The decision (Challenge #11). The concern was that high-exertion exercise (HIIT) is offset by compensatory NEAT downregulation or increased intake, so the extra expenditure does not translate. Two questions: does compensation happen, and does intensity make it worse?
Efficacy verdict — compensation is real and large, but intensity is NOT what drives it
Riou et al. 2015 (SR of 61 studies, 928 subjects; compensation = body-energy change over exercise energy expended):
- Compensation is real, and wildly variable: «The overall mean energy compensation was 18% ± 93%.» (Riou et al., 2015) So on average ~18% of the exercise energy deficit is offset — but the ±93% SD means the individual response ranges enormously, from over-compensation to none.
- Construct caveat — the measured 18% is not all biological compensation (corrected 2026-08-08): Riou’s ratio is body-energy change over estimated exercise energy, and the paper lists four contributors, only two of them true compensation: «some form of energy compensation, i.e., increased energy intake (EI), decreased energy expenditure, a small dose of exercise induced energy expenditure (ExEE) [] or simply a lack of compliance to the prescribed exercise [], is occurring.» (Riou et al., 2015) Under-delivered or overestimated ExEE and non-compliance inflate the number in a fixed direction — «reducing the total amount of ExEE and, thus, inflating the energy compensation» (Riou et al., 2015) — and Riou states the split is unrecoverable (no intake/NEAT data). So for a compliant exerciser the true physiological offset is plausibly below the quoted mean, and Riou alone reports no CI on the 18%; the existence claim is carried by Careau’s slope CI (below), not by this ratio.
- Duration is the dominant driver, and it gets worse over time: for short interventions compensation is low; «For studies of longer duration (about 80 weeks), the energy compensation approached 84%.» (Riou et al., 2015) The duration trend is Riou’s most solid predictor (r = 0.30, p < 0.002), but the 84% point figure is a fitted-model extrapolation on thin data — «only a few studies of longer duration were included in this analysis» (Riou et al., 2015) (the ~80-week anchor rests essentially on two Donnelly 2000 arms) (corrected 2026-08-08), so read the direction as solid and the 84% as indicative. This is a mechanism (among the four Riou names above) behind exercise alone underdelivers weight loss -> Does Weight Loss Reduce Cardiovascular Events (why the weight lever disappoints) and the constrained-energy-expenditure picture.
- The intensity-specific worry is NOT SUPPORTED — but by an underpowered test, not refuted (corrected 2026-08-08): «Sex, frequency, intensity and dose of exercise energy expenditure were not significant predictors of energy compensation.» (Riou et al., 2015) This null is weak, and Riou says so: intensity was dichotomised at 60% VO2max — pooling continuous jogging with intervals — the corpus contains few genuine interval protocols, and the group means carry enormous spread: «there was no significant difference for the degree of energy compensation between lower (11.8% ± 122.6%) and higher intensity (20.4% ± 81.4%) (p = 0.75)» (Riou et al., 2015). Riou flags «The dichotomisation of the variable intensity could have reduced the power of the statistical analyses.» (Riou et al., 2015) So this is insufficient evidence, not a refutation — and the non-significant trend even runs slightly toward more compensation at higher intensity. What predicts it is «the interaction between initial fat mass, age and duration» (48% of variance). So HIIT is not shown to compensate more than moderate work, but the intensity question is barely tested — the maintainer’s general concern (compensation is real) is upheld; the specific version (high exertion -> more NEAT downregulation) is neither supported nor adequately tested by this evidence.
Careau 2021 sharpens the magnitude, names the mechanism, and adds the adiposity gradient
The DLW landmark (n=1,754 free-living adults, the IAEA database) reaches the same conclusion by a wholly different method — cross-sectional decomposition of total vs basal energy expenditure, not exercise interventions:
- A cleaner magnitude, and the mechanism: «energy compensation by a typical human averages 28% due to reduced BEE; this suggests that only 72% of the extra calories we burn from additional activity translates into extra calories burned that day.» (Careau et al., 2021) So compensation is not only behavioural (eating more) — a measurable part is a drop in basal energy expenditure, a metabolic adjustment. (The page’s quote locates verbatim in Careau’s Summary — «…we burn from additional activity…»; the «spend on» variant is the In-brief duplicate. Prior REVIEW quote-drift marker closed as a false alarm, verified via srcgrep/cite.py 2026-08-08.)
- The adiposity gradient — the decision-relevant twist: «People that are at the 10th percentile of the BMI distribution compensate 27.7% of activity calories, whereas people at the 90th percentile compensate 49.2% of activity calories … individuals with greater fat levels … are stronger energy compensators or … become stronger compensators as they get fatter.» (Careau et al., 2021) The person carrying the most fat — who most wants to burn it off with exercise — compensates the most, so exercise “counts” least exactly where the weight goal is greatest. Careau flags the causality as undetermined (do compensators get fatter, or does getting fatter raise compensation?), so read it as an association, not a proven within-person law.
- But Riou’s adiposity term runs the OTHER WAY, and the disagreement is real (corrected 2026-08-08): in the same interaction the page credits, higher initial fat mass predicts lower, not higher, compensation: «for a shorter study duration, lower energy compensation was observed in younger individuals with higher initial fat mass (FM). In contrast, higher energy compensation was noted for younger individuals with lower initial FM.» (Riou et al., 2015) (Riou’s FM main effect is null, p = 0.12, and the interaction direction is short-duration only — gone from ~25 weeks — and study-level, not individual-level.) So the two sources point opposite ways on the one moderator both examine: Careau’s between-individual free-living gradient says more fat -> more compensation; Riou’s study-level intervention interaction says (short-term) more fat -> less. Treat the “higher body fat -> exercise counts less” reading as Careau’s free-living finding, not a settled law — the intervention evidence disagrees.
Independent backing [E-independent] — qualified, tells acknowledged. Riou (61 exercise-intervention
studies, body-composition change) and Careau (1,754 free-living adults, cross-sectional TEE-vs-BEE
regression on the IAEA DLW database, explicitly excluding athletes/prescribed-exercise cohorts) reach the
same qualitative claim — a substantial fraction (~1/5 to ~1/4) of activity energy is offset — by
genuinely different datasets and methods, with neither estimate derived from the other. The
independence is at the level of the estimate, and two tells are flagged rather than hidden: Careau
cites the exercise-intervention literature (Riou 2015 among its refs) as contextual motivation — but as
the contrast it is testing against (is compensation the default in normal life, not just under
prescribed exercise?), deriving no magnitude from it; and Careau (Ottawa, Biology) shares a university,
not a research programme or department, with Riou/Doucet (Ottawa, kinetics/nutrition). Under the strict
rule (no shared lineage of the estimate, no citing-as-antecedent-of-the-claim) this remains an E, not a
laundered one — but it is a qualified E. Do not pool the magnitudes — 18% (intervention) and 28%
(cross-sectional) measure related-but-distinct constructs; the convergence is on direction and rough
scale, not a shared number.
What this cashes and refines
- It cashes a named gap. Physical Activity Dose and Mortality recorded that the WHO PA annex ran no compensation analysis and called walking is better tolerated than intense exercise unsupported. Riou now supplies the held evidence: compensation exists and is substantial, but it is not intensity-graded — so neither HIIT is self-defeating via compensation nor walking compensates less is supported. The lever that matters is duration/adiposity, not modality.
- It refines the HIIT-vs-walking note on Measuring and Raising Cardiorespiratory Fitness: HIIT’s VO2max advantage is not eroded by extra compensation relative to moderate work; the real caveats on HIIT remain adherence (unheld comparative evidence) and the flattening mortality curve, not a compensation penalty specific to exertion.
- Two framings of the same finding. Compensation is why exercise for weight loss disappoints (the energy deficit is partly refilled), yet it says nothing against exercise for fitness/mortality — the CRF and mortality benefits do not run through the energy balance that compensation offsets.
- A third outcome where compensation does NOT erase the benefit — visceral fat. Recchia 2023 (SR+MA, 40 RCTs) finds exercise reduces visceral fat more than caloric restriction per unit of prescribed weekly deficit, and names the same under-reporting/overcompensation confound as something that «might explain the lack of differences in the treatment effects between groups» in head-to-head trials (Recchia et al., 2023) (hedge preserved, corrected 2026-08-08). Because exercise’s fat-loss is partly weight-independent, the compensated calories blunt the weight benefit without erasing the visceral-depot benefit — so the disappointing weight lever and a comparatively strong visceral lever coexist -> Exercise vs Caloric Restriction for Visceral Fat.
Decision relevance
- Do not expect exercise (of any intensity) to fully “count” toward an energy deficit, especially long-term — plan weight change around intake, and use exercise for fitness/function/mortality where its benefit does not depend on the un-compensated calories.
- Do not down-rank HIIT for a compensation reason — intensity is not a compensation predictor. If HIIT is deprioritized, let it be for adherence or the small incremental-outcome argument, not NEAT.
- Higher body fat may mean exercise “counts” less toward the deficit — but the sources disagree (corrected 2026-08-08): Careau’s free-living gradient runs ~28% -> ~49% compensation across the BMI range, while Riou’s intervention interaction runs the opposite way short-term (higher initial FM -> lower compensation). So this is a directional caution grounded in one source’s setting, not a settled law. To the extent it holds, it is a reason the higher-adiposity person should anchor weight change on intake and use exercise for its fitness/function/mortality benefits, which do not run through the compensated calories.
(inferred from Careau et al., 2021)
Limits
- Two sources,
confidence: low. Riou’s ±93% SD means its mean is a weak summary of a highly individual response; Careau’s headline 28% and 27.7%->49.2% gradient are between-individual cross-sectional estimates, so they cannot prove the adiposity link is within-person (its stated caveat). Both rest on energy measurement (DLW is the good instrument; intervention body-composition is noisier). The convergence raises confidence in direction and rough scale, not in any single magnitude. - Careau’s one within-person analysis points to stronger compensation, but is thin (added 2026-08-08). A longitudinal subset (n = 68 elderly, 70-90 y, measured 7 yr apart) partitions the slope and finds «energy compensation occurs only at the within-individual level» (Careau et al., 2021) — bwithin = 0.15 vs bbetween = 1.86 — such that «the calories they burn during bouts of activity are almost entirely compensated for» (Careau et al., 2021). This is the setting an individual planning exercise actually faces, and it suggests near-total within-person compensation, above the 28% between-individual figure — but it is elderly-only, small, and the TEE-BEE within-individual covariance was itself non-significant (chi2 = 0.80, p = 0.37), so it sharpens the caution without settling a magnitude.
(inferred from Riou et al., 2015)
Self-critique [run 2026-07-29, before commit]
- Anti-flip-flop / Challenge #11: this does not reverse the challenge’s core (compensation is real); it refines it — upholding the general limb, refuting the intensity-specific limb on the source’s own predictor analysis. The CRF-page HIIT note stands; this adds the compensation evidence it lacked.
- Over-claim check: the ±93% SD and the measurement-error limit are stated so the 18%/84% figures are not read as precise; the intensity-null is quoted, not paraphrased.
- Scope guard: compensation is scoped to the energy-balance outcome, explicitly NOT extended to the fitness/mortality benefits that do not run through it.
[E-independent]verification (Careau added 2026-07-29; cold-audit-caught refinement): the routes are independent at the level of the estimate — different datasets (IAEA DLW cross-sectional vs exercise-intervention trials), different methods, neither deriving its number from the other. Cold audit flagged two tells the tag must not hide, now acknowledged in the body: Careau cites Riou 2015 (as the intervention-literature contrast it tests against, not as an antecedent it formalizes), and shares a university (Ottawa, different departments). Under the strict rule these do not break E, but they make it a qualified E — recorded, not laundered. Magnitudes (18% vs 28%) deliberately NOT pooled. Careau brings distinct new content (BEE mechanism, adiposity gradient), earning itssources:slot on the dual test.
Reading “18% ± 93%” - it is a standard deviation, not a range (deliverable-critique, 2026-08-01)
The «± 93%» is the dispersion (SD), not a 7-34% interval - and it is the actual finding. A mean of 18% with an SD ~93% (five times the mean) means compensation is so heterogeneous that the population mean barely predicts an individual: some people compensate >100% (eat back more than they burned), some near zero, some negatively. So “~one fifth on average” is right for the mean, but the decision-relevant message is the SPREAD - you cannot read your own compensation off the average.
The estimate also pools across energy-balance states and durations, which the source does not stratify: compensation rises with duration («approached 84%» at ~80 weeks), consistent with the intuition that a sustained deficit recruits a stronger compensatory (survival) response - so deficit-vs-maintenance-vs- surplus plausibly shifts it, and the pooled 18% hides that.
References
Careau, V., Halsey, L. G., Pontzer, H., Ainslie, P. N., Andersen, L. F., Anderson, L. J., Arab, L., Baddou, I., Bedu-Addo, K., Blaak, E. E., Blanc, S., Bonomi, A. G., Bouten, C. V. C., Buchowski, M. S., Butte, N. F., Camps, S. G. J. A., Close, G. L., Cooper, J. A., Das, S. K., … Speakman, J. R. (2021). Energy compensation and adiposity in humans. Current Biology, 31(20), 4659-4666.e2. https://doi.org/10.1016/j.cub.2021.08.016
Recchia, F., Leung, C. K., Yu, A. P., Leung, W., Yu, D. J., Fong, D. Y., Montero, D., Lee, C.-H., Wong, S. H. S., & Siu, P. M. (2023). Dose–response effects of exercise and caloric restriction on visceral adiposity in overweight and obese adults: a systematic review and meta-analysis of randomised controlled trials. British Journal of Sports Medicine, 57(16), 1035–1041. https://doi.org/10.1136/bjsports-2022-106304
Riou, M.-È., Jomphe-Tremblay, S., Lamothe, G., Stacey, D., Szczotka, A., & Doucet, É. (2015). Predictors of Energy Compensation during Exercise Interventions: A Systematic Review. Nutrients, 7(5), 3677–3704. https://doi.org/10.3390/nu7053677