Relations Between Multi-Informant Assessments of ADHD Symptoms,

DAT1, and DRD4

Ian R. Gizer, Irwin D. Waldman,

and Ann Abramowitz

Emory University

Cathy L. Barr

Toronto Western Research Institute

and Hospital for Sick Children

Yu Feng, Karen G. Wigg, and Virginia L. Misener

Toronto Western Research Institute

David C. Rowe

University of Arizona

Researchers conducting candidate gene studies of attention-deficit/hyperactivity disorder (ADHD) typ-
ically obtain symptom ratings from multiple informants (i.e., mothers, fathers, and teachers) and use a
psychologist’s best estimate or a simple algorithm, such as taking the highest symptom ratings across
informants, to construct diagnostic phenotypes for estimating association. Nonetheless, these methods
have never been empirically validated in the context of a molecular genetic study. In the current study,
the authors systematically evaluated several methods of operationalizing phenotypes and the resulting
evidence for association between ADHD and the candidate genes: dopamine transporter gene (DAT1) and
dopamine D4 receptor gene (DRD4). Use of symptom scores as continuous scales in regression analysis
suggested that the combination of mother and teacher ratings yielded the strongest evidence for
association between hyperactive–impulsive ADHD symptoms and DAT1 and between inattentive ADHD
symptoms and DRD4. Teacher ratings alone were sufficient for evaluating the association between
inattentive symptoms and DAT1. Further, this regression-based method consistently yielded stronger
evidence for association among ADHD symptoms, DAT1, and DRD4 than did three simple algorithms
(i.e., the and, or, and averaging rules). The implications of these results for future molecular genetic
studies of ADHD are discussed.

Keywords: ADHD, DAT1, DRD4, multiple informants

Attention-deficit/hyperactivity disorder (ADHD) is estimated to

occur in approximately 3%–7% of children, making it one of the
most prevalent childhood psychiatric disorders (American Psychi-
atric Association, 2000). As described in the Diagnostic and
Statistical Manual of Mental Disorders, 4th edition (DSM–IV;
American Psychiatric Association, 2000), ADHD consists of two
distinct but correlated inattentive and hyperactive–impulsive
symptom dimensions. Quantitative behavior genetic studies (i.e.,
twin and adoption studies) have suggested substantial genetic
influences are involved in the etiology of ADHD with heritabilities
ranging from 60%–90% (Waldman & Rhee, 2002). As a result,

researchers have begun to search for susceptibility genes for this
disorder with some success (see Faraone et al., 2005, and Wald-
man & Gizer, 2006 for reviews). Nonetheless, all initial reports
claiming a relation between a specific candidate gene and ADHD
have been followed by a mix of successful and failed replication
attempts. It is important to note that many of these studies have
identified such susceptibility genes by testing for an association
between specific candidate genes and ADHD defined as a discrete
diagnostic category. Such an approach is warranted, given that
twin studies suggest substantial genetic influences on ADHD
operationalized as a diagnostic category (Thapar, Holmes, Poulton,
& Harrington, 1999), but similar studies also suggest that the
continuous, inattentive, and hyperactive–impulsive symptom di-
mensions that underlie ADHD show similarly strong genetic in-
fluences (e.g., Sherman, Iacono, & McGue, 1997). Thus, research-
ers have begun to explore whether the susceptibility genes that have
been identified for the diagnosis of ADHD might be better concep-
tualized as quantitative trait loci (QTLs) that underlie the hyperactive–
impulsive and/or inattentive symptom dimensions (see Asherson &
Image Consortium, 2004 for a review). Given the statistical power
that is retained when one does not artificially dichotomize a contin-
uum, the QTL approach may provide a more powerful method for
identifying susceptibility genes (Cohen, 1983).

Although the QTL approach may present specific advantages

for examining the association of candidate genes with ADHD,
studies of this type present unique difficulties regarding the as-

Ian R. Gizer, Irwin D. Waldman, and Ann Abramowitz, Department of

Psychology, Emory University; Cathy L. Barr, Genetics and Development
Division, Toronto Western Research Institute, Toronto, Ontario, Canada,
and Neurosciences and Mental Health Program, Hospital for Sick Children,
Toronto, Ontario, Canada; Yu Feng, Karen G. Wigg, and Virginia L.
Misener, Genetics and Development Division, Toronto Western Research
Institute; David C. Rowe, Family and Consumer Resources, University of
Arizona.

Preparation of this article was supported in part by National Institute of

Mental Health Grant F31-MH072083-01 to Ian R. Gizer and National
Institute of Mental Health Grant K01-MH01818 to Irwin D. Waldman.

Correspondence concerning this article should be addressed to Ian R.

Gizer, Department of Psychology, Emory University, 532 N. Kilgo Circle,
Atlanta, GA 30322. E-mail: [e-mail]

Journal of Abnormal Psychology

Copyright 2008 by the American Psychological Association

2008, Vol. 117, No. 4, 869 – 880

0021-843X/08/$12.00

DOI: 10.1037/a0013297

869

sessment of hyperactive–impulsive and inattentive symptoms. The
diagnostic criteria for ADHD require that presenting symptoms be
a pervasive aspect of the child’s behavior and, therefore, observ-
able in at least two settings (e.g., home and school) (American
Psychiatric Association, 2000). Thus, researchers often assess
ADHD symptoms by collecting data from multiple informants—
most often the child’s parents and teachers—in order to establish
pervasiveness. Nonetheless, researchers have shown via meta-
analysis that the correlations between informants’ symptom ratings
are low (e.g., parent–teacher, r

⫽ .27; Achenbach, McConaughy,

& Howell, 1987). Studies conceptualizing ADHD as a discrete
diagnostic category often account for such discrepancies by use of
best-estimate procedures for combining assessment data from mul-
tiple informants in a manner that mirrors the decision process
clinicians use when determining clinical diagnoses. For studies
with a QTL approach, however, a similar method does not exist
and may not be preferable. As a result, specific methods have been
used in studies examining the association between a candidate
gene and a continuously distributed trait, such as taking the highest
score among informants (e.g., Rowe et al., 1998) or taking an
average score across informants (e.g., Mill et al., 2005), but these
methods have yet to be systematically evaluated.

Investigations of methods for use of multi-informant data in

molecular genetic studies are informed by two important lines of
research. First, many studies have demonstrated that parent and
teacher reports provide unique and valid diagnostic information
(see Achenbach, McConaughy, & Howell, 1987 and Meyer et al.,
2001 for reviews), and further studies have demonstrated the
incremental usefulness of having multiple informants to diagnose
ADHD (e.g., Power et al., 1998) as well as diagnosing ADHD
subtypes in research settings (e.g., Gadow et al., 2000; Weiler,
Bellinger, Marmor, Rancier, & Waber, 1999). Additional research
has suggested that combining assessment data from multiple in-
formants into a single clinical picture can obscure important clin-
ical data. For example, Gadow et al. (2004) demonstrated that
parent and teacher ratings of ADHD differ in terms of their
sensitivity to ADHD symptoms as well as their relations with
measures of social, academic, and cognitive functioning. This led
the authors to conclude researchers studying childhood psychiatric
disorders should evaluate informant-specific diagnoses as well as
diagnoses derived from the combination of informant reports.

Second, and of direct relevance to genetic association studies,

twin studies of ADHD suggest that parent and teacher ratings of
ADHD symptoms show genetic influences that are unique to each
informant, in addition to those that are common across informants
(Martin, Scourfield, & McGuffin, 2002; Nadder, Silberg, Rutter,
Maes, & Eaves, 2001). For example, one study reported that
although approximately 31% of the variability in parent and
teacher ratings of ADHD was due to a common genetic etiology,
as much as 40% of the variability in parent ratings and 50% of the
variability in teacher ratings were due to genetic influences unique
to that informant’s ratings (Martin, Scourfield, & McGuffin,
2002). This suggests that although parent and teacher ratings are to
some extent measuring a common trait influenced by a shared set
of genes, each informant’s ratings are also measuring distinct
aspects of ADHD symptomatology influenced by distinct genes.

Such findings have obvious implications for molecular genetic

studies of ADHD. If different informants’ symptom ratings reflect
specific as well as common genetic influences, studies that test for

association with phenotypes based on a single informant could
result in negative findings or in failures to replicate previous
findings if they do not have the same informant as the initial study.
Further, studies that combine data across informants without care-
ful consideration of the methods used might be losing valid diag-
nostic data, which could also lead to negative findings. Thus,
testing for association for each informant separately and in com-
bination should provide a clearer understanding of the relation
between a candidate gene and a disorder. In the current study,
ADHD symptom scale scores were used to evaluate the unique and
incremental effects of mother, father, and teacher ratings of
ADHD, as well as the interactions between informant ratings, in
their associations with two candidate genes that have shown rep-
licable evidence of association with ADHD, the dopamine trans-
porter gene (DAT1) and the dopamine D4 receptor gene (DRD4).

DAT1 was first suggested as a candidate gene for ADHD given

that stimulant medications prescribed to treat ADHD symptoms,
such as methylphenidate, appear to bind to the dopamine trans-
porter and block the reuptake of dopamine from the synapse
thereby increasing the amount of available dopamine (Ritz, Lamb,
Goldberg, & Kuhar, 1987; Volkow et al., 1995). Cook et al. (1995)
conducted the first study to test for association between DAT1 and
ADHD, and they reported evidence of association between a 40
base pair (bp) variable number of tandem repeats (VNTR) se-
quence in the 3

⬘ untranslated region (UTR) of DAT1 and ADHD.

More specifically, they found that the 10-repeat allele was prefer-
entially transmitted to children with ADHD. This relation has been
replicated several times (e.g., Curran et al., 2001; Gill, Daly,
Heron, Hawi, & Fitzgerald, 1997; Waldman et al., 1998), but there
have also been several failures to replicate (e.g., Holmes et al.,
2000; Todd et al., 2001). Nonetheless, a recent meta-analysis of
published and unpublished studies suggested a significant associ-
ation between ADHD and the 10-repeat allele of DAT1 (Faraone
and Kahn, 2006). These findings were extended by a study con-
ducted in our own lab, suggesting that the severity of hyperactive–
impulsive symptoms, but not inattentive symptoms, increased as
the number of 10-repeat alleles (i.e., 0, 1, or 2 alleles) increased
and, further, that DAT1 was associated with the combined ADHD
subtype but not the inattentive ADHD subtype (Waldman et al,
1998).

Psychiatric genetic studies have also suggested an association

between ADHD and DRD4. Initial interest in DRD4 was sparked
by association studies linking the gene to the personality trait of
novelty seeking (Benjamin et al., 1996; Ebstein et al., 1996).
LaHoste et al. (1996) conducted the first test for association
between ADHD and a 48-bp VNTR in exon 3 of DRD4. They
reported an association between the seven-repeat allele of this
polymorphism and ADHD. As with DAT1, this initial finding has
been replicated several times (e.g., Barr et al., 2000; Faraone et al.,
1999; Rowe et al., 1998), though there have also been several
failures to replicate (e.g., Castellanos et al., 1998; Hawi et al.,
2000). Faraone, Doyle, Mick, and Biederman (2001) conducted a
meta-analysis of these and additional studies and found significant
evidence suggesting that the seven-repeat allele of DRD4 is asso-
ciated with increased risk for ADHD, which has remained a
significant result even as the number of studies examining this
relation has continued to increase (Faraone et al., 2005). These
findings have also been extended by studies suggesting that DRD4
is more strongly associated with the inattentive than with the

870

GIZER ET AL.

hyperactive–impulsive ADHD symptoms in both clinical samples
(Lasky-Su et al., 2008; McCracken et al., 2000; Rowe et al., 1998)
and in the general population (Schmidt, Fox, Perez-Edgar, Hu, &
Hamer, 2001).

The purpose of the current study was to evaluate alternative

methods of using multi-informant data within the context of a
molecular genetic study, specifically to evaluate the extent to
which these methods provide evidence for an association between
ADHD, DAT1, and DRD4. Because the majority of studies that
examined the relation between ADHD and these two genes relied
on best-estimate or consensus diagnoses to classify their samples,
the current study provided an opportunity to evaluate different
methods for using multi-informant data and to investigate the
nature of the relations between these genes and ADHD.

Methods for using multi-informant data were evaluated in two

steps. First, informants’ ratings of ADHD symptoms were evalu-
ated with a regression-based optimal informant method (Bird,
Gould, & Staghezza, 1992). In this approach, each informant’s
symptom ratings are evaluated independently and in combination
within a regression framework that includes tests for main effects
and interactions. Entering these terms hierarchically into the re-
gression model allows for the estimation of the incremental con-
tribution of each informant’s symptom ratings and the interactions
between informants’ ratings. Thus, the optimal informant or set of
informants can be determined by testing whether each informant’s
ratings or interaction of ratings yields a significant increase in
explained variance when predicting the criterion variable (Bird et
al., 1992). In the context of a molecular genetic study, mother,
father, and teacher symptom ratings would be evaluated in terms of
their association with a specific genetic marker. Such analyses
were conducted as a first step to determine which informants’
inattentive and hyperactive–impulsive ADHD symptom ratings
yielded the strongest associations with DAT1 and DRD4.

In the second step of the analyses, the results obtained with the

regression-based method were contrasted against the evidence for
association yielded by three simpler and more frequently used
algorithms. The first algorithm, the or rule, designates a symptom
as present if any informant rates it as present, thus providing a
liberal estimate of diagnosis (Cohen, Velez, Kohn, Schwab-Stone,
& Johnson, 1987; Lahey et al., 2000). When applied to continuous
symptom scales, the or rule translates into use of the highest rating
of each symptom dimension given by any informant. The second
algorithm, the and rule, designates a symptom as present only if all
informants (e.g., the parent and the teacher) rate it as present
(Offord et al., 1996), which yields a conservative estimate of
diagnosis. With continuous symptom scales, taking the lowest
score given for each symptom dimension by each of the informants
translates into a similarly conservative symptom rating. The third
simple algorithm takes the average of the ratings from all available
informants. Rather than simply the highest or lowest rating, with
this method one considers each informant’s ratings, and therefore,
this method represents a compromise between the or and the and
algorithms.

Method

Participants

Ninety-nine children were recruited through the Center for

Learning and Attention Deficit Disorders (CLADD) at the Emory

University School of Medicine in Atlanta, Georgia, a clinic spe-
cializing in the assessment and treatment of childhood externaliz-
ing disorders such as ADHD, oppositional defiant disorder (ODD)
and conduct disorder (CD). All children included in the study were
probands referred to the clinic for an assessment related to exter-
nalizing behavior problems. Children diagnosed with autism, trau-
matic brain injury, or neurological conditions (e.g., epilepsy) were
excluded from the study, as were children with IQs

⬍ 75. Any

other diagnosis assigned to a child remained confidential and did
not influence their inclusion in the study, though a post hoc
analysis indicated that 77 of the 99 children met full criteria for
ADHD. More specifically, 37 children met criteria for the com-
bined subtype, 39 met criteria for the predominantly inattentive
subtype, and 1 met criteria for the predominantly hyperactive–
impulsive subtype. Further, 45 children met diagnostic criteria for
a diagnosis of ODD, and 10 children met diagnostic criteria for
CD. Diagnoses of internalizing disorders were not obtained for the
sample.

The participants represent a subset of a larger sample, from

which findings have been reported previously (e.g., Rowe et al.,
1998; Waldman et al., 1998), as well as new participants in our
ongoing study. It is important to note that the recruitment method
did not change with the addition of the new participants, and
further the participants did not significantly differ with respect to
ADHD symptom levels as rated by mothers, fathers, or teachers ( p
values ranged from .293 to .686). They also did not differ with
respect to age, gender, ethnic composition, or level of parental
education ( p values

⬎ .10). Participants ranged in age from 5 years

to 17 years (M

⫽ 10.1, SD ⫽ 2.96) at the time of assessment. The

sample consisted of 68 boys (68%) and 31 (32%) girls. The ethnic
composition of the sample was 83% Caucasian, 8% African Amer-
ican, 8% Hispanic, and 1% Asian. The level of parental education
was 2% with some high school, 10% who were high school
graduates, 10% with some college but without a degree, 20% with
a two-year college degree, 50% with a four-year college degree,
and 8% with an advanced-level degree.

Assessment Procedures

Mother, father, and teacher ratings were obtained whenever

possible for each child with the Emory Diagnostic Rating Scale
(EDRS; Waldman et al., 1998). For those cases in which reports
were obtained from multiple teachers, we used data from the
teacher who spent the most time per week with the child. If there
was no clear difference in the amount of time spent together, the
homeroom teacher was selected. Because participants were not
selected with respect to medication status, all informants were
asked to rate the child’s behavior as though the child were not
currently taking medication. Self-report data were not collected
from the children.

The EDRS is a symptom checklist developed to assess symp-

toms of the major DSM–IV childhood psychiatric disorders, in-
cluding disruptive disorders such as ADHD, ODD, and CD, and
internalizing disorders such as depression and anxiety disorders.
Parents and teachers rated symptoms on a 0 – 4 scale, with a score
of 0 indicating that the symptom is not at all characteristic of the
child and a score of 4 indicating that the symptom is very much
characteristic of the child. A previous study showed that diagnoses
derived by counting the number of symptoms present (i.e.,

ⱖ6

871

RELATIONS BETWEEN MULTI-INFORMANT ASSESSMENTS

inattentive symptoms or

ⱖ6 hyperactive–impulsive symptoms)

yielded diagnostic rates in a control population similar to the
prevalence rates described in the DSM–IV, suggesting the EDRS
provides valid assessments of ADHD symptoms (Waldman et al.,
1998). Internal consistencies of the hyperactive–impulsive scale,
which contains the nine DSM–IV ADHD hyperactive–impulsive
symptoms, and the inattentive symptom scale, which contains the
nine DSM–IV ADHD inattentive symptoms, were independently
evaluated for each informant to ensure that the scales possessed
acceptable reliabilities. These values ranged from

␣ ⫽ .83 to ␣ ⫽

.96. The scores for each symptom were then summed to create
hyperactive–impulsive and inattentive symptom dimensions (see
Table 1 for means and standard deviations displayed by genotype).

Genotyping

The DNA collection and extraction procedures have been de-

scribed in detail in previous publications (Rowe et al., 1998;
Waldman et al., 1998). For the 40-bp VNTR in the 3

⬘ UTR of

DAT1, samples were genotyped by polymerase chain reaction
(PCR), with primers described by Vandenbergh et al. (1992),
either according to the protocol described by Waldman et al.
(1998) or, more recently, according to an alternate protocol. For
the alternate protocol, PCR reactions were carried out in a 20 ul
volume containing 60 –100 ng DNA, 25 ng each primer, a set of
Invitrogen PCR

⫻ Enhancer System reagents consisting of 1⫻

amplification buffer, 1

⫻ enhancer solution, 1.5 mM MgSO4 (Ca-

nadian Life Technologies, Burlington, Ontario, Canada), 0.2 mM
dNTPs (0.05 mM each), and 0.5 units Taq polymerase, with
thermocycling conditions as follows: (a) initial denaturation (4 min
at 94°C), (b) 35 cycles of denaturation (40 sec at 94°C), annealing
(40 sec at 68°C), and extension (30 sec at 72°C), and (c) final
extension (10 min at 72°C). PCR products were resolved on 3%
agarose gels and visualized by ethidium bromide staining. For the
48-bp VNTR in exon 3 of DRD4, samples were genotyped ac-
cording to Lichter et al. (1993).

Analyses

To test for association and provide a direct test of the incremen-

tal contribution of each informant over and above every other
informant, ordinal logistic regression analyses were conducted on
each informant’s ratings uniquely and in combination with every
other informant’s ratings. The number of high-risk alleles (i.e.,
number of 10-repeat alleles for DAT1 and number of 7-repeat
alleles for DRD4) that a child possessed (i.e., 0, 1, or 2) served as
the dependent variables in these analyses. Mother, father, and
teacher ratings and interactions between these ratings served as
independent variables. Participants’ ethnicities were included as
covariates in these analyses to control for possible population
stratification biases. Such biases occur when ethnic groups differ
in both allele frequencies and rates (or symptom levels) of the
disorder, which can produce spurious evidence for an association
in the absence of any true relation between gene and disorder. Age
and gender were also evaluated as potential covariates. The overall
pattern of results were similar whether ethnicity, age, and gender
were included as covariates with only slight changes in the effect
sizes (data not shown). Thus, to simplify the presentation of
findings, the results without covariates included are reported be-

Table

1

Means

and

Standard

Deviations

of

Symptom

Scale

Ratings

for

Each

Dimension

by

Genotype

for

DAT1,

DRD4,

Associated

R

2

Values,

and

Number

of

Participants

No.

of

high-risk

alleles

DAT1

DRD4

Inattentive

sxs

Hyperactive–impulsive

sxs

Inattentive

sxs

Hyperactive–impulsive

sxs

Mother

Father

Teacher

Mother

Father

Teacher

Mother

Father

Teacher

Mother

Father

Teacher

MS

DMS

D

MS

DMS

D

MS

DMS

D

MS

DMS

D

MS

DMS

D

MS

DMS

D

0

2.17

0.97

2.45

0.75

2.07

0.73

1.76

1.14

1.88

0.95

1.96

0.92

2.44

0.96

2.29

0.89

2.12

1.08

1.57

1.13

1.44

1.07

1.57

1.32

1

2.54

0.99

2.32

1.05

2.00

1.04

1.58

1.05

1.48

1.07

1.36

0.22

2.49

0.97

2.45

1.02

2.74

0.90

1.90

1.20

1.65

1.16

2.06

1.44

2

2.52

0.94

2.45

0.75

2.65

0.97

1.88

1.30

1.58

1.15

2.05

1.46

1.86

0.84

2.41

0.05

2.60

0.22

1.44

1.24

1.89

0.31

1.19

0.98

R

2

.00

.00

.10

.02

.00

.04

.00

.01

.09

.01

.01

.02

N

93

69

93

93

69

93

88

67

90

87

67

89

Note.

DAT1

⫽

dopamine

transporter

gene;

DRD4

⫽

dopamine

D4

receptor

gene;

sxs

⫽

symptoms.

872

GIZER ET AL.

low. Additionally, because directional a priori predictions were
made regarding which allele would be associated with ADHD for
each candidate gene, one-tailed p values were reported for the
main effects. No strong predictions could be made in regard to the
nature of the interactions between informant ratings in predicting
genotype, and thus, two-tailed p values were reported for these
results.

As individual predictors are entered hierarchically into an ordi-

nal regression model, the Wald

␹

2

for the added predictor acts as

a test of the incremental contribution of this predictor over and
above previously entered predictors (Cohen, Cohen, West, &
Aiken, 2003). Thus, to evaluate the incremental contribution of
each rater and of each of the interaction terms, the significance of
the Wald

␹

2

and the associated change in the Nagelkerke R

2

were

evaluated for each step. This method determined whether the gain
achieved by adding each successive predictor significantly im-
proved the fit of the model or whether the results of the more
parsimonious model were equivalent and, therefore, preferred. It
should also be noted that the independent variables were mean-
centered prior to calculation of the interaction term, to control for
problems related to multicollinearity.

A key assumption underlying ordinal regression is that the slope

of the regression line is approximately equal across each level of
the dependent variable. To examine whether this assumption was
met, the test of parallel lines was conducted for each analysis.
When the assumption was violated, these results were reported,
and binary logistic regression was used as an alternative. For
DAT1, children with zero or one high-risk alleles were combined
into a single group and contrasted against children with two
high-risk alleles, given that very few children possessed zero
copies of the high-risk allele. For DRD4, children with one or two
high-risk alleles were combined into a single group and contrasted
against children with zero high-risk alleles given that very few
children possessed two copies of the high-risk allele.

In the current study, there were equal numbers of participating

mothers and teachers (93 mothers and 93 teachers), but far fewer
fathers (n

⫽ 67). Thus, evaluating the incremental contribution of

teacher ratings over and above mother ratings was accomplished
by excluding a few participants due to missing data. Estimating the
incremental contribution of father ratings was problematic, how-
ever, because almost one third of the sample (32 participants)
would be excluded due to missing father data. Therefore, initial
ordinal regression analyses were run with the full sample to

estimate the effects of mother and teacher ratings uniquely and in
combination, and a second set of analyses were conducted limiting
the sample to participants with complete data from mothers, fa-
thers, and teachers. The set of informants that yielded the best
fitting model and demonstrated incremental contributions over a
more parsimonious model was considered the optimal set of in-
formants. As a final step, the best fitting model identified by
regression analysis was evaluated against models with composite
ratings produced by each of the simple algorithms (i.e., the or, the
and, and the averaging rules) by comparing the effect sizes yielded
by each method.

Results

Correlations between informant’s ratings were conducted to

estimate the degree of overlap across informants (see Table 2).
Mother and father ratings were highly correlated for both inatten-
tive (r

⫽ .72) and hyperactive–impulsive (r ⫽ .72) symptoms. The

degree of overlap was considerably lower between parent and
teacher ratings for each dimension, ranging from r

⫽ .23 for

father and teacher ratings of hyperactive–impulsive symptoms
to r

⫽ .44 for mother and teacher ratings of hyperactive–

impulsive symptoms.

As described in the Method section, estimating the incremental

contribution of father ratings over and above mother or teacher
ratings was problematic because father ratings could be collected
from only two thirds of the sample. Further, analyses including and
excluding father ratings yielded highly similar results for each
candidate gene, with father ratings failing to provide any increase
in explained variance over and above mother and teacher ratings
(data not shown). Thus, only the results from analyses with mother
and teacher ratings are reported.

DAT1 and Hyperactive–Impulsive Symptoms

To test for an association between hyperactive–impulsive

ADHD symptoms and DAT1, mother and teacher ratings were
evaluated uniquely and in combination with ordinal regression (see
Table 3). Teacher ratings showed significant evidence of associa-
tion with DAT1, Wald

␹

2

(N

⫽ 93) ⫽ 3.38, p ⫽ .033, Nagelkerke

R

2

⫽ .045, whereas mother ratings did not, Wald ␹

2

(N

⫽ 93) ⫽

1.16, p

⫽ .140, Nagelkerke R

2

⫽ .015. Further, teacher ratings

continued to show incremental evidence for an association after

Table 2
Correlations Among Informants’ Ratings of Inattentive and Hyperactive–Impulsive ADHD Symptoms

Informant/symptom scale

Mother

inattentive sxs

Father

inattentive sxs

Teacher

inattentive sxs

Mother

hyperactive–

impulsive sxs

Father

hyperactive–

impulsive sxs

Correlation

n

Correlation

n

Correlation

n

Correlation

n

Correlation

n

Father inattentive sxs

.72

b

69

—

Teacher inattentive sxs

.25

b

99

.30

b

70

—

Mother hyperactive–impulsive sxs

.41

b

99

.25

a

69

.26

b

99

—

Father hyperactive–impulsive sxs

.41

b

68

.41

b

71

.24

a

69

.72

b

68

—

Teacher hyperactive–impulsive sxs

.04

99

⫺.06

80

.51

b

99

.44

b

99

.23

a

69

Note.

Bold

⫽ significant correlation; sxs ⫽ symptoms.

a

Correlation is significant at the .05 level (two tailed).

b

Correlation is significant at the .01 level (two tailed).

873

RELATIONS BETWEEN MULTI-INFORMANT ASSESSMENTS

controlling for the presence of mother ratings, Wald

␹

2

(N

⫽ 88) ⫽

4.88, p

⫽ .014, ⌬R

2

⫽ .062, as indicated by the significant

Wald

␹

2

test statistic. It was concluded that the best-fitting

model, however, included the interaction between mother and
teacher ratings. The improvement in fit over and above the main
effects model approached significance, Wald

␹

2

(N

⫽ 88) ⫽

3.07, p

⫽ .080, and resulted in a substantial increase in effect

size (

⌬R

2

⫽ .042).

A visual inspection of the described interaction indicated that

teacher ratings of hyperactive–impulsive symptoms showed an
increasing linear relation with the number of DAT1 high-risk
alleles as mother ratings of these symptoms increased. This is

indicated in Figure 1 by the increasing slopes of the regression
lines, which were graphed with mother ratings set at

⫺1 standard

deviation, at the mean, and at

⫹1 standard deviation. Because only

4 participants possessed zero copies of the DAT1 high-risk allele,
this analysis was rerun excluding those participants to ensure they
were not solely responsible for the interaction. The results re-
mained highly similar after excluding these participants: interac-
tion term, Wald

␹

2

(N

⫽ 84) ⫽ 2.51, p ⫽ .056.

This regression-based optimal informant model was then com-

pared with the simple algorithms. As shown in Table 3, each of the
simple algorithms yielded significant evidence for association be-
tween DAT1 and hyperactive–impulsive symptoms: or rule, Wald

Table 3
Association Analyses Between DAT1 and Mother and Teacher Ratings of Hyperactive-Impulsive Symptoms using Ordinal Regression

Model

Overall model

Individual variables

N

␹

2

df

p

R

2

OR

Wald

␹

2

df

p

R

2

⌬R

2

M only

93

1.208

1.165

1

.140

.015

T only

93

1.328

3.385

1

.033

.045

M & T

88

5.678

2

.058

.077

M

.912

.195

1

.670

.032

T

1.486

4.884

1

.014

.062

T & M & T

ⴱ

M

88

9.022

3

.029

.119

M

.584

2.683

1

.950

T

.994

.000

1

.508

M

ⴱ

T

1.310

3.073

1

.080

b

.042

Simple algorithms

and rule

a

98

1.479

3.624

1

.028

.051

or rule

98

1.358

3.429

1

.032

.042

average method

98

1.449

3.580

1

.029

.044

Note.

Boldface indicates significance at the .05 level. Italic represents a trend at the .10 level. The overall model p is two tailed. The individual variables

p is one tailed, except where noted. For univariate regression analyses, individual variables R

2

represents the effect size for that model. For multiple

regression analyses,

⌬R

2

represents the incremental contribution of the individual variable over and above the other variables in the model; OR

⫽ odds

ratio; M

⫽ mother; T ⫽ teacher. DAT1 ⫽ dopamine transporter gene.

a

Assumption of equal slope coefficients across categories was violated; statistics represent the results of logistic regression analysis.

b

Two tailed p.

Figure 1.

Graph of the interaction between mother and teacher ratings of hyperactive–impulsive (Hyper-Imp)

symptoms in predicting the number of dopamine transporter gene (DAT1) high-risk alleles in which the linear
relation between teacher ratings and DAT1 genotype is shown for three different values of mother ratings: the
mean,

⫹1 standard deviation, and ⫺1 standard deviation.

874

GIZER ET AL.

␹

2

(N

⫽ 98) ⫽ 3.43, p ⫽ .032, Nagelkerke R

2

⫽ .042; averaging

rule, Wald

␹

2

(N

⫽ 98) ⫽ 3.58, p ⫽ .029, Nagelkerke R

2

⫽ .044;

and rule, Wald

␹

2

(N

⫽ 98) ⫽ 3.62, p ⫽ .028, Nagelkerke R

2

⫽

.051). The test of parallel lines, which checks the assumption that
the slope coefficients are equivalent across the levels of the de-
pendent variable, was violated for the analysis of the and rule,
␹

2

(1, N

⫽ 98) ⫽ 4.67, p ⫽ .031, and as a result, this algorithm was

evaluated with a binary logistic regression in which participants
with zero or one high-risk alleles were combined into a single
group.

Despite the significant results produced by the simple algo-

rithms, the regression-based optimal informant model that in-
cluded mother and teacher ratings as well as their interaction
yielded an effect size (i.e., Nagelkerke R

2

) that was

⬃2 times as

large as that yielded by the and rule and

⬃3 times as large as that

yielded by the or and averaging rules. Thus, the optimal informant
method appeared to yield stronger evidence for association be-
tween hyperactive–impulsive symptoms and DAT1 than did the
simple algorithms.

DAT1 and Inattentive Symptoms

A similar set of analyses were conducted evaluating the relation

between DAT1 and mother and teacher ratings of children’s inat-
tentive symptoms (see Table 4). Teacher ratings of inattentive
symptoms showed significant evidence for association with DAT1
when analyzed alone, Wald

␹

2

(N

⫽ 93) ⫽ 7.26, p ⫽ .004,

Nagelkerke R

2

⫽ .100, but mother ratings did not, Wald ␹

2

(N

⫽

93)

⫽ .06, p ⫽ .400, Nagelkerke R

2

⫽ .001. Further, teacher

ratings continued to show incremental evidence for association
after controlling for the presence of mother ratings, Wald

␹

2

(N

⫽

88)

⫽ 8.78, p ⫽ .002, ⌬R

2

⫽ .131, and adding the interaction term

into the model failed to yield a statistically significant increase in
explained variance, Wald

␹

2

(N

⫽ 88) ⫽ 1.24, p ⫽ .264, ⌬R

2

⫽

.016. Thus, the model that included just teacher ratings of inatten-
tive ADHD symptoms was found to be the best fitting.

This optimal informant model was then compared with each of

the simple algorithms. As shown in Table 4, the optimal informant
model including only teacher ratings yielded an effect size
(Nagelkerke R

2

⫽ .100) that was ⬃2 –3 times as large as that of

any of the simple algorithms (Nagelkerke R

2

⫽ .032–.046). Thus,

the regression-based method appeared to yield stronger evidence
for association between inattentive symptoms and DAT1 than did
the simple algorithms.

DRD4 and Hyperactive–Impulsive Symptoms

The relations between mother and teacher ratings of ADHD

hyperactive–impulsive symptoms and DRD4 were evaluated
uniquely and in combination with ordinal regression (Table 5).
Neither mother ratings nor teacher ratings of hyperactive–
impulsive symptoms showed a relation with the number of DRD4
high-risk alleles, Wald

␹

2

(N

⫽ 87) ⫽ 0.87, p ⫽ .176, Nagelkerke

R

2

⫽ .013, and Wald ␹

2

(N

⫽ 89) ⫽ 1.46, p ⫽ .113, Nagelkerke

R

2

⫽ .023, respectively, nor did any of the additional models

tested. Among the simple algorithms, only the or rule showed a
trend suggesting an association between DRD4 and hyperactive–
impulsive symptoms, Wald

␹

2

(N

⫽ 92) ⫽ 2.65, p ⫽ .052,

Nagelkerke R

2

⫽ .043. It should be noted that because the test of

parallel lines neared significance,

␹

2

(1, N

⫽ 92) ⫽ 3.58, p ⫽ .059,

for the analysis of the or rule, this algorithm was evaluated a with
binary logistic regression in which participants with one or two
high-risk alleles were combined into a single group. Thus, with the
exception of the or rule, these results provide little evidence to
suggest an association between hyperactive–impulsive symptoms
and DRD4.

DRD4 and Inattentive Symptoms

Ordinal regression analyses were conducted to test the asso-

ciation between mother and teacher ratings of inattentive

Table 4
Association Analyses Between DAT1 and Mother and Teacher Ratings of Inattentive Symptoms Using Ordinal Regression

Model

Overall model

Individual variables

N

␹

2

df

p

R

2

OR

Wald

␹

2

df

p

R

2

⌬R

2

M only

93

1.055

.064

1

.400

.001

T only

93

1.761

7.263

1

.004

.100

M & T

88

10.021

2

.007

.132

M

.866

.367

1

.728

.032

T

1.948

8.779

1

.002

.131

T & M & T

ⴱ

M

88

11.309

3

.010

.148

M

.506

1.601

1

.897

T

1.077

.017

1

.448

M

ⴱ

T

1.274

1.245

1

.264

a

.016

Simple algorithms

and rule

98

1.502

3.562

1

.030

.046

or rule

98

1.508

2.610

1

.053

.032

average method

98

1.629

3.516

1

.030

.044

Note. Boldface indicates significance at the .05 level. Italic represents a trend at the .10 level. The overall model p is two tailed. The individual variables
p is one tailed, except where noted. For, univariate regression analyses, individual variables R

2

represents the effect size for that model. For multiple

regression analyses,

⌬R

2

represents the incremental contribution of the individual variable over and above the other variables in the model; OR

⫽ odds

ratio; M

⫽ mother; T ⫽ teacher; DAT1 ⫽ dopamine transporter gene.

a

Two-tailed p.

875

RELATIONS BETWEEN MULTI-INFORMANT ASSESSMENTS

ADHD symptoms and DRD4 (Table 6). Teacher ratings were
significantly associated with DRD4 when analyzed alone, Wald
␹

2

(N

⫽ 90) ⫽ 5.28, p ⫽ .007, Nagelkerke R

2

⫽ .088, whereas

mother ratings were not, Wald

␹

2

(N

⫽ 88) ⫽ 0.04, p ⫽ .578,

Nagelkerke R

2

⫽ .001. Further, teacher ratings continued to

show incremental evidence for association after controlling for
mother ratings, Wald

␹

2

(N

⫽ 85) ⫽ 4.93, p ⫽ .013, ⌬R

2

⫽

.086. It was concluded that the best-fitting model, however,
included the interaction between mother and teacher ratings.
The improvement in fit over and above the main effects model
was significant, Wald

␹

2

(N

⫽ 85) ⫽ 4.34, p ⫽ .036, and

resulted in a substantial increase in effect size (

⌬R

2

⫽ .082). A

visual inspection of this interaction (see Figure 2) suggested
that teachers’ ratings showed an increasing linear relation with
the number of DRD4 high-risk alleles as mother ratings of these
symptoms decreased. This is indicated by the decrease in slopes
of the regression lines, which were graphed with mother ratings
set at

⫺1 standard deviation, at the mean, and at ⫹1 standard

deviation. Because only 2 participants possessed two copies of
the DRD4 high-risk allele, this analysis was rerun excluding
those participants to ensure they were not solely responsible for
the significant interaction. The results remained highly similar
after excluding these participants: interaction term, Wald
␹

2

(N

⫽ 83) ⫽ 4.41, p ⫽ .036.

Table 5
Association Analyses Between DRD4 and Mother and Teacher Ratings of Hyperactive–Impulsive Symptoms Using Ordinal Regression

Overall model

Individual variables

Model

N

␹

2

df

p

R

2

OR

Wald

␹

2

df

p

R

2

⌬R

2

M only

87

1.210

.870

1

.176

.013

T only

89

1.241

1.463

1

.113

.023

M & T

84

2.239

2

.326

.035

M

1.169

.441

1

.254

.012

T

1.198

.850

1

.178

.022

T & M & T

ⴱ

M

84

2.869

3

.412

.045

M

1.481

1.067

1

.151

T

1.451

1.399

1

.118

M

ⴱ

T

.882

.606

1

.426

a

.010

Simple algorithms

and rule

92

1.455

.476

1

.245

.007

or rule

b

92

1.397

2.650

1

.052

.043

average method

92

1.311

1.327

1

.124

.019

Note.

Italic represents a trend at the .10 level. The overall model p is two tailed. The individual variables p is one tailed, except where noted. For univariate

regression analyses, individual variables R

2

represents the effect size for that model. For multiple regression analyses,

⌬R

2

represents the incremental

contribution of the individual variable over and above the other variables in the model; OR

⫽ odds ratio; M ⫽ mother; T ⫽ teacher. DRD4 ⫽ dopamine

D4 receptor gene.

a

Two tailed p.

b

Assumption of equal slope coefficient across categories was violated; Statistics represent the results of logistic regression analysis.

Table 6
Association Analyses Between DRD4 and Mother and Teacher Ratings of Inattentive Symptoms Using Ordinal Regression

Model

Overall model

Individual variables

N

␹

2

df

p

R

2

OR

Wald

␹

2

df

p

R

2

⌬R

2

M only

88

1.049

.039

1

.578

.001

T only

90

1.804

5.283

1

.007

.088

M & T

85

5.687

2

.058

.087

M

.953

.032

1

.571

.000

T

1.781

4.932

1

.013

.086

T & M & T

ⴱ

M

85

11.408

3

.010

.169

M

6.679

3.600

1

.008

T

12.718

5.931

1

.024

M

ⴱ

T

.478

4.339

1

.036

a

.082

Simple algorithms

and rule

93

1.300

1.071

1

.150

.016

or rule

b

93

1.624

2.265

1

.066

.038

average method

93

1.380

1.041

1

.154

.015

Note.

Boldface indicates significance at the .05 level. Italic represents a trend at the .10 level. The overall model p is two tailed. The individual variables

p is one tailed, except where noted. For univariate regression analyses, individual variables R

2

represents the effect size for that model. For multiple

regression analyses,

⌬R

2

represents the incremental contribution of the individual variable over and above the other variables in the model. OR

⫽ odds

ratio; M

⫽ mother; T ⫽ teacher. DRD4 ⫽ dopamine D4 receptor gene.

a

Two tailed p.

b

Assumption of equal slope coefficients across categories was violated; statistics represent the results of logistic regression analysis.

876

GIZER ET AL.

This best-fitting optimal informant model was then compared

with each of the simple algorithms (Table 6). Of these algorithms,
only the or rule yielded evidence for a trend toward association
between DRD4 and inattentive symptoms, Wald

␹

2

(N

⫽ 93) ⫽

2.26, p

⫽ .066, Nagelkerke R

2

⫽ .038. Because the test of parallel

lines was violated,

␹

2

(1, N

⫽ 93) ⫽ 4.99, p ⫽ .026, for the ordinal

regression analysis evaluating the or rule, this algorithm was
evaluated with binary logistic regression. Nonetheless, the optimal
informant model that included the interaction between mother and
teacher ratings yielded an effect size (Nagelkerke R

2

⫽ .169) that

was

⬃4 times as large as that of the or algorithm (Nagelkerke

R

2

⫽ .038). Thus, the optimal informant method appeared to yield

stronger evidence for association between inattentive symptoms
and DRD4 than the simple algorithms.

Discussion

Candidate gene studies of ADHD frequently use clinicians’ best

estimate diagnoses or simple algorithms, such as the or and and
rules, to combine data from multiple informants and construct
phenotypes for studying linkage and association, but these meth-
ods have never been empirically validated or contrasted within the
context of a molecular genetic study. In the current study, we
systematically evaluated several methods of operationalizing
ADHD phenotypes by examining their association with two can-
didate genes, DAT1 and DRD4. Analyzing symptom scores from
multiple informants provided two lines of evidence suggesting that
a regression-based optimal informant approach yields more infor-
mative phenotypes than do those produced by simple algorithms
such as the or, and, and averaging methods.

First, the regression-based optimal informant method allowed

for the empirical evaluation of the data provided by each rater. A
significant association between DAT1 and teacher ratings of
hyperactive–impulsive ADHD symptoms, which yielded an effect
size of R

2

⫽ .045, was strengthened when mother ratings and the

interaction between teacher and mother ratings were added to the

regression model, as indicated by a substantial increase in effect
size (R

2

⫽ .119). In contrast to this interaction model, the associ-

ation between DAT1 and inattentive ADHD symptoms was suffi-
ciently estimated by teacher ratings alone, yielding an effect size of
R

2

⫽ .100. For DRD4, a model including the interaction between

mother and teacher ratings provided the strongest evidence for
association with inattentive symptoms (Nagelkerke R

2

⫽ .169),

but no relation was detected between DRD4 and the hyperactive–
impulsive symptoms.

Taken together, these results serve to highlight the potential

complexities that underlie the relations between informant ratings.
For example, the interaction between mother and teacher ratings of
hyperactive–impulsive symptoms in predicting DAT1 genotype
suggested that the strength of the association between teacher
ratings and DAT1 increased as mother ratings increased. In con-
trast, an examination of the interaction between mother and teacher
ratings of inattentive symptoms in predicting DRD4 genotype
suggested that the strength of the association between teacher
ratings and DRD4 increased as mother ratings decreased. The
former finding provides an example of how informant ratings
might act synergistically to provide evidence for genetic associa-
tion, whereas the latter result might be indicative of key differ-
ences in how parents and teachers rate inattentive symptoms.
Support for such an interpretation comes from previous research
suggesting that teachers show greater sensitivity than do mothers
when assessing inattention in children (Gadow et al., 2004). Thus,
these results demonstrate the usefulness of a regression-based
approach to evaluate the relations between informant ratings when
testing for association between candidate genes and ADHD.

The second line of evidence suggesting that a regression-based

optimal informant approach yields more informative phenotypes
than do simple algorithms comes from the stronger and more
consistent evidence for association produced by the former ap-
proach. For example, the regression-based approach yielded stron-
ger evidence for association with DAT1 than did each of the simple

Figure 2.

Graph of the interaction between mother and teacher ratings of inattentive symptoms in predicting

the number of dopamine D4 receptor gene (DRD4) high-risk alleles, in which the linear relation between teacher
ratings and DRD4 genotype is shown for three different values of mother ratings: the mean,

⫹1 standard

deviation, and

⫺1 standard deviation.

877

RELATIONS BETWEEN MULTI-INFORMANT ASSESSMENTS

algorithms with effect sizes (i.e., Nagelkerke R

2

s) that were two-

to threefold larger. Further, this evidence was consistent with
previous research in most cases. More specifically, the regression-
based approach showed stronger evidence of association between
DRD4 and inattentive symptoms than did hyperactive–impulsive
symptoms, which agrees with previous reports (Lasky-Su et al.,
2007; McCracken et al., 2000; Rowe et al., 1998), whereas the
simple algorithms failed to yield significant evidence for such an
association. Further, the regression-based approach yielded strong
evidence suggesting an association between DAT1 and
hyperactive–impulsive symptoms as well as inattentive symptoms.
Though this latter finding contradicts some previous research from
our own lab suggesting that DAT1 is more highly associated with
hyperactive–impulsive symptoms than with inattentive symptoms
(Waldman et al., 1998), it is consistent with previous research
suggesting a relation between the 10-repeat allele of DAT1 and
increased risk for ADHD. Thus, it will be of interest to establish
whether this finding can be replicated in future studies.

These results have direct implications for how researchers as-

sess symptoms of childhood psychiatric disorders, such as ADHD,
for genetic analysis. For example, parent ratings did not show
significant evidence for association between ADHD and DAT1 or
DRD4 when analyzed alone, whereas teacher ratings yielded sig-
nificant evidence for an association between hyperactive–
impulsive and inattentive symptoms and DAT1 and between inat-
tentive symptoms and DRD4. This result clearly demonstrates that
teacher ratings, independent of parent ratings, can be highly infor-
mative for establishing association between candidate genes and
ADHD. Nonetheless, the negative results produced by parent
ratings were surprising given the demonstrated validity of both
parent and teacher ratings as a measure of a child’s externalizing
behaviors (e.g., Bank, Duncan, Patterson, & Reid, 1993; Loeber,
Gren, Lahey, & Stouthamer-Loeber, 1991; Verhulst, Koot, & Van
der Ende, 1994) and candidate gene studies that yielded evidence
for associations between ADHD, DAT1, and DRD4, with only
parent ratings as phenotypes, including studies from our own lab
(e.g., Rowe et al., 1998; Waldman et al., 1998). Though surprising,
these results indicate that relying on a single informant, such as the
mother, may yield a limited perspective on child psychopathology
that can reduce a study’s power to detect genetic association. Thus,
these findings demonstrate the importance of collecting data from
both parents and teachers to assess childhood symptoms of ADHD
and suggest that future studies should include a test for informant-
specific evidence of genetic association as well as evidence of
genetic association derived from the combination of informant
reports.

As a further illustration of this conclusion, one potential expla-

nation for the negative findings produced by parent ratings in the
current study could be due to the modest sample size. Effect sizes
for the relations among DAT1, DRD4, and ADHD have consis-
tently been small (explaining

⬃3%–5% of the variance; Faraone

and Kahn, 2006; Waldman & Gizer, 2006), and thus, failures to
replicate findings for association are common. This could account
for the negative findings produced by parent ratings in the present
study, and this emphasizes the need for methods that use pheno-
typic data in a way that maximizes a study’s ability to detect the
relation between a candidate gene and a trait or disorder.

The current study also suggests that future molecular genetic

studies should carefully consider how to use father ratings. First,

the high degree of overlap between mother and father ratings (r

⫽

.72) in the current study suggests that mother and father ratings
may provide largely redundant information for establishing an
association between candidate genes and ADHD. Second, the low
rate of father participation makes it difficult to conduct tests of
association with sufficient statistical power. As a result, future
studies with the regression-based optimal informant approach
should carefully evaluate whether father ratings provide sufficient
independent information when testing for a genetic association that
compensates for the reduction in statistical power resulting from
their inclusion.

Limitations and Conclusion

Finally, there are some potential limitations of the current study

that should be addressed. First, the study’s exploratory nature
represents an important limitation. To evaluate several methods for
constructing phenotypes, multiple statistical tests were conducted
and corrections were not used, given that the primary aim was to
systematically compare the extent to which each method provided
evidence for association among ADHD symptom ratings, DAT1,
and DRD4. Although, in the current study, we attempted to control
for chance findings by making predictions regarding the nature of
the expected associations on the basis of previous research, the
need for replication in independent samples clearly remains. Sec-
ond, the current study sample was clinic-referred, and thus, the
participants selected represent the upper ends of hyperactive–
impulsive and inattentive behavior. It would be interesting to test
whether the study results could be extended to the general popu-
lation, but given that a comparison sample was not available, this
remains an empirical question for future study. As a third caveat,
it should be noted that although the regression-based approach
proved useful in the current research context, it is difficult to
relate these findings to studies investigating the use of multiple
informants in the assessment of ADHD. Thus, the findings of
the present study are unlikely to exhibit usefulness in clinical
assessment.

These limitations aside, the current study demonstrates the im-

portance of carefully constructing phenotypes for molecular ge-
netic studies. Because the effect sizes for the relations among
DAT1, DRD4, and ADHD in prior research have been relatively
small (explaining

⬃3%–5% of the variance), failures to replicate

findings for association are common. Thus, it is critical that such
studies use phenotypic data in a way that maximizes a study’s
ability to detect the relation between a candidate gene and a trait or
disorder. The results of the current study suggest that regression-
based methods of combining multi-informant data may provide a
more powerful technique for using such data than do simple
algorithms such as the or, and, and averaging methods.

References

Achenbach, T., McConaughy, S., & Howell, C. (1987). Child/adolescent

behavioral and emotional problems: Implications of cross-informant
correlations for situational specificity. Psychological Bulletin, 101, 213–
232.

American Psychiatric Association. (2000). Diagnostic and statistical man-

ual of mental disorders (4th ed. text rev.). Washington, DC: American
Psychiatric Association.

Asherson, P., & Image Consortium. (2004). Attention-deficit/hyperactivity

878

GIZER ET AL.

disorder in the post-genomic era. European Child & Adolescent Psychi-
atry, 13(Suppl. 1), 50 –70.

Bank, L., Duncan, T., Patterson, G., & Reid, J. (1993). Parent and teacher

ratings in the assessment and prediction of antisocial and delinquent
behaviors. Journal of Personality, 61, 693–709.

Barr, C., Wigg, K., Bloom, S., Schachar, R., Tannock, R., Roberts, W., et

al. (2000). Further evidence from haplotype analysis for linkage of the
dopamine D4 receptor gene and attention-deficit/hyperactivity disorder.
American Journal of Medical Genetics, 96, 262–267.

Benjamin, J., Li, L., Patterson, C., Greenberg, B., Murphy, D., & Hamer,

D. (1996). Population and familial association between the D4 dopamine
receptor gene and measures of novelty seeking. Nature Genetics, 12,
81– 84.

Bird, H., Gould, M., & Staghezza, B. (1992). Aggregating data from

multiple informants in child psychiatry epidemiological research. Jour-
nal of the American Academy of Child and Adolescent Psychiatry, 31,
78 – 85.

Castellanos, F., Lau, E., Tayebi, N., Lee, P., Long, R., Giedd, J., et al.

(1998). Lack of an association between a dopamine-4 receptor polymor-
phism and attention-deficit/hyperactivity disorder: Genetic and brain
morphometric analyses. Molecular Psychiatry, 3, 431– 434.

Cohen, J. (1983). The cost of dichotomization. Applied Psychological

Measurement, 7, 249 –253.

Cohen, J., Cohen, P., West, S., & Aiken, L. (2003). Applied multiple

regression/correlation analysis for the behavioral sciences (3rd ed.).
Mahwah, NJ: Erlbaum.

Cohen, P., Velez, N., Kohn, M., Schwab-Stone, M., & Johnson, J. (1987).

Child psychiatric diagnosis by computer algorithm: Theoretical issues
and empirical tests. Journal of the American Academy of Child &
Adolescent Psychiatry, 26, 631– 638.

Cook, E., Jr., Stein, M., Krasowski, M., Cox, N., Olkon, D., Kieffer, J., et

al. (1995). Association of attention-deficit disorder and the dopamine
transporter gene. American Journal of Human Genetics, 56, 993–998.

Curran, S., Mill, J., Tahir, E., Kent, L., Richards, S., Gould, A., et al.

(2001). Association study of a dopamine transporter polymorphism and
attention-deficit/hyperactivity disorder in UK and Turkish samples. Mo-
lecular Psychiatry, 6, 425– 428.

Ebstein, R., Novick, O., Umansky, R., Priel, B., Osher, Y., Blaine, D., et

al. (1996). Dopamine D4 receptor (DRD4) exon III polymorphism
associated with the human personality trait of novelty seeking. Nature
Genetics, 12, 78 – 80.

Faraone, S., Biederman, J., Weiffenbach, B., Keith, T., Chu, M., Weaver,

A., et al. (1999). Dopamine D4 gene 7-repeat allele and attention deficit
hyperactivity disorder. American Journal of Psychiatry, 156, 768 –770.

Faraone, S., Doyle, A., Mick, E., & Biederman, J. (2001). Meta-analysis of

the association between the 7-repeat allele of the dopamine D(4) recep-
tor gene and attention deficit hyperactivity disorder. American Journal
of Psychiatry, 158, 1052–1057.

Faraone, S. V., & Khan, S. A. (2006). Candidate gene studies of attention-

deficit/hyperactivity disorder. Journal of Clinical Psychiatry, 67, 13–20.

Faraone, S., Perlis, R., Doyle, A., Smoller, J., Goralnick, J., Holmgren, M.,

et al. (2005). Molecular genetics of attention-deficit/hyperactivity dis-
order. Biological Psychiatry, 57, 1313–1323.

Gadow, K. D., Drabick, D. A. G., Loney, J., Sprafkin, J., Salisbury, H.,

Azizian, A., & Schwartz, J. (2004). Comparison of ADHD symptom
subtypes as source-specific syndromes. Journal of Child Psychology and
Psychiatry, 45, 1135–1149.

Gill, M., Daly, G., Heron, S., Hawi, Z., & Fitzgerald, M. (1997). Confir-

mation of association between attention-deficit/hyperactivity disorder
and a dopamine transporter polymorphism. Molecular Psychiatry, 2,
311–313.

Hawi, Z., McCarron, M., Kirley, A., Daly, G., Fitzgerald, M., Gill, M., et

al. (2000). No association of the dopamine DRD4 receptor (DRD4) gene
polymorphism with attention-deficit/hyperactivity disorder (ADHD) in

the Irish population. American Journal of Medical Genetics, 96, 268 –
272.

Holmes, J., Payton, A., Barrett, J., Hever, T., Fitzpatrick, H., Trumper, A.,

et al. (2000). A family based and case-control association study of the
dopamine D4 receptor gene and dopamine transporter gene in attention
deficit hyperactivity disorder. Molecular Psychiatry, 5, 523–530.

Lahey, B., Pelham, W., Loney, J., Trapani, C., Stein, M., Lee, S., et al.

(2000). Comparison of methods of using parent and teacher reports of
symptoms in the diagnosis of DSM–IV attention-deficit/hyperactivity
disorder. Unpublished manuscript.

LaHoste, G., Swanson, J., Wigal, S., Glabe, C., Wigal, T., King, N., et al.

(1996). Dopamine D4 receptor gene polymorphism is associated with
attention deficit hyperactivity disorder. Molecular Psychiatry, 1, 121–
124.

Lasky-Su, J., Lange, C., Biederman, J., Tsuang, M., Doyle, A. E., Smoller,

J. W., et al. (2008). Family based association analysis of statistically
derived quantitative traits for ADHD reveal an association in DRD4 with
inattentive symptoms in ADHD individuals. American Journal of Med-
ical Genetics, 147, 100 –106.

Lichter, J. B., Barr, C. L., Kennedy, J. L., Van Tol, H. H., Kidd, K. K.,

Livak, K. J., et al. (1993). A hypervariable segment in the human
dopamine receptor D4 (DRD4) gene. Human Molecular Genetics, 2,
767–773.

Loeber, R., Green, S., Lahey, B., & Stouthamer-Loeber, M. (1991). Dif-

ferences and similarities between children, mothers, and teachers as
informants on disruptive child behavior. Journal of Abnormal Child
Psychology, 19, 75–95.

Martin, N., Scourfield, J., & McGuffin, P. (2002). Observer effects and

heritability of childhood attention-deficit hyperactivity disorder symp-
toms. British Journal of Psychiatry, 180, 260 –265.

McCracken, J. T., Smalley, S. L., McGough, J. J., Crawford, L.,

Del’Homme, M., Cantor, R. M., et al. (2000). Evidence for linkage of a
tandem duplication polymorphism upstream of the dopamine D4 recep-
tor gene (DRD4) with attention deficit hyperactivity disorder (ADHD).
Molecular Psychiatry, 5, 531–536.

Meyer, G. J., Finn, S. E., Eyde, L. D., Kay, G. G., Moreland, K. L., Dies,

R. R., et al. (2001). Psychological testing and psychological assessment:
A review of evidence and issues. American Psychologist, 56, 128 –165.

Mill, J., Xu, X., Ronald, A., Curran, S., Price, T., Knight, J., et al. (2005).

Quantitative trait locus analysis of candidate gene alleles associated with
attention deficit hyperactivity disorder (ADHD) in five genes: DRD4,
DAT1, DRD5, SNAP-25, and 5HT1B. American Journal of Medical
Genetics, 133, 68 –73.

Nadder, T., Silberg, J., Rutter, M., Maes, H., & Eaves, L. (2001). Com-

parison of multiple measures of ADHD symptomatology: A multivariate
genetic analysis. Journal of Child Psychology & Psychiatry & Allied
Disciplines, 42, 475– 486.

Offord, D., Boyle, M., Racine, Y., Szatmari, P., Fleming, J., Sanford, M.,

et al. (1996). Integrating assessment data from multiple informants.
Journal of the American Academy of Child & Adolescent Psychiatry, 35,
1078 –1085.

Power, T. J., Anderews, T. J., Eiraldi, R. B., Doherty, B. J., Ikeda, M. J.,

DuPaul, G. J., & Landau, S. (1998). Evaluating attention-deficit/
hyperactivity disorder using multiple informants: The incremental utility
of combining teacher with parent reports. Psychological Assessment, 10,
250 –260.

Ritz, M., Lamb, R., Goldberg, S., & Kuhar, M. (1987, September 4).

Cocaine receptors on dopamine transporters are related to self-
administration of cocaine. Science, 237, 1219 –1223.

Rowe, D., Stever, C., Giedinghagen, L., Gard, J., Cleveland, H., Terris, S.,

et al. (1998). Dopamine DRD4 receptor polymorphism and attention
deficit hyperactivity disorder. Molecular Psychiatry, 3, 419 – 426.

Schmidt, L., Fox, N., Perez-Edgar, K., Hu, S., & Hamer, D. (2001).

879

RELATIONS BETWEEN MULTI-INFORMANT ASSESSMENTS

Association of DRD4 with attention problems in normal childhood
development. Psychiatric Genetics, 11, 25–29.

Sherman, D., Iacono, W., & McGue, M. (1997). Attention-deficit/

hyperactivity disorder dimensions: A twin study of inattention
impulsivity-hyperactivity. Journal of the American Academy of Child &
Adolescent Psychiatry, 36, 745–753.

Thapar, A., Holmes, J., Poulton, K., & Harrington, R. (1999). Genetic basis

of attention deficit and hyperactivity. British Journal of Psychiatry, 174,
105–111.

Todd, R. D., Jong, Y. J., Lobos, E. A., Reich, W., Heath, A. C., Neuman,

R. J., et al. (2001). No association of the dopamine transporter gene 3

⬘

VNTR polymorphism with ADHD subtypes in a population sample of
twins. American Journal of Medical Genetics, 105, 745–748.

Vandenbergh, D. J., Persico, A. M., Hawkins, A. L., Griffin, C. A., Li, X.,

Jabs, E. W., et al. (1992). Human dopamine transporter gene (DAT1) maps
to chromosome 5p15.3 and displays a VNTR. Genomics, 14, 1104 –1106.

Verhulst, F., Koot, H., & Van der Ende, J. (1994). Differential predictive

value of parents’ and teachers’ reports of children’s problem behaviors: A
longitudinal study. Journal of Abnormal Child Psychology, 22, 531–546.

Volkow, N., Ding, Y., Fowler, J., Wang, G., Logan, J., Gatley, J., et al.

(1995). Is methylphenidate like cocaine? Archives of General Psychia-
try, 52, 456 – 463.

Waldman, I., & Gizer, I. (2006). The genetics of attention-deficit/

hyperactivity disorder. Clinical Psychology Review, 26, 396 – 432.

Waldman, I., & Rhee, S. (2002). Behavioral and molecular genetic studies.

In S. Sandberg (Ed.), Hyperactivity and attention disorders of childhood
(2nd ed., pp. 290 –335). New York: Cambridge University Press.

Waldman, I., Rowe, D., Abramowitz, A., Kozel, S., Mohr, J., Sherman, S.,

et al. (1998). Association and linkage of the dopamine transporter gene
and attention-deficit hyperactivity disorder in children: Heterogeneity
owing to diagnostic subtype and severity. American Journal of Human
Genetics, 63, 1767–1776.

Weiler, M. D., Bellinger, D., Marmor, J., Rancier, S., & Waber, D. (1999).

Mother and teacher reports of ADHD symptoms: DSM–IV questionnaire
data. Journal of the American Academy of Child and Adolescent Psy-
chiatry, 38, 1139 –1147.

Received July 5, 2007

Revision received May 27, 2008

Accepted May 30, 2008

䡲

880

GIZER ET AL.