This is a hot topic, and I don’t think there is a firm consensus in the child psychiatry world about the answer. I think a good summary of the literature would leave you with:
X, Y, and Z studies showed an association; Χ, Ψ, and Ω studies showed no association.
Excellent. Ιn lieu of that, I'm going to use one study to illustrate a couple methodological points to show you why this answer is difficult to know. My goal is to help the reader develop a skeptical eye when research of this type is presented.
One study used teacher questionnaires about hyperactivity symptoms and student questionnaires about dietary habits to see if a relationship existed between high sugar intake and hyperactivity. They found that children were at statistically higher risk for ADHD if they consumed:
- less sugar from fruit snacks
- low Vitamin C
However, the total simple sugar intake did not correlate with hyperactivity symptoms.
This makes very little sense. Why should fruit snacks be different from other sugar, and where did Vitamin C come from?
Statistical noise. Data interpretation relies on a ratio of signal:noise to find the signal (in this case, an association). A 5-second brainstorm will bring to mind dozens of unrelated (to usual dietary patterns) topics that may affect what students write down on questionnaires: literacy, gender, upbringing, vocabulary, organization, memory, day-of-the-week, recent holidays, who won the Red Sox game, etc. etc. etc. Some of these may reflect recent dietary patterns that skew the memory of usual dietary patterns; others simply affect the accuracy of report. One could generate an equally diverse list of factors that may affect teacher reports. Either way, the result is the same: a noisy signal.
Confounding. A confound exists when an extraneous factor exists that correlates with both the independent (dietary report) and the dependent (inattentiveness) variable. In this case, one could imagine that students who tend to be careless about reporting might also be those displaying symptoms of hyperactivity (which tends to run with inattentiveness). It’s not obvious whether that would lead to under or over reporting of sugar intake.
Generally, obvious confounders can be adjusted for statistically (here, I don't see that they did much of that, although they did do separate analyses for boys vs girls). However, no amount of math can adjust for factors that aren’t measured. There was no “carelessness independent of hyperactivity” score obtained here, and we couldn’t expect it — it’s a basically unknowable parameter. Certain confounders are inherent to the methodology. Math never fixes this.
Multiple comparisons. The basic principle is: if you look at enough factors, something is going to be statistically associated with something else for reasons that have nothing to do with reality. That’s because we define statistical significance as a result that would occur by chance <5% of the time. If you run 20 tests, you have a good chance of coming up with one of them that appears to be positive simply by chance. In data that has a low signal:noise ratio (see #1), this problem is all the more apparent. This is something to think about when unexpected results pop up from a study that was designed to look at something completely different. Vitamin C??!
Please note: this is not meant to be (primarily) a critique of this particular study. My goals here were to help the reader understand:
- Why do I find different answers to this question every time I do a google search?
- How should we think critically about epidemiological data presented on Health.SE?
Kim, Y. Chang, H. Correlation between attention deficit hyperactivity disorder and sugar consumption, quality of diet, and dietary behavior in school children. Nutr Res Pract. 2011 Jun;5(3):236-45.