The Verification Venue · a thing everyone gets wrong, reversed
The Sugar High That Was in the Parents
Double-blind trials went looking for the sugar high in the children and never found it. Pool sixteen controlled studies and sugar's effect on a child's behavior sits at zero. So where does the "sugar high" live? In 1994, an experiment gave 35 boys a placebo and lied to half their mothers: told them their sons had just been dosed with sugar. Those mothers rated their sons as significantly more hyperactive, and hovered and criticized more. The effect is real. It's just measurable in the parent, not the child.
Below you can set two things independently: how much sugar the child really got, and what the watching parent was told. Watch the two lines. The child's measured behavior stays flat across every dose. The parent's perceived rating jumps the instant the label says "sugar," and doesn't care what the real dose was. The gap between them is the sugar high, and you can see exactly where it lives.
Child's measured hyperactivity
d = 0.00
flat: no dose effect
Parent's perceived rating
d = 0.00
set by the label
The "sugar high" (perceived − measured)
d = 0.00
lives in the observer
0 to 50 g of pure sugar (a can of cola is about 39 g). Drag it anywhere. The green line does not move: that is the whole finding.
Flip this and the pink line jumps, even though the child's actual dose hasn't changed. In Hoover & Milich (1994) every child got a placebo; only the label the mother heard was randomized.
How many children the trial measured. This sets the pale band: the smallest real effect a trial this size could still detect. Bigger N shrinks the band toward zero, but never to zero. This is the honest whisker: "no effect found" means "none bigger than the band," not "exactly zero."
Two facts, from two different kinds of study, sit on top of each other here. The green line is the child's actual behavior, pinned flat at the pooled result of every double-blind sugar trial: Wolraich et al., JAMA 1995 meta-analyzed 16 reports (23 within-subject studies) across 14 behavioral and cognitive measures. Every one of the 14 mean effect sizes had a 95% confidence interval that included zero; the range of the means ran from −0.14 to +0.30. The pink line is the parent's rating, and it tracks the label because that is what Hoover & Milich (1994) found when they held the child's dose fixed at placebo and moved only the words the mother heard.
The check: every number recomputed in front of you
The band (the smallest effect a trial could still catch) and the label floor (how big the expectancy effect had to be to reach significance in Hoover & Milich) are computed live from the standard power arithmetic. Nothing here is asserted; it is calculated:
The band is verified two independent ways: the power formula MDE = (z.975 + z.80)/√N, and by inverting it: plugging that MDE back into a power calculation must return exactly 80%. Run it yourself offline: node research/does-sugar-make-kids-hyperactive/verify-does-sugar-make-kids-hyperactive.mjs.
Three studies, one verdict, in a table
| study | design | who / how many | what varied | result |
|---|---|---|---|---|
| Wolraich, Wilson & White JAMA 1995 |
meta-analysis of double-blind, placebo-controlled trials | 16 reports · 23 within-subject studies · 14 measures | real sugar vs. placebo, everyone blinded | no effect on behavior or cognition; all 14 CIs include 0 |
| Wolraich et al. NEJM 1994 |
double-blind, diet-controlled crossover | ~48 children (incl. "sugar-sensitive") | 3-week diets: high-sucrose / aspartame / saccharin | no difference across 39 behavioral & cognitive variables |
| Hoover & Milich J. Abn. Child Psych. 1994 |
randomized expectancy manipulation | 35 boys aged 5–7 & their mothers | the label only: every child got placebo | mothers told "sugar" rated sons significantly MORE hyperactive; hovered & criticized more |
The twist: the effect isn't absent. It moved.
It would be too easy to file this under "sugar does nothing." The honest, complete answer is stranger: the sugar high is real and measurable. It is simply relocated out of the child and into the observer. Hoover & Milich caught it in the mother's ratings and in her body: told "sugar," she stayed physically closer, watched harder, and drifted toward criticism. Nothing about the boy changed. The whole phenomenon was a fact about the watcher's expectation. That is why fifty bland explainers that stop at "it's a myth" leave the best part out.
And there are two guardrails the tidy version skips, both of which this instrument keeps honest:
Guardrail 1 · a bounded null, not a proof of zero
Wolraich's own paper says it plainly: "a small effect of sugar or effects on subsets of children cannot be ruled out." That is why the pale band never closes to a line. Drag the trial size and watch: even 1,000 children only shrink the smallest-detectable effect to about d = 0.09 (small, but not nothing). "No effect found" is a statement bounded by the study's power, not a mathematical zero. An effect inside the band, or one confined to a subset of children with ADHD, is exactly what these trials were unable to exclude.
Guardrail 2 · "no behavior effect" is not "sugar is fine"
This page is about one specific claim: behavior and attention. It is silent on metabolism, and the metabolic case against sugar is not silent at all: cavities, weight gain, and type-2 diabetes risk are real and well evidenced. The takeaway is narrow on purpose: sugar doesn't wind kids up. It says nothing about whether the third slice is a good idea.
What's exactly true here, and what's a modeled stand-in
Exactly true (sourced & recomputed). The band is the minimum detectable effect of a paired trial with N children at 80% power and α = 0.05 (two-sided): MDE = (z.975 + z.80)/√N = 2.8016/√N. At N = 48 that is d ≈ 0.40; the verifier confirms it two ways. The label floor (d ≈ 0.69) is the smallest expectancy effect that could have reached significance in Hoover & Milich's two-group design (n ≈ 17 and 18): t.975(df=33)·√(1/17+1/18). The Wolraich envelope (means −0.14 to +0.30, all 14 CIs spanning 0) is transcribed from the published abstract.
A modeled stand-in, clearly labelled. The pink line's height uses that 0.69 significance floor as its magnitude. Hoover & Milich's abstract reports the direction (mothers told "sugar" rated sons significantly more hyperactive) and the significance, but not an exact effect size; 0.69 is therefore the smallest expectancy effect consistent with their reported result at their sample size: a defensible floor, not their measured value, which was very likely larger. The green line is pinned at exactly 0 because that is the meta-analytic point estimate for behavior; the band around it carries the uncertainty.
Free choices named. Effect sizes are in standardized (Cohen's d) units. The MDE assumes a within-subject/paired design (the shape of the real sugar trials), 80% power, α = 0.05 two-sided, and the effect expressed in SD-of-differences units; change any of those and the band's width changes, though never its qualitative story. "Dose" is shown in grams of sugar; the trials dosed roughly 1.5–1.75 g/kg, well above a normal snack, and still found nothing.