Excerpt from:

Food Additives and Environmental Chemicals
as Sources of Childhood Behavior Disorders

Bernard Weiss, PhD
Journal of the American Academy of Child Psychiatry, 1982

This paper was prepared under grants by the National Institute of Mental Health (NIMH), the National Institute of Environmental Health Sciences, and the U.S. Department of Energy at the University of Rochester Department of Radiation Biology and Biophysics.


The Feingold hypothesis postulates that many children who exhibit disturbed behavior improve on a diet devoid of certain food additives. Its validity has been examined on the basis of controlled trials. The total evidence, although not wholly consistent, nevertheless suggests that the hypothesis is, in principle, correct. Such a conclusion poses difficult problems and new issues for etiology, treatment, toxicology, and regulation.

A theory that seemed ludicrous a few years ago - that childhood hyperactivity (attention deficit disorder) might have dietary origins - has irrupted into the debate about the etiology and treatment of childhood behavior disorders

. . .

Although the literature is too sparse to permit a conclusive review, analysis of all the studies published to date indicates the presence of a real effect.

Total Diet Intervention

Harley et al. 1978a. The most extensive diet cross-over study was undertaken by a group at the University of Wisconsin. They provided all the food for each participating family to keep parents blind to dietary manipulations and to minimize violations. The experimental regimen excluded most, but not all, sources of artificial colors (sweet potatoes in particular), artificial flavors, and those foods designated by Feingold as sources of "natural salicylates." Subjects were boys who had been referred to the University of Wisconsin hospitals for "hyperactivity." ... A baseline period lasting 2 weeks was followed either by two 3-week or two 4-week observation periods. During one of these observation periods the children consumed the elimination diet; during the other they consumed a control diet designed to duplicate the conventional American diet. The diet order was counterbalanced so that half the subjects followed the sequence Feingold-control, and the other half the sequence control-Feingold. At the end of the baseline, and of each individual diet period, the older boys were tested by various psychological and other procedures. Student observers recorded behavior in the clasroom 3 times weekly. Once a week parents and teachers completed the 10-item Conners Parent-Teacher Questionnaire (Conners, 1970).

Although the authors concluded that their results contradict the Feingold hypothesis, closer examination discloses discrepancies between such an interpretation and the published findings. One enigma is the parent ratings. Thirteen of the 36 mothers, and 14 of the 30 fathers in the first tudy, recorded substantially improved behavior on the experimental compared to the control diet. Of the 36 teachers, 6 rated the children as less hyperactive on the diet. Four of the 36 children on the Feingold diet were consistently rated as improved by both parents and teachers. The concordance of mothers' and fathers' ratings was not presented in the published paper, but data from a preliminary yielded the entries in Figure 1. These suggest a greater degree of agreement among observers about dietary as opposed to control observations, which, in itself, suggests a more powerful effect from the diet. Although analyses of variance confirmed a significant diet effect for both the fathers' and mothers' ratings, the authors dismissed this finding because of a statistically significant effect of sequence. That is, 12 of the 13 subjects showing a positive response by mothers' ratings, and 11 of the 14 subjects showing a positive response by fathers' ratings, had been assigned the sequence of control diet first, experimental diet second. I replotted the data with the help of a pair of calipers and averaged mothers' ratings during baseline, experimental, and control diet, periods. The sequence control-experimental in figure 2 shows a sharp drop of about 4 points on the Conners scale, a change of nearly 23%. The drop from baseline to control was only about 10%. The sequence Experimental-control shows almost no change from baseline to control.