The Effect of Sugar on Behavior

or Cognition in Children
A Meta-analysis
Mark L. Wolraich, MD; David B. Wilson, PhD; J. Wade White, MD

Objective.\p=m-\Toexamine the effects of sugar on the behavior or cognition of behavioral extension to the vomiting re¬
children by using meta-analytic techniques on reported studies. action to milk proteins.5 Three more re¬
Data Sources.\p=m-\Studieswere identified through a literature search of the cent investigations have used this theory
MEDLINE and PsychINFO databases and the authors' files using sugar, sucrose, to study the effects of placing children
on restricted diets.68 These studies have
and attention deficit disorder as the search terms.
found beneficial effects from placing chil¬
Study Selection.\p=m-\Studieswere required to (1) intervene by having the subjects dren on restricted diets compared with
consume a known quantity of sugar; (2) use a placebo (artificial sweetener) con-
placebo diets,6 or they found worsening
dition; (3) blind the subjects, parents, and research staff to the conditions; and (4) behaviors in children on the restricted
report statistics that could be used to compute the dependent measures effect sizes. diets when they are challenged with of¬
Data Extraction.\p=m-\Variablesincluded publication year, study setting, subject fending foods compared with placebo
type and number, gender, age, sugar and placebo type and dose, prior dietary challenges.7·8 However, those dependent
condition, measurement construct, means and SDs for the sugar and placebo con- measures that found significant effects
were more subjective in nature (paren¬
ditions, and direction of effect.
Data Synthesis.\p=m-\Sixteenreports met the inclusion criteria for a total of 23 tal or clinician ratings of behavior). In
within-subject design studies. The weighted mean effect size and related statistics addition, sugars were not singled out
for individual evaluation.
for each of the 14 measurement constructs revealed that although the range for The second theory suggested that
these means was from -0.14 for direct observations and up to +0.30 for academic some children may experience functional
tests, the 95% confidence interval for all 14 mean effect sizes included 0. reactive hypoglycemia similar to that
Conclusion.\p=m-\The meta-analytic synthesis of the studies to date found that seen in adults.9 Adults with this condi¬
sugar does not affect the behavior or cognitive performance of children. The strong tion are reported to experience glucose
belief of parents may be due to expectancy and common association. However, a levels in the hypoglycémie range while
small effect of sugar or effects on subsets of children cannot be ruled out. on diets high in carbohydrates. Consum¬
(JAMA. 1995;274:1617-1621) ing diets high in proteins seems to pre¬
vent the effect. It was theorized that
children would display increased motor
REFINED sugar first became a nutri¬ tension-fatigue syndrome. Sugar later ap¬ activity at low blood glucose levels. This
tional concern in the United States after peared in the 1970s as a major offending theory became prominent after a study
the Civil War. At that time the main con¬ agent when the lay literature provided a of 271 hyperactive children found a large
cern was that sugar had lower nutritive good deal of coverage to the condition number of children who, during glucose
value.1 The idea that food including sugar called "functional reactive hypoglyce- tolerance tests, had patterns of blood
might have an adverse effect on behavior mia."1 The hypothesis in the relationship glucose similar to the pattern seen in
was first raised in 1922 by Shannon.2 This between sugar and behavior has since adults with functional reactive hypogly¬
concept was further elaborated in 1947 become so strong that it has been used in cemia.10 For children, subsequent study
by Randolph3 in his description of the court cases for a self-defense, referred to has shown that the patterns found by
as the "Twinkie defense," and reference Langseth and Dowd10 are actually nor¬
to it appears in cartoons and on television mal variations in children.11
From the Department of Pediatrics, Vanderbilt Uni- shows.4 Separate from any of the underlying
versity Child Development Center (Dr Wolraich), and The hypothesis in the relationship be¬ etiologies, the interest in sugar as a
the Institute for Public Policy Studies (Dr Wilson),
Vanderbilt University, Nashville, Tenn; and US Naval tween sugar and hyperactivity is based single offending agent peaked in the mid
Hospital, Yokosuka, Japan (Dr White). on two theories. The first, a possible 1970s because of several correlational
Reprint requests to Department of Pediatrics, allergic response to refined sugar, was studies. One early cross-sectional study
Vanderbilt University, Child Development Center, 2100
Pierce Ave, 426 MCS, Nashville, TN 37232-3573 (Dr conceived of in the first half of this cen¬ suggested an association between sugar
Wolraich). tury as the tension-fatigue syndrome, a intake and hyperactivity.12 It appeared

Downloaded From: http://jama.jamanetwork.com/ by a Northern Ontario School of Medicine User on 09/28/2013

 that hyperactive children who consumed
more sugar seemed to display more hy¬
peractive behavior. A similar but smaller
effect was found in a second study.13 Since
the major limitation of cross-sectional
studies is that causal relationships can¬
not be determined, it is equally possible
that the behavior caused the increased
sucrose intake as that the increased su¬
crose intake caused the behavior. There¬
fore, more definitive examination of the
issue has required research with inter¬
vention methods.
Most intervention research has en¬
tailed controlled double-blind challenge
studies. In these studies, the children
receive challenges with foods or drinks
containing either sugar or an artificial
sweetener. The children, their parents,
and the researchers are unaware of the
composition of the foods or drinks when
given, and the children's behavior and
cognitive performance are closely as¬
sessed within the few hours after the
ingestion. It would be helpful to deter¬
mine if the cumulative results of these
studies identify any effects of sugar on
the behavior or cognitive performance
of children.
METHODS
Literature Search
Studies were identified through the
authors' files and a literature search of
the MEDLINE and PsychlNFO data¬
bases using sugar, sucrose, and atten¬
tion deficit disorder as the search terms.
Studies were only included if they met
the following criteria: (1) they intervened
by having the subjects consume a known
quantity of sugar; (2) they used a pla¬
cebo (aspartame or saccharin) condition;
(3) their design had the subjects, par¬
ents, and research staff blinded to the
conditions; and (4) they reported means
and SDs or other statistics that could be
used to compute an effect size for the
dependent measures. This search pro¬
cess identified 16 articles1429 reporting
23 separate studies. All identified stud¬
ies used a within-subject design.
Coding
The variables coded reflected two lev¬
els of analysis: that of the study and the
dependent measure. Study level vari¬
ables included the following: publication
year; study setting (ie, laboratory, school,
hospital, other); subject type included in
sample (ie, normal, attention deficit hy-
peractivity disorder [ADHD], opposi-
tional defiant disorder [ODD]/aggres-
sive, adversely responsive to sugar based
on parent report, ADHD and ODD/ag-
gressive, or other); proportion of males
in the study sample; mean age and age
group of sample; sugar and placebo type;
Downloaded From: http://jama.jamanetwork.com/ by a Northern Ontario School of Medicine User on 09/28/2013
sugar and placebo doses; and prior di¬
etary condition (ie, none, overnight fast,
sugar restricted prior to challenge, post¬
prandial, or other). The dependent mea¬
sure variables included measurement
construct, number of subjects, means
and SDs for the sugar and placebo con¬
ditions, and direction of effect. The mea¬
surement constructs were groupings of
the dependent measures that measured
similar behavioral or cognitive charac¬
teristics.
Calculation of Effect Sizes
The unbiased effect size estimator (d)
was applied to the data for each depen¬
dent measure reported in the studies.
The statistic d is the mean difference
between the sugar condition and the pla¬
cebo condition divided by the pooled SD,
adjusted for the small sample bias.
Hedges and Olkin30 showed that the stan¬
dardized mean difference effect size is
upwardly biased for small samples and
derived an adjustment function to re¬
move this bias. The 23 studies yielded
532 effect sizes. The largest number of
effect sizes from a single study was
7828; the smallest number of effect sizes
was l.13
In meta-analysis it is important to
maintain statistical independence be¬
tween the effect sizes included in an
analysis.31 Effect sizes are considered
dependent or correlated if they are based
on the same subject sample. Indepen¬
dence among the effect sizes can be
achieved in several ways.32 This study
used a less conservative approach that
makes better use of the data by calcu¬
lating a single effect size for each theo¬
retically meaningful measurement con¬
struct. Multiple effect sizes from a single
study within a measurement construct
were averaged. The analyses were per¬
formed for each measurement construct
separately, thus maintaining statistical
independence. The 532 effect sizes were
categorized into one of 14 measurement
constructs. Three constructs were sub¬
jective ratings of observers (parents,
teachers, or experimenters). One was
direct counting of defined behaviors ob¬
served. Eight were measures of skills
and activities using standardized mea¬
sures (neuropsychological tests, actom-
eter, continuous performance test, paired
associate learning, figure matching,
drawing figures, motor skills, and aca¬
demic tests). Aggression was left as a
separate construct because it reflected
a different type of behavior, and mood
consisted of self-report measures com¬
pleted by the subjects about how they
felt. This reduced the number of effect
sizes to 91 across the 14 measurement
constructs for the 23 studies.
An additional potential source of de-
pendence in the data are the multiple
studies reported by a single researcher
or research group, regardless of whether
the studies were reported in a single
article or separately.31 However, because
the analyses performed in this meta-
analysis weight each effect size by the
inverse of its variance, which is primar¬
ily a function of sample size, a combined
effect size based on separate studies
within a report would receive greater
weight in the analyses than the corre¬
sponding individual effect sizes. There¬
fore, treating each subject sample as
independent has little effect on the over¬
all mean effect size estimates. Further¬
more, it is conventional in a meta-analy-
sis to treat effect sizes from different
subject samples as independent. To ex¬
plore the possibility of a dependency, a
sensitivity analysis was performed. In¬
cluding only a single average effect size
per author per analysis produced com¬
parable results to treating the 23 sub¬
ject samples as independent.
On some measures, a positive sugar-
placebo difference reflected that sugar
improved behavior in the sugar condi¬
tion (eg, increased performance on aca¬
demic tests) or it reflected worsened
behavior (eg, increased score on an ag¬
gression measure). Therefore, the di¬
rection of effect was standardized across
effect sizes so that a positive sign on the
effect size indicated a desirable effect of
sugar, and a negative sign indicated an
undesirable effect of sugar on behavior.
Thus, for the aggression measure, a pos¬
itive effect size indicated a decrease
in aggression (ie, a desirable effect of
sugar).
Analysis
A weighted mean d for each of the 14
measurement constructs was calculated,
weighting by the inverse variance of d
to reflect the sampling error inherent in
each effect size. Confidence intervals
(CIs) and the homogeneity statistic, Q,
were computed using the formulas pro¬
vided by Hedges and Olkin.30 The ho¬
mogeneity analysis tests the hypothesis
that the sample of effect sizes estimates
a single underlying population param¬
eter for each measurement construct.
Essentially, this meta-analysis can be
viewed as 14 different meta-analyses,
one on each measurement construct.
RESULTS
The 16 studies that met the inclusion
criteria were conducted over a period of
12 years from 1982 to 1994. All of the
studies used a within-subject design. All
studies but one used challenges with
sugar and placebo. The one study29 that
did not used provision of total diets to
the subjects for three 3-week intervals
 Table 1.—Characteristics of the Studies

No. of Age
Source, y Subjects* Subjects Range, y Male, % Preconditions Sugars Placebo
Gross,14 1984 ADHD 50 72 None Sucrose Saccharin
Beharetal,151984 Sugar reactors 21 6-14 100 Overnight fast Sucrose Saccharin
Beharetal,'51984 Sugar reactors 21 100 Overnight fast Glucose Saccharin
Kruesietal,,61987 Sugar reactors 18 2-7 100 None Sucrose Saccharin
Glucose Aspartame
Kruesl et al,16 1987 Normal 2-7 100 None Sucrose Aspartame
Glucose Saccharin
Rosen et al," 1988 Normal 45 5-8 58 With breakfast Sucrose Aspartame
Ferguson et al,'8 1986 Sugar reactors 5-13 None Sucrose Aspartame
Ferguson et al,'8 1986 Normal 18 3-5 50 None Sucrose Aspartame
Manan et al,'31988 Sugar reactors 3-10 80-100 None Sucrose Aspartame
Wolraich et al,20 1985 ADHD 16 7-12 100 Postprandial Sucrose Aspartame
Wolraich et al,20 1985 ADHD 16 8-12 100 Overnight fast Sucrose Aspartame
Milich and Pelham,2' 1986 ADHD 16 6-9 Overnight fast Sucrose Aspartame
Goldman et al,22 1986 Normal 3-7 50 None Sucrose Aspartame
Roshon and Hagen,231989 Normal 12 None Sucrose Aspartame
Bachorowskl et al,2" 1990 Delinquent 14-19 100 With breakfast Sucrose Aspartame
Bachorowski et al,24 1990 Normal 39 14-19 100 With breakfast Sucrose Aspartame
Wender and Solante,25 1991 ADHD/aggressive 5-7 With breakfast Sucrose Aspartame
Wender and Solante,25 1991 Normal 5-7 With breakfast Sucrose Aspartame
Conners,26 1983 Psychiatric ¡npatients 37 6-12 76 None Sucrose Aspartame
Sarvais et al,271990 Normal 20 9-10 50 With breakfast Sucrose Aspartame
Otto et al,28 1982 Prader-WIIII syndrome 10-30 100 With breakfast Sucrose Saccharin
Wolraich et al,29 1994 Sugar reactors 6-10 78 3-wk diet Sucrose Aspartame
Saccharin
Wolraich et al,281994 Normal 25 3-5 48 3-wk diet Sucrose Aspartame
Saccharin

*ADHD indicates attention deficit hyperactivity disorder, and ellipses, sex not given in article.

Table 2.—Weighted Mean Effect Size (d) by Measurement Construct and Related Statistics for 23 Within-Subject Sugar Challenge Studies
95% Confidence
Interval Range of d
No. of No. of I
Measurement Construct* Mean d Lower Upper Effect Sizest Subjects^ Lower Upper
Neuropsychological tests 0.04 -0.20 0.27 135 -0.02 0.24 0.50
Teacher ratings 0.03 0.25 161 -0.24
Experimenter ratings 0.05 -0.17 0.27 -0.01 0.08 0.05
Direct observation -0.14 0.05 13 229 -0.62 0.03 5.43
Parent ratings 0.03 -0.20 0.25 -0.09 0.10 0.27
Actometer 0.04 0.27 142 -0.50
Aggression -0.10 -0.38 0.17 105 -0.51 0.11 2.17
Continuous performance tests -0.04 -0.25 169 -1.20
Paired associate learning 0.02 -0.21 0.24 149 -0.19 0.16 0.62
Academic test 0.30 84 -0.24 2177§
Figure matching 0.04 -0.32 0.39 -0.08 0.16 0.27
Drawing figures 0.04 -0.45 0.53 32 0.02 0.06 0.00
Mood -0.02 -0.35 0.32 68 -0.18 0.21 0.85
Motor skills 0.20 -0.03 0.43 158 -0.42 2.43 33.76§
"The direction of the effect size for each construct is such that a positive value reflects a more desirable outcome on the measure and a negative value reflects a less desirable
outcome. Therefore, the negative effect size for aggression indicates slightly more aggression for the sugar condition,
tlf there were multiple effect sizes for a measurement construct within a study, the average of these effect sizes was used.
iThe total number of subjects across studies contributing to the mean d.
§P<.05; distribution heterogeneous.

with diet high in sucrose. In some

one
cases the reports used more than one
subject group so that there were 23 sepa¬
rate within-subject design studies. Sev¬
enteen of the studies had a laboratory
setting, three took place in a school set¬
ting, two in a psychiatric inpatient set¬
ting, and one in another setting. All of
the studies used sucrose challenges. In
addition, three studied glucose and two
studied fructose. Most (15) used doses
per kilogram of body weight ranging
from 1.25 to 5.6 g, while eight used fixed
doses from 35 to 78 g. Thirteen used
aspartame as the placebo, two used sac¬
charin, and six used both saccharin and
aspartame. The types of subjects, the
age distribution, and the preconditions
are presented in Table 1.
The weighted mean effect size (d) and
related statistics for each of the 14 mea¬
surement constructs are shown in Table 2.
Although the range for these means is
from -0.14 for direct observation to +0.30
for academic tests, the 95% CI for all 14
mean effect sizes includes 0, as can more
easily be seen in Figure 1. An examina¬
tion ofthe homogeneity statistic, Q, shows
that all but two ofthese distributions are
homogeneous; thus each set ofeffect sizes
shows no more variation than expected
from sampling error and their mean can
be presumed to estimate a single under-

Downloaded From: http://jama.jamanetwork.com/ by a Northern Ontario School of Medicine User on 09/28/2013


Figure 1.—Weighted mean effect sizes and 95% confidence intervals
Figure 2.—Effect sizes and 95% confidence intervals
lying population effect size. Choosing a
more liberal of less than .10 does not
change the findings.
For purposes of illustration, Figure 2
presents the effect size and associated
95% CIs for each study within three
measurement constructs. Shown in this
Figure are one homogeneous distribu¬
tion, direct observation, and the two het¬
erogeneous distributions, motor skills
and academic tests.
Examining the two heterogeneous dis¬
tributions more closely shows that both
include an outlier (an effect size of 1.86
for academic tests and 2.43 for motor
Downloaded From: http://jama.jamanetwork.com/ by a Northern Ontario School of Medicine User on 09/28/2013
(CIs) by measurement construct.
(CIs) by study for three measurement constructs.
skills—both of these effect sizes are
greater than 2 SDs from the mean). If
we assume that these outliers are the
occasional statistically rare observation
from a common underlying population
and recalculate the mean d without these
values, the two distributions become ho¬
mogeneous and the mean d is substan¬
tially reduced, becoming negative in one
case ( -0.05 and 0.01). Taking a different
line of reasoning and assuming that the
outliers were produced by real differ¬
ences between the studies suggest the
need to apply a random effects model,
rather than the fixed effects model as-
sumed thus far.38·34 Under the random
effects model, the width of the CIs in¬
creases substantially and the mean d
moves toward 0 (0.13 and 0.09). These
different assumptions lead to the same
conclusion: a null finding of the effect of
sugar on these two measurement con¬
structs.
The homogeneity of the effect size
distributions for each measurement con¬
struct indicates that the observed dif¬
ferences between the effect sizes within
each measurement construct cannot be
attributed to methodological or substan¬
tive differences between the studies. In
other words, the observed differences
in effect size magnitude across studies
does not exceed what would be expected
from chance variation. Unfortunately,
given the small number of studies, the
statistical power ofthe homogeneity test
is low. Therefore, several exploratory
analyses of breakouts of the larger ef¬
fect size distributions by substantive
variables, eg, subject type and age group,
were performed. These analyses failed
to identify any important modifiers of
the effect of sugar on these measure¬
ment constructs.
Of interest was the possibility that
sugar might have a larger effect on some
behaviors than others. The general in¬
ability to reject the null hypothesis for
any of the measurement constructs cau¬
tions against any such interpretation. Ad¬
ditionally, because each mean effect size
reported in Table 2 is based on a differ¬
ent subset of the 23 studies, any differ¬
ences in the mean effect sizes may also
reflect differences between these stud¬
ies other than the outcome measured.
A threat to the validity of the above
findings is the possibility that studies
showing stronger effects were somehow
missed by the search process. Concep¬
tually, this threat is the opposite of the
file-drawer problem, which asks how
many studies with null findings would
need to exist in "file drawers" to reduce
the observed mean effect size to a non¬
significant or nonmeaningful level.35,36
Here we need to ask how many studies
with nonnull findings would need to ex¬
ist to raise the observed mean to a sig¬
nificant or meaningful level. One ap¬
proach to this issue is to modify Orwin's
fail-safe formula35 to derive the num¬
ber of additional studies of a given mean
effect size (eg, 0.60) needed to raise the
observed mean to a specified level (eg,
0.40). This formula assumes that the ad¬
ditional studies have a mean sample size
equal to the mean sample size of the
observed studies. Applied to these data,
between three additional studies using
academic tests and 35 additional studies
using direct observation measures are
needed with a mean d of 0.60 to raise the
 overall mean value of d to 0.40 (these
would be negative values for the direct
observation studies). This represents a
60% to 200% increase in the number of
studies per measurement category. The
measurement category of academic tests
was the only category with less than a
100% increase in the number of studies
needed (these "reverse" fail-safe Ns are
presented in Table 2). The target popu¬
lation mean d of 0.40 was chosen as a
small but meaningful difference between
the sugar and placebo conditions. The
0.60 value was chosen as the magnitude
of the additional unobserved effect sizes
because it represents a moderate effect
size. Furthermore, less than 5% of the
532 effect sizes computed for this study
had an absolute value greater than 0.60,
indicating that an effect size of 0.60 is
rare in this population of studies. Al¬
though it is conceivable that several mod¬
erate-to-large effect size studies within
a given measurement construct have
been unintentionally omitted, it seems
unlikely.
COMMENT
This meta-analysis of the reported
studies to date found that sugar (mainly
sucrose) does not affect the behavior
References
1. Deutsch RM. The New Nuts Among the Berries.
Palo Alto, Calif; Bull Publishing Co; 1977:34, 235-238.
2. Shannon WR. Neuropathologic manifestations
in infants and children as a result of anaphylactic
reaction to foods contained in their dietary. AJDC.
1922;24:89-94.
3. Randolph TG. Allergy as a causative factor of
fatigue, irritability and behavior problems in chil-
dren. J Pediatr. 1947;31:560-572.
4. Wurtman JJ. Carbohydrate craving, mood
changes, and obesity. J Clin Psychiatry. 1988;49:
37-39.
5. Speer F. The allergic tension-fatigue syndrome.
Pediatr Clin North Am. 1954;1:1029-1037.
6. Egger J, Graham PJ, Carter CM, et al. Con-
trolled trial of oligoantigenic treatment in the hy-
perkinetic syndrome. Lancet. 1985;1:540-545.
7. Kaplan BJ, McNicol RD, Conte RA, et al. Di-
etary replacement in preschool aged hyperactive
boys. Pediatrics. 1989;83:7-17.
8. Boris M, Mandel FS. Foods and additives are
common causes of the attention deficit hyperactive
disorder in children. Ann Allergy. 1994;72:462-468.
9. Cott A. Treatment of learning disabilities. In:
William RJ, Kalita DK, eds. A Physician's Hand-
book on Orthomolecular Medicine. New York, NY:
Pergamon Press; 1977:90-91.
10. Langseth L, Dowd J. Glucose tolerance and
hyperkinesis. Fed J Cosmetic Toxicol. 1978;16:120\x=req-\
133.
11. Josefsberg Z, Vilunski E, Haukugh A, et al.
Glucose and insulin responses to an oral glucose
load in normal children and adolescents in Israel.
Isr J Med Sci. 1976;12:189-194.
12. Prinz RJ, Roberts WA, Hantman E. Dietary
correlates of hyperactive behavior in children.
J Consult Clin Psychol. 1980;48:760-769.
13. Wolraich ML, Stumbo P, Milich R, et al. Di-
etary characteristics of hyperactive and control boys
and their behavioral correlates. J Am Diet Assoc.
Downloaded From: http://jama.jamanetwork.com/ by a Northern Ontario School of Medicine User on 09/28/2013
or cognitive performance of children.
There is a high degree of homogeneity
and all the studies show effect sizes
whose CIs include 0, despite the fact
that there was a great deal of variation
in the types and ages of children stud¬
ied, dose, and preexisting diet condi¬
tions. However, the current number of
studies doesn't eliminate the possibility
of a small effect.
The question remains as to why the
results of controlled studies differ so
much from the impression of parents. A
possible explanation is expectancy. A
study by Hoover and Milich37 tested a
group of parents and children where the
parents believed that their children were
adversely affected by sugar. They chal¬
lenged all the children with a drink
sweetened with artificial sweetener (pla¬
cebo), but for half the parents they told
them the drink was sweetened with
sugar. They then asked the parents to
observe and rate their children's behav¬
ior, and the researchers observed and
recorded the parents' behaviors in in¬
teracting with their children. The effect
sizes (d) for the parent ratings and be¬
havioral coding of the parents ranged
from -0.57 to 1.11 with seven of the 11
measures greater than 0.42, which are
1986;86:500-504.
14. Gross M. Effect of sucrose on hyperkinetic chil-
dren. Pediatrics. 1984;74:876-878.
15. Behar D, Rapoport J, Adams C. Sugar chal-
lenge testing with children considered behaviorally
sugar reactive. Nutr Behav. 1984;1:277-288.
16. Kruesi MJ, Rapoport JL, Cummings M, et al.
Effects of sugar and aspartame on aggression and
activity in children. Am J Psychiatry. 1987;144:
1487-1490.
17. Rosen LE, Bender ME, Sorrell S, et al. Effects
of sugar (sucrose) on children's behavior. J Consult
Clin Psychol. 1988;56:583-589.
18. Ferguson HB, Stoddart C, Simeon J. Double
blind challenge studies of behavioral and cognitive
effects of sucrose-aspartame ingestion in normal
children. Nutr Rev. 1986;44:144-150.
19. Mahan LK, Chase M, Furukawa C, et al. Sugar
'allergy' and children's behavior. Ann Allergy. 1988;
61:453-458.
20. Wolraich ML, Milich R, Stumbo P, et al. Effects
of sucrose ingestion on the behavior of hyperactive
boys. J Pediatr. 1985;106:675-682.
21. Milich R, Pelham WE. Effects of sugar inges-
tion on the classroom playgroup behavior of atten-
tion deficit disordered boys. J Consult Clin Psy-
chol. 1986;54:714-718.
22. Goldman JA, Lerman RH, Contois JH, et al.
Behavioral effects of sucrose on preschool children.
J Abnorm Child Psychol. 1986;14:565-577.
23. Roshon MS, Hagen RL. Sugar consumption,
locomotion, task, orientation, and learning in pre-
school children. J Abnorm Child Psychol. 1989;17:
349-357.
24. Bachorowski J, Newman J, Nichols SL, et al.
Sucrose and delinquency: behavioral assessment.
Pediatrics. 1990;86:244-253.
25. Wender E, Solanto M. Effects of sugar on ag-
gressive and inattentive behavior in children with
attention deficit disorder with hyperactivity and
considerably higher than those found
for the sugar studies.
Parents' beliefs and expectancies can
have a significant effect on how they
perceive behavior and how they inter¬
act with their children. For example,
excited states in children are common
with birthday parties and holidays where
large sugar intake is likely. Since sugar
is commonly ingested and enjoyed by
children, variation in their behavior may
be mistakenly correlated with its con¬
sumption. In addition, all of the lay pub¬
licity suggesting an association between
sugar and behavior may prime parents
to expect the adverse behavior. This
combination could appear very convinc¬
ing to parents, thus perpetuating their
beliefs.
Although the current evidence gives
no support for the effect of sugar on
behavior, there are not currently enough
studies to reach a definitive conclusion.
With the small number of studies the
power to detect a small effect of sugar
is low. Furthermore, additional studies
would be needed to test for differential
effects of sugar on restricted subsets of
children.
We are grateful to Mark Lipsey, PhD, for his ad¬
vice and feedback with the study and manuscript.
normal children. Pediatrics. 1991;88:960-966.
26. Conners CK. The Effects of Sucrose, Fructose,
and Aspartame on Behavior, Cognitive Performance,
and Brain Function in Hospitalized Children: Fi-
nal Report to the Sugar Association. Washington,
DC: The Sugar Association; 1983.
27. Sarvais S, Schachar R, Zlotkin S, et al. Aspartame:
effects on learning, behavior and mood. Pediatrics.
1990;86:75-83.
28. Otto PL, Sulzbacher SI, Worthington-Roberts
BS. Sucrose-induced behavior changes of persons
with Prader-Willi syndrome. Am J Ment Defic.
1982;86:335-341.
29. Wolraich ML, Lindgren S, Stumbo P, et al.
Effects of diets high in sucrose or aspartame on the
behavioral and cognitive performance of children.
N Engl J Med. 1994;330:301-307.
30. Hedges LV, Olkin I. Statistical Methods for
Meta-analysis. Orlando, Fla: Academic Press; 1985.
31. Hedges LV. Directions for future methodol-
ogy. In: Wachter KW, Straf, ML, eds. The Future
of Meta-analysis. New York, NY: Russell Sage
Foundation; 1990.
32. Durlak JA, Lipsey MW. A practitioner's guide
to meta-analysis. Am J Community Psychol. 1991;
19:291-332.
33. Hedges LV. A random effect model for effect
sizes. Psychol Bull. 1983;93:388-395.
34. Shadish WR, Haddock CK. Combining estimates
of effect size. In: Cooper H, Hedges LV, eds. The
Handbook of Research Synthesis. New York, NY:
Russell Sage Foundation; 1994.
35. Orwin RG. A fail-safe N for effect size in meta\x=req-\
analysis. J Educ Stat. 1983;8:157-159.
36. Rosenthal R. The 'file drawer problem' and
tolerance for null results. Psychol Bull. 1979;85:
185-193.
37. Hoover DW, Milich R. Effects of sugar inges-
tion expectancies on mother-child interactions.
J Abnorm Child Psychol. 1994;22:501-515.