Journal of Child Psychology and Psychiatry 49:12 (2008), pp 1304–1312 doi:10.1111/j.1469-7610.2008.01942.

Construct validity of the five-factor

Strengths and Difficulties Questionnaire (SDQ)
in pre-, early, and late adolescence
Betty Van Roy,1,2* Marijke Veenstra,3 and Jocelyne Clench-Aas4
1
Institute of Psychiatry, University in Oslo, Norway; 2Akershus University Hospital, Division of Mental Health,
Norway; 3NOVA-Norwegian Social Research, Oslo, Norway; 4Norwegian Institute of Public Health, Oslo, Norway

Background: The Strengths and Difficulties Questionnaire (SDQ) is designed to measure psychological
adjustment in children and adolescents. Psychometric evaluations of the instrument have shown
satisfactory convergent and discriminant validity, while factor analysis studies have shown mixed
results across countries. In the present study, the construct validity of the five-factor SDQ is evaluated
in a large community sample of Norwegian pre-, early, and late adolescents. Methods: The sample
consisted of 26,269 children and adolescents (10–19 years) with valid answers on all 25 items of the
SDQ self-report. Complete parent/proxy data of respective pre-adolescent children was available for
6,645 cases. A Lisrel approach to Confirmatory Factor Analysis (CFA) was used to evaluate the five-
factor model and the presence of a positive construal factor. In the sample of pre-adolescents and their
parents/proxies, convergent and discriminant validity was evaluated by a CFA approach to multitrait-
multimethods (MTMM). Results: Fit statistics for the hypothesized five-factor model were satisfactory,
but introducing correlated error terms for some of the items led to significant model improvement in all
age groups. All factor loadings were higher than .30, except for item 11 (good friend). The loadings
differed across age groups and differed markedly between the parent/proxy and self-report measures.
The MTMM showed that the source of ratings made a difference on the validity of all subscale ratings,
with self-reports discriminating more on ratings of emotional and peer problems, and parents/proxies
discriminating more on hyperactivity symptoms. A positive construal factor was identified but had a
modest effect compared with the original five traits. Results suggested an unclear construct and
meaning of the Prosocial behaviour subscale. Conclusion: The results of the present study indicated
support for the proposed five-factor structure of the SDQ (Goodman, 2001) across a wide age range (10–19
years), including older adolescents and different informants. However, some improvements should be
considered to improve internal reliability and conceptual clarity. Keywords: SDQ, construct validity,
CFA, adolescents, psychometrics, questionnaires. Abbreviations: SDQ: Strengths and Difficulties
Questionnaire.

The Strengths and Difficulties Questionnaire (SDQ)
is designed to measure psychological adjustment in
children and adolescents (Goodman, 1997), and has
been established as a widely used instrument in
child mental health research (Vostanis, 2006). It is a
short questionnaire (25 items), easy to complete and
user friendly because of its briefness and positive
attributes items, and is sensitive to change. The five
subscales of the SDQ cover children’s emotional and
behavioural functioning (i.e., emotional symptoms,
conduct problems, hyperactivity–inattention, and
peer problems) as well as personal strengths (i.e.,
prosocial behaviour). The inclusion of both strength
and difficulty items in the SDQ supposedly makes
the questionnaire especially suitable for studies of
general populations where the majority of children
are healthy. The SDQ is available as both a self-
report and a proxy report questionnaire.
The instrument has been used in epidemiological,
developmental and clinical research in many coun-
tries and different cultures, increasing its utility for
cross-country comparison of children’s emotional
Conflict of interest statement: No conflicts declared.
Ó 2008 The Authors
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA
and behavioural difficulties (Goodman, Meltzer, &
Bailey, 2003; Marzocchi et al., 2004; Mathai,
Anderson, & Bourne, 2004; Obel et al., 2004; Wo-
erner et al., 2004). In Norway, the SDQ has been
used in several community screening programs to
study the overall well-being of large groups of young
people (Heiervang et al., 2007; Heyerdahl, 2003;
Ronning, Handegaard, Sourander, & Morch, 2004;
Van Roy, Groholt, Heyerdahl, & Clench-Aas, 2006).
One of these studies (Van Roy et al., 2006) suggested
that the SDQ could also be a valuable screening
instrument for older adolescents (16–19 years).
The usefulness of any measurement instrument
depends on its validity and reliability. Internal reli-
ability and test–retest stability of the SDQ have been
termed satisfactory, despite modest levels of internal
reliability for several subscales (Muris, Meesters, &
van den Berg, 2003; Palmieri & Smith, 2007). It has
been argued that low internal reliability may be due
to the positively worded reverse-scored items on the
conduct, peer problem and hyperactivity subscales
(Muris, Meesters, Eijkelenboom, & Vincken, 2004).
One of the most important properties of a meas-
urement tool is construct validity, which here refers

to the degree the questionnaire items are valid
measures of the five constructs of the SDQ. Con-
vergent and discriminant validity are main aspects of
construct validity. Because we cannot prove con-
struct validity, it is important to collect increasing
amounts of support that the measurements appear
to be sensible, that the postulated constructs behave
as anticipated, and that there are no grounds for
rejecting them (Fayers & Machin, 2007, p. 83).
Convergent validity of the SDQ has been evaluated
by showing substantial correlations with other
instruments of psychological adjustment, such as
the Rutter questionnaire, upon which it was
originally based, the Child Behaviour Check List and
the Youth Self-Report (Achenbach, 1991; Becker,
Woerner, Hasselhorn, Banaschewski, & Rothenber-
ger, 2004; Elander & Rutter, 1996; Goodman, 2001;
Goodman et al., 2003; Goodman & Scott, 1999;
Smedje, Broman, Hetta, & von Knorring, 1999).
Further, the SDQ discriminates well between
children with and without psychopathological
symptoms (Goodman, 2001).
Factor analysis studies have shown somewhat
mixed results across countries. In general, five-fac-
tor solutions that correspond with the hypothesised
domains of psychopathology and personal strengths
are indicated (Becker et al., 2006; Goodman, 2001;
Muris et al., 2003; Smedje et al., 1999). However,
results from a study from the United States (Dickey
& Blumberg, 2004) suggested a three-factor solution
corresponding to internalization, externalization and
prosocial problems. This was also found in Finnish
youth self-report data (Koskelainen, Sourander, &
Vauras, 2001). The presence of both positively and
negatively worded items can confound the factor
structure. Positively worded items tend to cluster
together in what can be termed a ‘positive construal
method factor’, irrespective of the trait they sup-
posedly measure. Several studies have pointed to the
presence of such a ‘positive construal method’ factor
in the SDQ, with substantial loadings for positively
worded items that belong to different sub-
scales (Dickey & Blumberg, 2004; Goodman, 2001;
Koskelainen et al., 2001; Palmieri & Smith, 2007;
Ronning et al., 2004).
The self-report version of the SDQ was originally
developed for children and adolescents between 11
and 16 years old. A previous study found support for
the validity and reliability of the self-report version of
the SDQ in 8–13-year-olds (Muris et al., 2004). A
Europe-wide study of children with ADHD, including
older adolescents (6–18 years), confirmed the valid-
ity and reliability of the parent-reported SDQ scale
structure for the sample as a whole (Becker et al.,
2004). Separate psychometric properties of the SDQ
self-report version in older adolescents (16–19 yrs)
have, to our knowledge, not been assessed
previously.
The overall objective of the present study is to
evaluate the construct validity of the five-factor SDQ
Ó 2008 The Authors
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
Construct validity of the SDQ 1305
as proposed by Goodman (2001) in a large com-
munity sample of Norwegian pre-, early, and late
adolescents. This involves the following steps: i)
analysis of the latent structure of the SDQ in
pre-adolescents, early, and late adolescents, and
parent/proxies; ii) evaluation of convergent and
discriminant validity in the sample of pre- adoles-
cents and their parent/proxies; and iii) assessment
of the presence of a ‘positive construal factor’ in
addition to the five hypothesized factors of the SDQ.
Methods
Sample
Pupils from the 5th grade in primary school to the last
year of high school (29,631 pupils, mean age: 14.7,
range: 10–19 years; boys: 50.9%, girls: 49.1%,
response rate: 84.3%) were invited to participate in
a health profile study, including SDQ, undertaken
in Akershus County in 2002 by the Norwegian Health
Services Research Centre (Van Roy et al., 2006). In
grades 5–7, parents and children completed a health
profile questionnaire (response rate = 78%). All 22
municipalities in the county, which surrounds the
Norwegian capital, participated. Classes at each school
level were selected at random to obtain a sample rep-
resentative of the county as a whole. Participation in the
study was voluntary. The parents were informed by the
local school and asked to give their consent. The chil-
dren completed the questionnaire at school during
regular classes, while the parents received the ques-
tionnaire at home via the child and returned materials
in a sealed envelope. The questionnaires of each parent
and their child had the same registration number, so
that they could be matched without violating the an-
onymity of the participants.
The SDQ scoring algorithms for subscale scores and
total difficulties scores allow scale scores to be prorated
if at least three of five scale items are complete. For
analytical purposes, in the present article, only
respondents with valid answers on all 25 items of the
SDQ (N = 26,269) were included in the analyses. Miss-
ing item responses varied between .7% and 2.3% per
item, with the highest percentage of missing responses
among late adolescents. The final number of respond-
ents in pre-adolescence (primary school, age range:
10–13 years), early adolescence (secondary school, age
range: 13–16 years) and late adolescence (high school,
age range: 16–19 years) was 8,320; 8,582 and 9,367,
respectively. Complete parent/proxy data of respective
pre-adolescent children was available for 6,645 cases.
These 6,645 cases were used for further analyses of the
parent/proxy sample and for the multitrait-multi-
methods (MTMM) analyses.
Measures
The Norwegian version of the SDQ self-report, obtained
by translation and back-translation, was used in the
study.
The SDQ (25 items) was designed by Goodman on the
basis of nosological concepts of the ICD-10 (WHO,
1994) and the DSM IV (APA, 1994). The questionnaire
1306 Betty Van Roy, Marijke Veenstra, and Jocelyne Clench-Aas
consists of five subscales of five items covering emo-
tional problems, conduct problems, hyperactivity
problems, peer problems and prosocial behaviour. Each
item uses a three-point ordinal Likert format and can be
answered with: ‘not true’, ‘somewhat true’ or ‘certainly
true’. Responses can be rated 0–2 for negatively worded
items and rated inversely 2–0 for positively worded
items. In this way, for all items, higher scores indicate
more problematic attributes. Subscores were generated
for each subscale (range: 0–10). All subscores except
the prosocial score were added up to a total difficulties
score (range: 0–40). The prosocial subscale measures
the adolescent’s ability to act prosocially, independent
of the difficulties measured by the other subscales.
An extended version of the SDQ includes a brief
‘impact supplement’ that assesses chronicity, distress,
social impairment and burden to others. In this study,
the impact supplement was not included in the psycho-
metric analyses.
Statistical analysis
We used a structural equation approach (SEM) to
Confirmatory Factor Analysis (CFA; Bollen, 1989) to
evaluate the latent structure of the hypothesized five-
factor model of the SDQ. Unlike traditional Exploratory
Factor Analysis (EFA), a SEM approach to CFA takes
measurement error into account and allows for explicit
modelling of the five hypothesized constructs. As such,
it proposes a specific test of whether the items of the
SDQ load on to the five hypothesised latent constructs
(factors) exactly as predicted. The analyses comprised
of several steps: i) Multigroup CFA to evaluate the five-
factor model and to test the equality of factor loadings
across samples of pre-adolescents, early adolescents,
late adolescents and parents/proxies, ii) CFA approach
to MTMM analysis to evaluate convergent and
discriminant validity; and iii) CFA to evaluate the
presence of a positive construal factor over and above
the hypothesized five trait factors.
To conduct CFA, we used LISREL 8.70 software,
which is suited for modelling ordinal, non-normally
distributed data. Based on the large sample size and the
ordinal nature of the data, we applied diagonally
weighted least squares (DWLS) estimation methods with
polychoric correlation and asymptotic covariance
matrices (Jöreskog, 2005). We used the following criteria
to evaluate the model fit (Bollen, 1989; Loehlin, 1998):
average residual correlations as measured by the root
mean square error of approximation (RMSEA), where an
RMSEA below .08 indicates an acceptable model fit, and
an RMSEA below .05 indicates good model fit; a Good-
ness-of-Fit Index (GFI) and Comparative Fit index (CFI)
above .90 and the root-mean-square-residual (SRMR)
below .10. The five-factor model (model 1), the MTMM
model (model 2) and the Positive Construal Factor model
(model 3) are presented in Figure 1.
Model 1 indicated the five-factor model consisting of
five correlated constructs with each construct reflecting
five items. Modification indexes (MI) were used to find
potential sources of significant model improvement.
Owing to the large sample size, a relatively small MI will
suggest a significant improvement of model fit. There-
fore, we only used the largest MI as indicators of model
improvement. The equivalence of constrained and
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
unconstrained factor models for the different grades
was evaluated with a chi-square difference test, using a
critical value (a) of 5%.
Model 2 indicated the CFA approach to MTMM
(Campbell & Fiske, 1959; Loehlin, 1998; Marsh &
Hocevar, 1983). This approach was used to attempt to
separate out the true variance on the five SDQ constructs
from variance resulting from measurement methods
(self-report versus parent/proxy). The logic of MTMM
designs is that proxy and self-report measures of the
same construct should be highly correlated, but that
measures of different constructs have low correlations.
In the hypothesized model, each subscale was measured
by both a method factor (parent/proxy or self-report) and
a trait-factor. By calculating squared factor loadings,
explained variance in subscales resulting from the
underlying trait and the reporting method was estim-
ated. The part left unexplained is termed uniqueness.
Trait factor correlations and method factor correlations
were estimated but the correlation between traits and
methods was constrained to zero.
Model 3 represented the positive construal factor
model. This model indicated that the positive construal
factor would contribute to an increase in explained
variance in the positively worded items that did not
belong to the Prosocial construct. Because all the items in
the Prosocial construct are positively phrased, this
construct already meets at least some of the criteria of a
positive construal factor. We let this sixth construct
correlate with the other five constructs of the SDQ. This
model posits that positively worded items might reflect
method variance rather than conceptually distinct
dimensions.
Cross-scale correlations were derived from the CFA
approach in LISREL and, as such, are error free. SPSS
14 was used for descriptive purposes and to assess the
internal reliability (Cronbach’s alpha) of the five sub-
scales.
Results
Table 1 presents the results from the CFA in all three
age groups and for the parent/proxy sample (model 1).
Fit statistics for the hypothesized five-factor model
were satisfactory. However, the MI suggested that
introducing correlated error terms for items 2
(Restless) and 10 (Fidgety) within the Hyperactivity
Problems construct, as well as for items 18 (Lies) and
19 (Bullied), would lead to significant model
improvement in all three age groups. This resulted in
RMSEA indices that indicated good fit for all age
groups. For the parent/proxy sample, only the cor-
related error terms for items 2 and 10 were sub-
stantial and the final model fit was good. Error-free
intercorrelations between the five constructs were
strongest between Conduct Problems and Hyperac-
tivity in all samples (.66–.84), followed by Peer
Problems and Emotional Problems (.62–.72). The
size of these correlations suggests considerable
conceptual overlap between these traits. Lowest
correlations were found between the constructs of
Prosocial and Emotional Problems (|.01–.16|).
Ó 2008 The Authors
 Construct validity of the SDQ 1307

Figure 1 Three hypothesized CFA models of the Strengths and Difficulties Questionnaire; five-factor model (model 1);
CFA approach to MTMM (model 2); A positive construal model (model 3). Error terms are not shown. Model 1. Five-
factor model SDQ. Item numbers as shown in table 2. Model 2. CFA approach to multitrait-multimethod analysis of
the five subscales (P = Parent/Proxy; S = Self-report). Model 3. Positive Construal Factor (PCF) model (Correlations
between PCF and SDQ traits not shown)

Ó 2008 The Authors

Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
1308 Betty Van Roy, Marijke Veenstra, and Jocelyne Clench-Aas

Table 1 Confirmatory factor analysis of the SDQ, model fit statistics in pre-adolescents (PA), early adolescents (EA), late adolescents
(LA) and parent/proxy (PP)

PA EA LA PP

N 8,320 8,582 9,367 6,645

Model 1: Five-factor model unadjusted
Satorra Bentler scaled v2(DF) 4,226 (265) 7,132 (265) 7,933 (265) 3,470 (265)
RMSEA .042 .055 .056 .043
GFI .96 .95 .95 .97
CFI .96 .96 .96 .98
SRMR .07 .08 .08 .07
Model 2: Five-factor model modified#
Satorra Bentler scaled v2(DF) 3,406 (263) 5,868 (263) 5,997 (263) 2,395 (264)
RMSEA .038 .049 .048 .035
GFI .97 .96 .96 .98
CFI .97 .97 .97 .99
SRMR .06 .07 .08 .07
Model 3: Positive construal factor model
Satorra Bentler scaled v2(DF) 1,957 (253) 3,790 (253) 4,581 (253) 1,872 (254)
RMSEA .029 .040 .043 .031
GFI .99 .98 .97 .99
CFI .98 .98 .98 .99
SRMR .05 .06 .07 .05
#
Correlated errors items 2 and 10; and 18 and 19. Correlated error items 2 and 10.

The standardized factor loadings corresponding to
these modified models are presented in Table 2. All
loadings were higher than .30, except for item 11
(Good friend) in the parent/proxy sample, which had
a loading of .29. The results in Table 2 suggested
that the factor loadings for the self-report groups
were relatively similar. However, chi-squared differ-
ence tests comparing the self-reports simultaneously
(v2dif = 1,101.46, df = 54, p < .0001), and pairwise
comparisons, indicated that factor loadings were not
equal across groups.
Cronbach’s alpha for the total difficulties score
was .74 in the pre-adolescent group, .78 in early
adolescence, .80 in late adolescence and .78 in the
parent/proxy sample. In all groups, Cronbach’s
alpha was lowest for the construct of Conduct
Problems, ranging from .44 to .59 (Table 2).
Hyperactivity had high internal consistency in the
parent/proxy sample (.76) but was, at best, modest
in the self-report samples.
Our next step was to evaluate the convergent
and divergent validity of the SDQ in the sample of
parent/proxy and preadolescents (model 2). The
CFA approach to MTMM with two method factors
and five trait factors had an acceptable model fit:
v2 = 288, df = 14; RMSEA = .056; CFI = .99;
AFI = .99 and SRMR = .02. Table 3 presents a
summary of the variance components attributed by
trait, method and uniqueness. The source of rat-
ings made a difference in the validity of all sub-
scale ratings. The relatively high trait variance
components suggested that self-report tend to be
more discriminating on ratings of Emotional
Symptoms and Peer Problems. Similarly, parent/
proxies seemed to be particularly discriminating on
ratings for Hyperactivity, where they were
practically not subject to any method effects. The
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
construct of Prosocial behaviour was characterised
by a relatively large uniqueness that underlined
the problems with interpreting the meaning of this
scale. We found modest correlations between the
method factors (.25).
Finally, we evaluated the presence of a positive
construal factor (model 3). The fit statistics of this
six-factor model are presented for the self-report
groups as well as the parent/proxy sample in the
lower part of Table 1. Introduction of a positive
construal factor contributed to significantly
improved model fit in all samples, as indicated by
the decrease in chi-square and the improved
RMSEA statistics. This factor was highly correlated
with the Prosocial construct (.63–.70). However, a
closer inspection of the variance components
attributable to the postulated traits indicated a
relatively modest role for the positive construal
factor (Table 4). For the self-report groups, factor
loadings for all five positively phrased items were
higher than .30. Yet, only item 21 (Reflective) and
item 7 (Obedient) were, to a larger extent, deter-
mined by the Positive Construal Factor rather than
their original trait (Hyperactivity/Conduct). The
relatively large amount of uniqueness related to
these two items indicated that their meaning may
be unclear in self-report groups. In particular, item
7 (‘Generally obedient, usually does what adults
request’) seemed to be less appropriate for late
adolescents as a measure of Conduct Problems.
For parent/proxies, item 11 (Good friend) was not
related to the Positive Construal Factor. The
exceptionally large uniqueness related to this item
(.91) emphasized the lack of clarity in its inter-
pretation. On the whole, ratings of parent/proxies
on the positively phrased items were little affected
by the Positive Construal Factor.
Ó 2008 The Authors
 Construct validity of the SDQ 1309

Table 2 Cronbach’s alpha and standardized factor loadings Table 4 Trait and method variance components in the Positive
for SDQ items (CFA) in pre-adolescents (PA), early adolescents Construal Factor CFA model in pre-adolescents (PA), early
(EA), late adolescents (LA) and parent/proxy (PP) adolescents (EA), late adolescents (LA) and parent/proxy (PP)

PA EA LA PP Item number

N 8,320 8,582 9,367 6,645 7 11 14 21 25

Emotional symptoms
Cronbach’s a .61 .71 .70 0,67 Trait PA .18 .54 .39 .17 .43
Somatic (3) .47 .60 .60 .51 EA .11 .65 .30 .13 .42
Worries (8) .57 .63 .64 .77 LA .06 .58 .31 .12 .44
Unhappy (13) .74 .84 .81 .81 PP .22 .08 .67 .35 .62
Clingy (16) .57 .57 .55 .69 Positive Construal Factor PA .19 .12 .23 .21 .11
Fears (24) .58 .74 .76 .67 EA .19 .22 .16 .25 .09
Conduct problems LA .17 .14 .11 .2 .09
Cronbach’s a .44 .59 .54 .5 PP .13 .01 .15 .16 .02
Tempers (5) .50 .52 .55 .68 Unique PA .63 .34 .38 .62 .46
Obedient (7) .44 .46 .37 .52 EA .70 .13 .54 .62 .49
Fights (12) .54 .71 .69 .73 LA .77 .28 .58 .68 .47
Lies (18) .58 .69 .70 .72 PP .65 .91 .18 .49 .36
Steals (22) .41 .66 .64 .54
Hyperactivity Items: 7 = obedient; 11 = good friend; 14 = popular;
Cronbach’s a .57 .65 .66 .76 21 = reflective; 25 = persistent.
Restless (2) .36 .46 .43 .62
Fidgety (10) .34 .41 .43 .59
Distractible (15) .70 .75 .80 .86
predicted five-factor structure in late adolescence
Thinks before .43 .49 .46 .64 (16–19 years). It also holds in a sample of parent/
acting (21) proxy raters.
Persistent (25) .64 .70 .73 .80 Although all models were acceptable without any
Peer problems modifications, several of our findings call for closer
Cronbach’s a .54 .64 .60 .56
inspection.
Solitary (6) .57 .55 .53 .68
Good friend (11) .67 .84 .79 .29 First, large and significant correlated error terms
Popular (14) .54 .59 .59 .80 for two pairs of items were found for all age groups.
Bullied (19) .69 .81 .86 .74 The correlated error terms between items 2 (Restless)
Adults (23) .40 .49 .44 .56 and 10 (Fidgety) correspond to a previous Norwegian
Prosocial behaviour
study. Ronning et al. (2004) argued that there is a
Cronbach’s a .64 .65 .64 .66
Considerate (1) .69 .78 .78 .87 close semantic similarity between these two items in
Shares (4) .47 .53 .56 .58 the Norwegian language. These items can possibly be
Caring (9) .71 .64 .67 .66 reformulated in order to obtain more distinct indi-
Kind to kids (17) .61 .65 .59 .68 cators; another possibility is to take out one of the
Often volunteers .58 .43 .42 .58
to help (20)
two items.
Second, we could not conclude that the factor
SDQ item numbers in parentheses. loadings were equal across age groups. This may be
partly the result of the very large sample, where quite
small differences from ideal measurement can lead
Table 3 CFA approach to multitrait-multimethod (MTM); Trait
and Method variance components (N = 6,645) to large chi-squares. Examination of the factor
loadings indicated that these were relatively similar
Measured variables Trait Method Uniqueness for early and late adolescents and generally higher
Self-report Emotional symptoms .41 .14 .45 than in the pre-adolescent group. One explanation
Conduct problems .19 .38 .42 may be that younger children experience problems in
Hyperactivity .10 .38 .51 a less differentiated way than older adolescents,
Peer problems .62 .07 .31 according to their cognitive capability. Factor load-
Prosocial behaviour .18 .08 .73
ings from the parent/proxy measures were markedly
Parent/ Emotional symptoms .19 .41 .40
proxy Conduct problems .25 .34 .41 different from self-reports, indicating that, for these
Hyperactivity .83 .16 .01 raters, items were weighing differently on their
Peer problems .18 .22 .60 underlying construct. The positively phrased item 11
Prosocial behaviour .32 .10 .57 (Good friend) had a surprisingly low factor loading in
the parent/proxy sample, and was not related to a
Positive Construal Factor. This calls for an inspec-
Discussion
tion of the Norwegian translation of this particular
The results of the present study indicate support for item in the proxy questionnaire.
the proposed five-factor structure of the SDQ Third, there was considerable conceptual overlap
(Goodman, 2001) in three age groups in a Norwegian between some subscales, which suggests a lack of
community sample, inclusive late adolescents. This discriminatory validity. High correlations between
is the first study showing support for Goodman’s Hyperactivity and Conduct Problems indicated that
Ó 2008 The Authors
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
1310 Betty Van Roy, Marijke Veenstra, and Jocelyne Clench-Aas
these scales at least partly share the same underly-
ing trait. The same phenomenon was found for the
Peer Problems and Emotional subscales. These
findings are consistent with the three-factor solu-
tions found by Dickey and Blumberg (2004) and
Koskelainen et al. (2001). Although we did not find
clear evidence for a three-factor solution, external-
izing and internalizing dimensions are likely to be
responsible for overlap between these particular
scales. On the other hand, our results supported the
findings of Goodman (2001), who concluded that
scales from the internalizing dimension are not
substantially correlated with scales from the exter-
nalizing dimension.
Fourth, relatively low correlations between Proso-
cial Behaviour and the other four constructs of the
SDQ seem to support Goodman’s claim (2003,
p. 174) that ‘the absence of prosocial behaviour is
conceptually different from the presence of psycho-
logical difficulties’. However, the Prosocial scale also
suffered from low agreement between parent/proxy
and self-rating, the presence of a method effect, and
was highly correlated with a positive construal fac-
tor. Our findings suggested difficulties with inter-
preting the meaning of the Prosocial Behaviour
subscale and indicated, consistent with the findings
of Palmieri and Smith (2007) that this subscale
might reflect a methodological artefact.
Fifth, similar to the recent findings of Palmieri
and Smith (2007), the results of the present study
indicated the presence of a positive construal factor,
in particular for self-reports. Ratings from parent/
proxies were least affected by this factor. Positively
worded items in the SDQ tend to cluster together
irrespective of the subscale they belong to. On the
one hand, this is a well-known detrimental method-
ological phenomenon. Detrimental, as it reduces the
clarity of the factor structure and may have con-
tributed to decreased internal reliability in some of
the subscales. On the other hand, the inclusion of
positively worded items is theoretically as well as
clinically motivated and refers to the strengths part
of the SDQ. As Rothenberger and Woerner (2004)
underline, it is necessary to include positive behav-
iours, protective factors, and compensatory
resources to gain broad and clinically meaningful
psychopathological profiles – otherwise, description
of a child’s psychiatric status will remain incomplete
and intervention planning may run into difficulties.
Also, as pointed out by previous studies (Dickey &
Blumberg, 2004; Ronning et al., 2004), removal
of the positively construed items could result in
decreased acceptability, and is therefore not
recommended.
Sixth, as indicated by different factor loadings,
parents and their children only modestly agree in
ratings on children’s strengths and difficulties
(Goodman, 2001). Method factors tend to make
large contributions to psychological measures, and
the results of the present study confirmed this is
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
also the case for the SDQ. The results indicated
that method factors had little influence on the
report of Peer Problems and Emotional Symptoms
in self-reports, and on the report of Hyperactivity
problems in parent/proxies. Goodman (2001)
demonstrated that the full combination of the
three-informant SDQ (parent, teacher, and self)
questionnaires yielded the highest sensitivity in
predicting psychiatric disorders. Our findings
support the use of multi-informant SDQ in the
assessment of children’s mental health.
Finally, internal reliability, examined with Cron-
bach’s alpha, was satisfactory for the summary
score of the SDQ consisting of 20 items but modest
for some of the subscales. The low Cronbach’s
alpha for the subscale measuring Conduct Problems
is problematic and has been reported in previous
studies as well (Bourdon, Goodman, Rae, Simpson,
& Koretz, 2005; Muris et al., 2003). Results
measured in this subscale should be interpreted
with caution. Low Cronbach’s alpha suggests that
the items in this subscale are not only tapping
Conduct Problems, which is also reflected by the
correlated error terms between item 18 (Lies) with
item 19 (Bullied) from Peer Problems. Cronbach’s
alpha is relatively modest for the Hyperactivity and
Peer problem subscales in preadolescence (<0.60).
This emphasises that, in this age group in particular,
the items in the different subscales not only reflect
their respective domains, but are related to several
domains. Higher cross-scale correlations in younger
children are not surprising, because possible
emotional and behavioural difficulties are less
distinct early in life, and many children have
problems in different areas at the same time. The
limited number of response categories (three) might
be another reason for low internal reliability (Fayers
& Machin, 2007, p. 126).
Study limitations
The five-factor SDQ has been widely adopted in epi-
demiological research in Norway. In the present
study we therefore choose a CFA approach to eval-
uate this particular latent structure. Although the
results indicated satisfactory model fit for the pro-
posed structure, there may be alternative models
that fit equally well or perhaps even better.
Another potential limitation is that no proxy data
were available for early and late adolescents. The
correspondence between self-report and proxies for
older age groups is likely to differ from findings in
preadolescence. Of that reason, the findings on
proxy data in our study cannot be generalized to all
age groups.
Finally, in a validation process, two questions
need to be addressed explicitly (Messick, 1975): Is
the SDQ any good as a measure of psychological
adjustment in children and adolescents and, should
the SDQ be used for the proposed purpose?
Ó 2008 The Authors

Evaluation of psychometric properties refers to the
first question, which has been the topic of the pres-
ent paper. However, the present study lacked infor-
mation on important psychometric properties such
as sensitivity, responsiveness and test–retest
repeatability. To be of best practical value for
screening, clinical purposes or in randomised trials,
it is important also to address these properties.
Answering the second question requires an evalua-
tion of the potential consequences of the use of SDQ
in epidemiological and clinical studies. It is neces-
sary to take into account ethical/social consider-
ations, balancing between a wish to investigate the
mental health of children and the availability of
effective treatment services, when knowledge on
possible psychopathology and burden is established.
How to manage this challenge deserves closer
attention in future research.
Conclusions
The results of the present study indicated support for
the proposed five-factor structure of the SDQ (Good-
man, 2001) across a wide age range (10–19 years),
including older adolescents and different informants.
However, some improvements should be considered
to improve internal reliability and conceptual clarity.
Our results indicated support for the suggestions
made by Ronning et al. (2004) that the Norwegian
version of the SDQ would probably benefit from
‘avoiding semantic similarities in item wording, posi-
tive and negative wording within the same construct,
and two sentences within the same item’.
However, as stated by Goodman, Iervolino, Colli-
shaw, Pickles, and Maughan (2007), potential
changes in item wording should only be adapted
after careful consideration and pilot testing as
seemingly minor changes to a questionnaire can
make a big difference to mean scores and create
problems for meaningful comparisons between the
original and the modified measure and for combining
results in meta-analyses.
Acknowledgements
This research was completed with assistance from
EXTRA funds from the Norwegian Foundation for
Health and Rehabilitation, and performed in coop-
eration with the Norwegian Health Services Research
Centre, incorporated in the Norwegian Knowledge
Centre for Health Services.
Correspondence to
Betty Van Roy, University in Oslo, Institute of Psy-
chiatry, PB 1078 Blindern, 0316 Oslo, Norway; Tel:
+47 23 49 21 20; Fax: +47 23 49 23 02; Email:
r.e.van@medisin.uio.no
Ó 2008 The Authors
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
Construct validity of the SDQ 1311
References
Achenbach, T.M. (1991). Manual for the Child Behavior
Checklist/ 4–18 and 1991 Profile. Burlington: Uni-
versity of Vermont, Department of Psychiatry.
American Psychiatric Association. (1994). Diagnostic
and statistical manual of mental disorders (DSM-IV)
(4th edn). Washington, DC: American Psychiatric
Association.
Becker, A., Steinhausen, H.C., Baldurson, G., Dalsg-
aard, S., Lorenzo, M.J., Ralston, S.J., Döpfner, M., &
Rothenberger, A. (2006). Psychopathological screen-
ing of children with ADHD: Strengths and Difficulties
Questionnaire in a pan-European study. European
Child and Adolescent Psychiatry, 15(Suppl. 1), 56–62.
Becker, A., Woerner, W., Hasselhorn, M., Banaschew-
ski, T., & Rothenberger, A. (2004). Validation of the
parent and teacher SDQ in a clinical sample. Euro-
pean Child and Adolescent Psychiatry, 13(Suppl. 2),
II11–II16.
Bollen, K.A. (1989). Structural equations with latent
variables. New York: John Wiley.
Bourdan, K.H., Goodman, R., Rae, D.S., Simpson, G., &
Koretz, D.S. (2005). The Strengths and Difficulties
Questionnaire: U.S. normative data and psycho-
metric properties. Journal of the American Academy
of Child and Adolescent Psychiatry, 44, 557–564.
Campbell, D.T., & Fiske, D.W. (1959). Convergent and
discriminant validation by the multitrait-multi-
method matrix. Psychological Bulletin, 2, 81–105.
Dickey, W.C., & Blumberg, S.J. (2004). Revisiting the
factor structure of the Strengths and Difficulties
questionnaire: United States, 2001. Journal of the
American Academy of Child and Adolescent Psychi-
atry, 43, 1159–1167.
Elander, J., & Rutter, M. (1996). Use and development
of the Rutter Parents’ and Teachers’ Scales. Inter-
national Journal of Methods in Psychiatric Research,
6, 63–78.
Fayers, P.M., & Machin, D. (2007). Quality of life: The
assessment, analysis and interpretation of patient
reported outcome. Chichester: John Wiley & Sons.
Goodman, R. (1997). The Strengths and Difficulties
Questionnaire: A research note. Journal of Child
Psychology and Psychiatry, 38, 581–586.
Goodman, R. (2001). Psychometric properties of the
Strengths and Difficulties questionnaire. Journal of
the American Academy of Child and Adolescent Psy-
chiatry, 40, 1337–1345.
Goodman, R., Iervolino, A.C., Collishaw, S., Pickles, A.,
& Maughan, B. (2007). Seemingly minor changes to a
questionnaire can make a big difference to the mean
scores: A cautionary tale. Social Psychiatry and
Psychiatric Epidemiology, 42, 322–327.
Goodman, R., Meltzer, H., & Bailey, V. (2003). The
Strengths and Difficulties Questionnaire: A pilot
study on the validity of the self-report version. Inter-
national Review of Psychiatry, 15, 173–177.
Goodman, R., & Scott, S. (1999). Comparing the
Strengths and Difficulties Questionnaire and the
Child Behavior Checklist: Is small beautiful? Journal
of Abnormal Child Psychology, 27, 17–24.
Heiervang, E., Stormark, K.M., Lundervold, A.J., Hei-
mann, M., Goodman, R., Posserud, M.B., et al.
(2007). Psychiatric disorders in Norwegian 8- to
1312 Betty Van Roy, Marijke Veenstra, and Jocelyne Clench-Aas
10-year olds: An epidemiological survey of preval-
ence, risk factors, and service use. Journal of the
American Academy of Child and Adolescent Psychi-
atry, 46, 438–447.
Heyerdahl, S. (2003). Strengths and Difficulties Ques-
tionnaire: A new questionnaire for mental health
assessment in children and adolescents, used in
recent Norwegian surveys: UNGHUBRO, OPPHED
and TROFINN. Nordic Journal of Epidemiology, 13,
127–135.
Jöreskog, K.G. (2005). Structural equation modeling
with ordinal variables using LISREL. Revised 2005.
Available at http://www.ssicentral.com/lisrel/tech
docs/ordinal.pdf.
Koskelainen, M., Sourander, A., & Vauras, M. (2001).
Self-reported strengths and difficulties in a commun-
ity sample of Finnish adolescents. European Child
and Adolescent Psychiatry, 10, 180–185.
Loehlin, J.C. (1998). Latent variable models. An intro-
duction to factor, path and structural Analysis. Mah-
wah, NJ: Lawrence Erlbaum Associates.
Marsh, H.W., & Hocevar, D. (1983). Confirmatory factor
analysis of multitrait-multimethod matrices. Journal
of Educational Measurement, 20, no 3, 231–248.
Marzocchi, G., Capron, C., Di Pietro, M., Tauleria, E.,
Duyme, M., Frigerio, A., et al. (2004). The use of the
Strengths and Difficulties Questionnaire (SDQ) in
Southern European countries. European Child and
Adolescent Psychiatry, 13(Suppl. 2), 40–46.
Mathai, J., Anderson, P., & Bourne, A. (2004). Com-
paring psychiatric diagnoses generated by the
Strengths and Difficulties Questionnaire with dia-
gnoses made by clinicians. Australian and New
Zealand Journal of Psychiatry, 38, 639–643.
Messick, S. (1975). The standard problem. Meaning and
values in measurement and evaluation. American
Psychologist, 75, 955–966.
Muris, P., Meesters, C., Eijkelenboom, A., & Vincken,
M. (2004). The self-report version of the Strengths
and Difficulties Questionnaire: Its psychometric
properties in 8 to 13-year-old non-clinical children.
British Journal of Clinical Psychology, 43, 437–448.
Muris, P., Meesters, C., & van den Berg, F. (2003). The
Strengths and Difficulties Questionnaire (SDQ) –
further evidence for its reliability and validity in a
community sample of Dutch children and adoles-
cents. European Child and Adolescent Psychiatry, 12,
1–8.
Journal compilation Ó 2008 Association for Child and Adolescent Mental Health.
Obel, C., Heiervang, E., Rodriguez, A., Heyerdahl, S.,
Smedje, H., Sourander, A., Guethmundsson, O.O.,
Clench-Aas, J., Christensen, E., Heian, F., Mathie-
sen, K.S., Magnusson, P., Njarethvik, U., Koskelai-
nen, M., Ronning, J.A., Stormark, K.M., & Olsen, J.
(2004). The Strengths and Difficulties Questionnaire
in the Nordic countries. European Child and Adoles-
cent Psychiatry, 13(Suppl. 2), II32–II39.
Palmieri, P.A., & Smith, G.C. (2007). Examining the
structural validity of the Strengths and Difficulties
Questionnaire (SDQ) in a U.S. sample of custodial
grandmothers. Psychological Assessment, 19, 189–
198.
Ronning, J.A., Handegaard, B.H., Sourander, A., &
Morch, W.T. (2004). The Strengths and Difficulties
Self-Report Questionnaire as a screening instrument
in Norwegian community samples. European Child
and Adolescent Psychiatry, 13, 73–82.
Rothenberger, A., & Woerner, W. (2004). Editorial.
Strengths and Difficulties Questionnaire (SDQ) –
evaluations and applications. European Child and
Adolescent Psychiatry, 13(Suppl. 2), II/1–II/2.
Smedje, H., Broman, J.E., Hetta, J., & von Knorring,
A.L. (1999). Psychometric properties of a Swedish
version of the ‘Strengths and Difficulties Question-
naire’. European Child and Adolescent Psychiatry, 8,
63–70.
Van Roy, B., Groholt, B., Heyerdahl, S., & Clench-Aas,
J. (2006). Self-reported strengths and difficulties in a
large Norwegian population 10–19 years: Age and
gender specific results of the extended SDQ-ques-
tionnaire. European Child and Adolescent Psychiatry,
15, 189–198.
Vostanis, P. (2006). Strengths and Difficulties Ques-
tionnaire: Research and clinical applications. Current
Opinion in Psychiatry, 19, 367–372.
Woerner, W., Fleitlich-Bilyk, B., Martinussen, R., Fletch-
er, J., Cucchiaro, G., Dalgalarrondo, P., Lui, M., &
Tannock, R. (2004). The Strengths and Difficulties
Questionnaire overseas: Evaluations and applications
of the SDQ beyond Europe. European Child and
Adolescent Psychiatry, 13(Suppl. 2), II47–II54.
World Health Organization. (1994). The ICD-10 Classi-
fication of Mental and Behavioural Disorders: Dia-
gnostic Criteria for Research. Geneva: WHO.
Manuscript accepted 8 April 2008
Ó 2008 The Authors