size study sample (k=26) with many unbalanced covariates, limiting power for moderator meta-regressions; 21 effects (n=21) with no computable effect size were entered as conservative zero effects (removing them raised r to 0.17).
• Adults only (18–65, mostly 18–24, mean age 22); the one older sample (Kimhy) confounds age with a long HRV↔task time lag, so the age moderator is uninterpretable. Clinical populations and children/adolescents excluded.
• Control of respiration during HRV assessment could not be coded (125 of 132 effects ‘no assessment of respiration’), so RSA’s respiratory confound is uncontrolled across the corpus.

Connections

• Methods: Respiratory sinus arrhythmia (RSA), Heart rate variability (HRV)
• Concepts: Respiratory sinus arrhythmia (RSA), Central autonomic network, Polyvagal theory, Autonomic specificity of emotion
• Debates: Is Polyvagal Theory scientifically valid?
• Related studies: Wesarg et al. (2022), Graziano & Derefinko (2013), Porges (2023), Grossman (2023), Kerdabadi et al. (2024)

The third and last of the RSA/HRV meta-analyses queued in raw/papers/Polyvagal/, and the one that tests the vagal literature’s most cognitive claim: that trait HRV indexes the prefrontal inhibitory capacity underlying self-control. It is less a Polyvagal test than a test of the neighbouring Neurovisceral Integration Model (Thayer & Lane) — both frameworks predict higher resting HRV → better self-regulation — and it delivers the same cautionary result as its two companions: a small positive effect that does not survive correction for publication bias.

What was tested

Both Porges’s Polyvagal Theory and Thayer & Lane’s Neurovisceral Integration Model hold that resting HRV is a proxy for the inhibitory capacity of a prefrontal–subcortical “central autonomic network” (CAN): the prefrontal cortex exerts vagally-mediated inhibitory control over the heart and, by the same circuitry, over impulses and behaviour. Higher tonic vagal tone should therefore mean better self-control. Zahn et al. pool the studies that put this to a laboratory test — trait (resting) HRV correlated with performance on self-control tasks — and ask two questions: how big is the association, and does anything moderate it.

Self-control tasks were sorted two ways: by content domain (emotion regulation, attention control, cognitive control) and by executive function (Miyake’s shifting / updating / inhibiting). HRV was restricted to the Task-Force-endorsed metrics — HF (high-frequency power), RMSSD, SDNN — which is where this paper quietly diverges from Porges (see below).

The central result, and its reversal

analysis	r	95% CI	note
Pooled (all 132 effects)	0.15	[0.088, 0.221]	p < .001, I² = 56.10%
Excluding 21 zero-imputed effects	0.17	[0.10, 0.240]	p < .001
Trim-and-fill adjusted	0.04	[−0.038, 0.110]	p = .334, ns
Excluding Kimhy outlier	0.16	[0.098, 0.223]	bias signal disappears

The headline is a small, significant positive effect (r = 0.15). The story is what happens to it under scrutiny. Egger’s regression test flagged funnel asymmetry (p = .006), and the trim-and-fill correction — imputing 10 studies missing from the left side — pushed the pooled effect down to r = 0.04, nonsignificant. The authors’ own summary: “there might not be an association between HRV and self-control performance.” Two independent lines converge on the same deflation: study quality trends negatively with effect size (higher-quality and confounder-adjusted studies show smaller effects), exactly the signature of an inflated small-study literature.

The one caveat cutting the other way is the outlier. Kimhy et al. (2013) — the largest, oldest, most carefully confounder- and respiration-adjusted study, with an effect near zero — is where the funnel’s empty corner would be. Remove it and the publication-bias evidence vanishes (Egger’s p = .823), while the pooled effect holds at r = 0.16. So one can read the corpus either as bias-inflated noise around zero, or as a genuine small effect with one influential careful study dragging the funnel test — the paper cannot decide between these and says so.

Why it matters here

For the polyvagal debate. This lands on Is Polyvagal Theory scientifically valid?’s evidential-robustness axis, completing the trio with Wesarg (adversity ↮ vagal tone) and Graziano & Derefinko (RSA-W ↮ social functioning). Each takes a different outcome the vagal frameworks stake themselves on — a developmental risk factor, social engagement, executive self-control — and finds the real-world association thinner than advertised, here thin enough to disappear under bias correction. None tests PVT’s dorsal/ventral brainstem anatomy (the mechanism axis Porges and Grossman fight over), so all three are nulls within the vagal-reactivity paradigm rather than refutations of it — but together they erode the predictive props the theories lean on.

One scope subtlety worth keeping. Unlike its two companions, Zahn et al. explicitly restricted HRV to Task-Force metrics and excluded Porges–Bohrer RSA, conceding the results “may only partially be generalizable to Polyvagal Theory.” This is the mirror image of Grossman’s complaint: where Grossman argues Porges over-claims what RSA can mean, here a mainstream HRV meta-analysis pre-emptively brackets Porges’s preferred quantification out of scope. The effect is that this paper speaks most directly to the Neurovisceral Integration Model — the HRV-as-prefrontal-inhibition account — and only glancingly to PVT proper. The wiki records it as evidence about the predictive robustness of resting HRV for cognition/regulation, in the same key as the reliability and domain-generality deflations it keeps meeting for other body-based measures.

RSA vs interoceptive accuracy. As with the other vagal sources, note the category: HRV/RSA here is an efferent autonomic readout (the vagal control of the heart, requiring no awareness), not an interoceptive-perception score. This meta-analysis is about what resting vagal tone predicts about behaviour, not about whether people can feel their hearts — a distinction the Respiratory sinus arrhythmia (RSA) page keeps deliberately.

Reliability of the finding, and its brakes

Strengths: comprehensive multi-database search (5551 records screened to 24 articles), dual-coder reliability (ICC/κ mostly > .85), robust variance estimation to handle the many dependent effects per study, and a genuine publication-bias examination that the authors let overturn their own headline. Brakes, mostly theirs: a medium study sample (k=26) underpowers the moderator analyses, so the flat moderator profile is as likely low power as true absence; 21 uncomputable effects were entered as conservative zeros; respiration was uncontrolled across almost the whole corpus (the standing RSA confound); the scope is resting HRV and laboratory tasks only, leaving phasic/state cardiac change and everyday self-control untested; and the whole thing is correlational, so even the surviving small effect carries no directional claim.