Content uploaded by Sherry H Stewart
Author content
All content in this area was uploaded by Sherry H Stewart on May 19, 2015
Content may be subject to copyright.
Original Research
Early Temperament Prospectively Predicts Anxiety in
Later Childhood
Valerie V Grant, BA (PhD Candidate)1; Alexa L Bagnell, MD, FRCPC2;
Christine T Chambers, PhD3; Sherry H Stewart, PhD4
Key Words: anxiety, temperament, unadaptable, behavioural inhibition, fearful distress,
fussy–difficult, irritable distress, population
WLa Revue canadienne de psychiatrie, vol 54, no 5, mai 2009320
Objective: To investigate the contribution of early childhood temperamental constructs
corresponding to 2 subtypes of general negative emotionality—fearful distress
(unadaptable temperament) and irritable distress (fussy–difficult temperament)—to later
anxiety in a nationally representative sample.
Method: Using multiple linear regression analyses, we tested the hypothesis that
caregiver-reported child unadaptable temperament and fussy–difficult temperament scales
of children aged 2 to 3 years (in 1995) would prospectively predict caregiver-reported
child anxiety symptoms at ages 4 to 5, 6 to 7, 8 to 9, and 10 to 11 years, and child-reported
anxiety at 10 to 11 years (controlling for sex, age, and socioeconomic status) in a
nationally representative sample from Statistics Canada’s National Longitudinal Survey of
Children and Youth (initial weighted n= 768 600).
Results: Only fussy–difficult temperament predicted anxiety in children aged 6 to 7 years.
In separate regressions, unadaptable temperament and fussy–difficult temperament each
predicted anxiety at 8 to 9 years, but when both were entered simultaneously, only
unadaptable temperament remained a marginal predictor. Temperament did not
significantly predict caregiver- or child-reported anxiety at 10 to 11 years, suggesting that
as children age, environmental factors may become more important contributors to anxiety
than early temperament.
Conclusion: Our results provide the first demonstration that early temperament is related
to later childhood anxiety in a nationally representative sample.
Can J Psychiatry. 2009;54(5):320–330.
Clinical Implications
·Biologically based traits such as temperament can predispose children to anxiety.
·The current results are consistent with calls for increased efforts to intervene directly with
temperament in the prevention of childhood anxiety pathology.
Limitations
·The anxiety scales used in the analyses were very brief.
·The anxiety scales measured anxiety-relevant symptoms, not anxiety disorders.
·Temperament was measured with caregiver-reported subscales, which could potentially be
biased by caregiver motivations or perceptions.
Many children and adolescents experience fears and anxi-
ety during the course of normal development.1For
some, anxiety reaches a clinically significant level. A recent
review revealed that anxiety disorders are common in preado-
lescents.2Anxiety disorders in children are important not only
due to their prevalence but also because of the significant
impairment in social and academic domains that they can
cause.3Further, there is retrospective4and prospective5evi-
dence to indicate that childhood anxiety disorders are not sim-
ply transient problems that will resolve on their own.
Although current psychological treatments for childhood
anxiety disorders are effective, many children still meet diag-
nostic criteria for an anxiety disorder at treatment
completion.6Therefore, recent research has focused on pre-
vention: intervention that takes place before the onset of an
anxiety disorder with the aim of averting new cases.7
It is important to determine the risk factors for anxiety because
they could be useful for identifying targets of selective pre-
vention strategies.8One prominent line of anxiety risk factor
research has focused on temperament—biologically based
behavioural traits appearing very early in life.9A wide array
of measures and terms have been used to describe tempera-
ment, but a review indicates that the findings converge.10 Two
different types of “general negative emotionality” constructs
consistently emerge, in both infancy and childhood tempera-
ment research: “fearful distress” and “irritable distress.”10, p 113
Behavioural inhibition, which corresponds to unadaptable
temperament or fearful distress, is the temperamental con-
struct that has received the most attention in the anxiety risk
factor literature. Behaviourally inhibited children tend to be
unadaptable to new people or situations11 and display shyness,
cautiousness, and emotional restraint in the face of unfamiliar
people or situations.12 Fussy–difficult temperament corre-
sponds to the irritable distress construct and is exemplified by
infants and children who are irritable, easy to upset, difficult
to soothe, and who frequently display negative affect.9
There have been 2 primary approaches to the study of
temperament–anxiety relations: relatively small-scale inves-
tigations involving convenience samples (for example, in the
laboratory), and large-scale epidemiologic explorations of
population-representative samples. In the laboratory, studies
using small samples selected to have high or low behavioural
inhibition (assessed through behavioural observation) have
demonstrated that behavioural inhibition early in life predicts
later anxiety,13 particularly when the behavioural inhibition
is stable across time.14 Research using parent reports of child
temperament has also revealed a positive relation between
behavioural inhibition and anxiety.15 In addition, some
cross-sectional16 and prospective13 findings suggest that
behavioural inhibition confers a specific vulnerability for
social anxiety.
Compared with the literature on behavioural inhibition
(unadaptable temperament or fearful distress), information
concerning fussy–difficult temperament (irritable distress) is
relatively sparse. Research findings with convenience sam-
ples have generally indicated that parent-reported
fussy–difficult temperament is associated concurrently17 and
prospectively18 with both internalizing9(for example,
depression and anxiety) and externalizing9(for example,
aggressiveness) problems. However, several of the studies in
this area were limited in that they examined only one compo-
nent of fussy–difficult temperament19 or used broad-based
measures of internalizing problems that are not specific to
anxiety.18
In comparison with the many smaller-scale studies, there are
relatively few epidemiologic investigations that report the
relation between temperament and anxiety using large
population-representative samples. One such study investi-
gated prospective mother-rated temperamental predictors of
child anxiety in adolescence, including fearful distress
(shy-inhibited temperament) and a single aspect of irritable
distress (reactive temperament), in a cohort of children
selected to be representative of one Australian state.20
Shy-inhibited children were significantly more likely than
non–shy-inhibited children to be clinically anxious (by child-
and mother-reported composite ratings) in adolescence.
However, reactive temperament (an indicator of
difficultness9) did not predict anxiety problems in
adolescence.
The temperamental precursors of child and adolescent
psychopathology were assessed in a longitudinal sample
recruited from a complete cohort of consecutive births occur-
ring during a 1-year period in one New Zealand city.21
Experimenter-rated approach temperament (the conceptual
inverse of unadaptable temperament) in early childhood was
inversely related to aggregated parent and teacher ratings of
anxiety in late childhood and to parent-rated anxiety in
adolescence. “Lack of control, irritability, and
distractibility”21,p59(a temperamental construct including
some elements of irritable distress), measured in early child-
hood was positively related to parent and teacher ratings of
Early Temperament Prospectively Predicts Anxiety in Later Childhood
The Canadian Journal of Psychiatry, Vol 54, No 5, May 2009 W321
Abbreviations used in this article
BSSE bootstrap standard error
ICQ Infant Characteristics Questionnaire
OCHS Ontario Child Health Study
NLSCY National Longitudinal Survey of Children and Youth
SES socioeconomic status
anxiety in late childhood and parent-rated anxiety in
adolescence.
In the extant literature, there are no investigations of the rela-
tion between early childhood temperament and later anxiety
symptoms in a nationally representative sample. The current
study is the first to investigate the contribution of early child-
hood temperamental constructs corresponding to both sub-
types of general negative emotionality—fearful distress and
irritable distress—to later anxiety in a nationally representa-
tive sample (from the NLSCY22).
Basic demographic characteristics that have been shown to be
related to internalizing problems in children (that is, sex,18
SES,23 and age24) were statistically controlled in the data anal-
yses to ensure a stringent test of the temperamental predictors,
in contrast to prior research that has failed to account for the
potential impact of some of these basic demographics charac-
teristics (for example, Prior et al20 did not control for SES in
their analyses). The hypothesis was that caregiver-reported
child unadaptable temperament (fearful distress) and
fussy–difficult temperament (irritable distress) at ages 2 to
3 years would prospectively predict caregiver-reported child
anxiety symptoms at ages 4 to 5 , 6 to 7 , 8 to 9 , and 10 to 11
years, and child-reported child anxiety symptoms at ages 10 to
11 years, while controlling for child sex, age, and SES.
Method
The National Longitudinal Survey of Children and Youth
The NLSCY is a national, longitudinal, probabilistic survey,
designed to track the development of a sample representative
of the Canadian civilian population living in the 10 prov-
inces.22 The first cycle of data collection was executed in
1994/95, with additional cycles of data collection occurring
every 2 years. For the sample used in the current investigation,
data collection was conducted by telephone or in person using
computer-assisted interviewing, with the exception of
child-reported anxiety information at ages 10 to 11 years,
which was collected through written questionnaire. The over-
all response rate to the NLSCY was 86.3%.25 The current
investigation was a component of a larger project, headed by
one of the authors, for which data access was approved by an
adjudicating committee with representation from the Social
Sciences and Humanities Research Council and from Statis-
tics Canada. Ethical approval for the entire study was pro-
vided by the IWK Health Centre’s Research Ethics Board.
Statistics Canada created survey weights that are used to
ensure that the NLSCY data represent the Canadian popula-
tion at the time of selection into the survey (that is, January
1995 for the longitudinal cohort that began in Cycle 1).22 Each
person in the sample is assigned a weight equivalent to the
number of people in the population that it represents. For the
current investigation, the Cycle 5 longitudinal funnel weights
were applied, which are appropriate for respondents who par-
ticipated in each of the first 5 cycles of data collection.
To conform to the release guidelines put forth by Statistics
Canada, a minimum of 5 contributing respondents was
required for the estimates reported here.22 Further, the Statis-
tics Canada rounding guidelines were followed herein.22
Bootstrap Approach to Variance Estimation
The NLSCY has a complex sample design that involves strat-
ification, multiple selection stages, and unequal probabilities
of selecting respondents.22 Correct point estimates (for exam-
ple, means and correlations) can be calculated by simply
applying the appropriate survey weights. However, typical
methods for estimating variance, based on the assumption of
a simple random sample, are unsuitable and may lead to
underestimates of true variance, which might in turn lead to
inappropriate rejection of the null hypothesis in statistical
testing.22
To permit proper inference, bootstrap weights must be used
for variance estimation. The bootstrap approach is a
replicate-based variance estimation technique.26 This tech-
nique involves the random selection (from the full sample) of
synthetic subsamples, or replicates, each of which has the
same design as the entire sample.22 Then, weights for the
cases within each subsample are recalculated in the same
manner as is used for the full sample (see previous paragraph)
to create a final set of, in the case of the NLSCY, 1000 boot-
strap weights per person. Population estimates for a given
characteristic are calculated for each subsample using the
bootstrap weights. Then, to estimate the sampling variance
(see definition in the following paragraph) in the population,
the variance among the 1000 subsample estimates is calcu-
lated. In the current investigation, the set of 1000 Cycle 5 lon-
gitudinal funnel bootstrap weights was used for all analyses
requiring variance estimation.
Sampling variance refers to the variability in an estimate
resulting from use of a sample instead of a census.22 It reflects
sampling error, which is influenced by population variance,
sample design, sample size, and response rate to the survey.
The standard error of an estimate is equivalent to the square
root of the sampling variance and is the variance estimation
statistic used in the present data analyses. The confidence
intervals presented herein are based on standard errors that
were derived using bootstrap weights (that is, BSSE).
Participants
Respondents were people most knowledgeable about (that is,
caregivers of) children from the longitudinal cohort that
began participation in the NLSCY in Cycle 1 (that is, 1995).
Child-report data from Cycle 5 were also included. Respon-
dents who had participated in all 5 cycles were selected to
WLa Revue canadienne de psychiatrie, vol 54, no 5, mai 2009322
Original Research
heighten the comparability of the sample across cycles. The
sample was further restricted to caregiver respondents whose
children were aged 2 to 3 years at Cycle 1 (and 10 to 11 years
at Cycle 5), leaving weighted n= 790 400 (unweighted n=
2037). The rationale for this restriction was that information
about temperament (collected about young children in Cycle
1 only) and about child-reported anxiety (collected at Cycle 5
only from children aged 10 years and older) was provided by
this subgroup.
In the resultant sample, none of the respondents were missing
any of the demographic predictor variables for the multiple
regression analyses used to test the hypothesis. Next, the fol-
lowing missing value exclusion rule was applied: respon-
dents were required to have at least 60% of the component
items for each of the predictor temperament subscales
employed in the regression analyses (that is, 3 of the 5
unadaptable subscale items and 6 of the 9 fussy–difficult
subscale items) and to have a minimumof 67% of the compo-
nent items for at least one of the possible anxiety outcome
subscales (that is, 2 of the 3 items for the caregiver-reported
child anxiety items for either Cycle 2, Cycle 3, Cycle 4, or
Cycle 5, or the child-reported anxiety items for Cycle 5).
Following the application of this missing value exclusion
rule, a weighted n= 768 600 (unweighted n= 2001)
remained. Table 1 displays the sample demographic charac-
teristics. Among those who had enough data to be included in
the regression analyses, 38.1% (95% CI 34.6% to 41.6%) had
at least one missing value on one of the variables used for the
scales that were entered in the regression analyses. The mean
number of missing values on variables comprising the regres-
sion subscales was 1.2 (95% CI 1.1 to 1.3). A small percent-
age (6.3%; 95% CI 3.6% to 9.0%) of child targets shared a
household with another child in this sample. Except for the
weighting variable, the Canadian Research Institute for
Social Policy-NLSCY files (version 1.1), which are essen-
tially groomed versions of the original Statistics Canada
NLSCY files, were analyzed herein.27
Materials
Child Sex and Age. Child sex was a dichotomous variable
(0 = boys; 1 = girls). The child’s age in years (that is, 2 or 3
years) at the time of the interview in Cycle 1 was used in the
analyses.
Socioeconomic Status. This composite variable was derived
from the level of education and occupation(s) of the target
child’s parent(s), along with the target child’s household
income, scaled to 1994 values.27 This measure was standard-
ized on the entire sample of households surveyed in the
NLSCY (that is, it is a zscore). The mean SES in the current
sample was 0.0 (95% CI –0.1 to 0.0).
Early Temperament Prospectively Predicts Anxiety in Later Childhood
The Canadian Journal of Psychiatry, Vol 54, No 5, May 2009 W323
Table 1 Description of Canadian subpopulation
studied in multiple regression analyses
Variable (weighted na) % (95% CI)
Sex of child (768 600)
Girls 50.0 (48.7 to 51.4)
Boys 50.0 (48.6 to 51.3)
Age of child (initial) (768 600), years
2 47.7 (45.5 to 50.0)
3 52.3 (50.0 to 54.5)
Ethnicity of child (data collected at ages 4
to 5 years) (623 700)b
White 90.3 (87.4 to 93.2)
South Asian (for example, Indian,
Pakistani)
2.1 (0.5 to 3.7)
Chinese 1.8 (0.7 to 3.0)
Filipino 1.8 (–0.1 to 3.7)
Native and (or) Aboriginal 0.9 (0.4 to 1.4)
Black (for example, African, Haitian) 0.7 (–0.2 to 1.7)
Latin American 0.4 (0.0 to 0.9)
Other (Arab and [or] West Asian,
Southeast Asian, Japanese, and
Other)c
2.8 (1.0 to 4.7)
Relationship of caregiver to child (768 600)
Birth mother 91.2 (89.1 to 93.3)
Adoptive mother 0.5 (0.0 to 1.1)
Stepmother, foster mother, or other
related female
0.8 (0.0 to 1.5)
Birth father, stepfather, foster father, or
other related male
7.5 (5.5 to 9.5)
Age of caregiver (768 600), years
15 to 24 8.4 (6.6 to 10.1)
25 to 29 24.1 (21.0 to 27.1)
30 to 34 39.8 (35.9 to 43.7)
35 to 39 20.6 (17.6 to 23.6)
³40 7.2 (4.7 to 9.6)
aWeighted using Cycle 5 longitudinal funnel weights.
bRace categories are not mutually exclusive.
cThe Arab and (or) West Asian, Southeast Asian, and Japanese
categories were combined with the original Other category to follow
Statistics Canada guidelines that there be at least 5 unweighted cases
per category.
Temperament Measures. The NLSCY measured child tem-
perament at Cycle 1 only. The (caregiver-reported) tempera-
ment items used in the NLSCY were derived from the ICQ for
infants aged 6 months, which has subscales corresponding to
fussy–difficult, unadaptable, dull, and unpredictable tempera-
ments.11,25 The 6-month version of the ICQ has shown ade-
quate factor structure, internal consistency, and test–retest
reliabilities for the subscales.11 In addition, parent-reported
fussy–difficultness was correlated with objective observa-
tions of fussiness, demonstrating the validity of this
subscale.11 Further, there was convergence of the
fussy–difficult and unadaptable subscales with conceptually
related subscales on other measures.11
The ICQ items were slightly modified to be more appropriate
for NLSCY children aged 3 years; thus the wording of tem-
perament items differed somewhat for 2- compared with
3-year-olds included in our study.25 In our investigation,
fussy–difficult and unadaptable subscales were created from
means of component items. Each item was scored on a scale of
1 (more favourable) to 7 (more unfavourable, inhibited, or dif-
ficult). The fussy–difficult subscale was comprised of 9 items
(for example, “How easy or difficult is it for you to calm or
soothe [child’s name] when he/she is upset?”). The
unadaptable subscale had 5 items (for example, “How does
he/she respond to being in a new place?”). Descriptive statis-
tics and Cronbach’s alphas (internal consistencies) for the
temperament subscales are presented in Table 2. Though the
unadaptable subscale internal consistency reliability is less
than the widely accepted cut-off of 0.70, it is adequate accord-
ing to Loewenthal’s standard, which suggests that a
Cronbach’s alpha of greater than 0.60 is acceptable for short
scales.28
Anxiety Scales. There were 5 anxiety scales analyzed in our
study (that is, caregiver-reported at ages 4 to 5, 6 to 7, 8 to 9,
and 10 to 11 years, and child-reported at ages 10 to 11 years).
Each anxiety scale was comprised of the mean of 3 anxiety-
relevant items taken from the broader NLSCY emotional dis-
order scale,25 which was adapted from the original OCHS
emotional disorder scale.29 The original OCHS emotional dis-
order scale consists of items selected to reflect essential com-
ponents of affective disorder, overanxious disorder, and
obsessive–compulsive disorder.30 Boyle et al29 found that the
emotional disorder scale had satisfactory internal consistency
and test–retest reliability. In addition, the validity of the emo-
tional disorder scale was demonstrated by its prediction of
psychiatric status (that is, whether or not a child attended a
mental health clinic) in 6- to 16-year-olds.29 Further, Boyle
et al29 found the emotional disorder scale to be significantly
associated with overanxious disorder scale scores on a revised
version of the OCHS scales. In turn, both parent-report31 and
youth-report32 versions of the revised OCHS were found to
have reliability and validity comparable with a structured
diagnostic interview.
Three subscales comprise the original OCHS emotional dis-
order scale: dysphoric mood, strong feelings of tension, and
compulsive, obsessive behaviour.29 The caregiver- and
child-reported anxiety-relevant items selected for the current
study (for example, “How often would you say that [child’s
name] is too fearful or anxious?” [caregiver-reported]) corre-
sponded to the 3 items on the OCHS strong feelings of ten-
sion subscale, which each showed good convergent and
discriminant validity in item analyses.29 The anxiety items
used in the current investigation were each scored on a scale
of 1 = never or not true to 3 = often or very true.22 See Table 2
for descriptive statistics of the anxiety scales, which all had
acceptable internal consistencies.
Data Analyses
Data analysis for inferential tests and for variance estimation
(that is, calculating BSSEs) was conducted with Stata soft-
ware.33 This software has the capacity to analyze complex
survey data using a set of survey (svy) commands. Software
packages (such as Stata Release 9) that can accommodate
balanced repeated replication weights in their analyses can
use a balanced repeated replication variance estimator along
with bootstrap weights to generate bootstrap variance esti-
mates.26 In our investigation, the svy:regress command was
used with the Stata balanced repeated replication variance
estimator to produce regression coefficients based on boot-
strap weights.
Stata does not have a svy command for correlation. Follow-
ing recent recommendations, the point estimates for correla-
tions were derived using the correlate command weighted
with the longitudinal Cycle 5 funnel weights.34 Then the
svy:regress command (as described above) was used to
derive the associated significance level by regressing each of
the 2 values to be correlated on each other and taking the larg-
est (that is, most conservative) Pvalue of the 2.34
The central analyses involved a series of multiple linear
regression analyses in which child sex, age, SES, and temper-
amental subscales (that is, the unadaptable subscale or the
fussy–difficult subscale) measured at ages 2 to 3 years were
entered as predictors. Criteria variables were the caregiver-
reported anxiety scales for children at ages 4 to 5, 6 to 7, 8 to
9, and 10 to 11 years. The child-reported anxiety scale at ages
10 to 11 years served as an additional criterion variable.
Results
The correlations among the temperament and anxiety vari-
ables are displayed in Table 2. Unadaptable and
fussy–difficult temperament had a small, significant positive
correlation. Fussy–difficult temperament was positively
WLa Revue canadienne de psychiatrie, vol 54, no 5, mai 2009324
Original Research
Early Temperament Prospectively Predicts Anxiety in Later Childhood
The Canadian Journal of Psychiatry, Vol 54, No 5, May 2009 W325
Table 2 Descriptive statistics and correlations among temperament variables and child anxiety variables
Correlations (P) among temperament and child anxiety variablesb
Scale, ages in years
Weighted na
(Weighted nfor a)cM(95% CI) a12345 6 7
1. Unadaptable temperament
2to3
768 600
(767 500)
2.5 (2.4 to 2.5) 0.65 — 0.24
(<0.001)
0.00
(0.90)
0.05
(0.32)
0.11
(0.05)
–0.02
(0.60)
–0.02
(0.73)
2. Fussy–difficult temperament
2to3
768 600
(767 400)
3.1 (3.0 to 3.1) 0.75 — 0.04
(0.44)
0.12
(0.004)
0.09
(0.03)
0.03
(0.60)
0.01
(0.78)
3. Caregiver-reported child anxiety
4to5
768 600
(766 000)
1.3 (1.3 to 1.4) 0.64 — 0.44
(<0.001)
0.26
(<0.001)
0.28
(<0.001)
0.06
(0.13)
4. Caregiver-reported child anxiety
6to7
719 300
(718 200)
1.4 (1.4 to 1.4) 0.72 — 0.50
(<0.001)
0.35
(<0.001)
0.19
(<0.001)
5. Caregiver-reported child anxiety
8to9
739 800
(738 000)
1.5 (1.4 to 1.5) 0.70 — 0.50
(<0.001)
0.21
(<0.001)
6. Caregiver-reported child anxiety
10 to 11
732 900
(730 500)
1.5(1.4to1.5) 0.75 —0.16
(<0.001)
7. Child-reported anxiety
10 to 11
579 200
(551 100)
1.6(1.6to1.7) 0.67 —
aWeighted using Cycle 5 longitudinal funnel weights.
bThe weighted nfor each correlation may be less than either of the weighted ns for the correlated scales as only those cases with both scales could be included in a given bivariate correlation. Point estimates of
correlations are calculated with the Cycle 5 longitudinal funnel weights applied. To derive significance levels for correlations, each variable in a particular bivariate correlation was regressed on the other (using
bootstrap weights) and the higher Pvalue (most conservative) from the 2 regressions was followed.
cThe weighted nfor each scale’s Cronbach’s ais less than that for the corresponding mean as only cases without missing values for any of the component variables for each scale were included in the Cronbach’s
aanalyses.
correlated with caregiver-reported child anxiety at 6 to 7
years. Also, both fussy–difficult and unadaptable tempera-
ment were positively correlated with caregiver-reported child
anxiety at 8 to 9 years.
The results of the multiple linear regression analyses (Table 3)
are consistent with the pattern of correlations. In the multiple
regression analyses, caregiver-reported fussy–difficult tem-
perament at ages 2 to 3 years predicted caregiver-reported
child anxiety at ages 6 to 7 years and at ages 8 to 9 years, while
controlling for basic demographics. Caregiver-reported
unadaptable temperament at ages 2 to 3 years also predicted
caregiver-reported child anxiety at ages 8 to 9 years, while
controlling for basic demographics. However, when
unadaptable and fussy–difficult temperament (along with
child sex, age, and SES) were entered in the same regression
equation to predict caregiver-reported child anxiety at ages 8
to 9 years, unadaptable temperament remained a marginally
significant predictor (B = 0.05, t= 1.71, P= 0.09), but
fussy–difficult temperament did not (B = 0.03, t= 1.45, P=
0.15). Neither of the temperamental subscales predicted
caregiver-reported child anxiety at ages 4 to 5 years or at ages
10 to 11 years, while controlling for basic demographics.
Likewise, temperament did not predict child-reported anxiety
at ages 10 to 11 years.
Discussion
The results of the current investigation represent the first dem-
onstration in a nationally representative sample that 2 types of
early temperament, both broadly classified as general nega-
tive emotionality,10 each prospectively predicted anxiety in
later childhood, even with the effects of basic demographic
characteristics statistically controlled. As hypothesized,
caregiver-reported fussy–difficult temperament at ages 2 to
3 years prospectively predicted caregiver-reported child anxi-
ety at ages 6 to 7 and 8 to 9 years, controlling for basic demo-
graphic variables. In addition, consistent with expectation,
caregiver-reported unadaptable temperament at ages 2 to 3
years predicted caregiver-reported anxiety at ages 8 to 9 years,
controlling for basic demographic characteristics. The results
of the multiple regression and correlation analyses suggest
that unadaptable temperament was a relatively more powerful
predictor of anxiety at ages 8 to 9 years than fussy–difficult
temperament.
One possible reason why early fussy–difficult temperament
predicted anxiety at ages 6 to 7 years whereas early
unadaptable temperament was a more powerful predictor of
anxiety at ages 8 to 9 years is related to the changing social
demands that children experience as they age. Social worries
begin to increase at age 8 years, reflecting a growing impor-
tance of social interactions.35 The fact that social fears mount
as children age, taken with prior findings that behavioural
inhibition appears to confer a specific vulnerability for social
anxiety,13 is consistent with the current result that
unadaptable temperament predicts anxiety in older (but not
younger) children.
Contrary to expectation, early caregiver-reported child tem-
perament did not prospectively predict child anxiety at ages 4
to 5 years (caregiver-reported) or at ages 10 to 11 years
(caregiver- or child-reported). Perhaps temperament at ages
2 to 3 years did not predict anxiety at ages 4 to 5 years because
of the challenges inherent in assessing anxiety at that young
age.36 Indeed, a review of the internal consistency results for
the caregiver-reported child anxiety scales indicates that reli-
ability improves after ages 4 to 5 years. Early temperament
might not have predicted child- or caregiver-reported anxiety
at ages 10 to 11 years because, as children age, environmental
factors (for example, peer influences) become more impor-
tant contributors to anxiety than early temperament. In addi-
tion, the failure of temperament to predict anxiety at ages 10
to 11 years might be related to the fact that this occasion of
anxiety measurement was the most temporally distal from the
measurement of temperament (at ages 2 to 3 years).
Given that the objective of the NLSCY was to cover a broad
range of factors that influence or provide an index of chil-
dren’s development,22 practical constraints precluded the use
of lengthy, comprehensive measures of any one construct
(for example, anxiety). The brevity of the anxiety scales in
our study is a limitation, as this brevity is likely related to the
somewhat low reliabilities of the anxiety scales and to the
fairly low proportion of variance in anxiety scores explained
by the predictor variables. Even so, it is impressive that early
temperament significantly predicted later anxiety, despite
being measured by short anxiety scales.
The current anxiety scales are further limited in that they sim-
ply measure anxiety-relevant symptoms, not anxiety disorder
diagnoses, and they were not designed to distinguish among
subtypes of anxiety (for example, social anxiety or panic).
Nonetheless, as mentioned in our Method section, Boyle
et al29 found that scores on the emotional disorder scale from
which our anxiety scales were derived may be appropriate
proxies of psychological disorder. In addition, Ferdinand
et al37 determined that preadolescent children could not be
divided into distinct groups displaying similar constellations
of anxiety symptoms, thereby supporting the use of a more
general measure of anxiety as in our study.
In our investigation, temperament was measured with care-
giver-reported subscales, which may be biased by caregiver
perceptions or motivations.38 Thus some researchers (for
example, Biederman et al16 and Schwartz et al13) have relied
on direct observation of behaviour to determine tempera-
ment, which has the advantage (compared with caregiver
WLa Revue canadienne de psychiatrie, vol 54, no 5, mai 2009326
Original Research
Early Temperament Prospectively Predicts Anxiety in Later Childhood
The Canadian Journal of Psychiatry, Vol 54, No 5, May 2009 W327
Table 3 Multiple linear regression analyses prospectively predicting child anxiety at ages 4 to 5, 6 to 7, 8 to 9, and 10 to 11 years from predictor variables at
ages 2 to 3 years
With unadaptable temperament at ages 2 to 3 years as a predictor With fussy–difficult temperament at ages 2 to 3 years as a predictor
Outcome variable, age in years
(Weighted na)
Predictor
variable B tPR
2(P)
Predictor
variables B tP R
2(P)
Caregiver-reported child anxiety
4 to 5 (768 600) (Constant) 1.16 11.79 <0.001 (Constant) 1.13 8.79 <0.001
Child sexb–0.08 2.42 0.02 Child sex –0.08 2.42 0.02
Child agec0.08 2.55 0.01 Child age 0.08 2.59 0.01
SES –0.02 1.21 0.23 SES –0.02 1.23 0.22
Unadaptable 0.00 0.05 0.96 0.025 (<0.001) Fussy–difficult 0.01 0.60 0.55 0.026 (<0.001)
Caregiver-reported child anxiety
6 to 7 (719 300) (Constant) 1.05 12.03 <0.001 (Constant) 0.98 9.70 <0.001
Child sex 0.02 0.42 0.67 Child sex 0.02 0.41 0.68
Child age 0.11 3.31 0.001 Child age 0.10 2.93 0.003
SES 0.00 0.02 0.98 SES 0.00 0.10 0.92
Unadaptable 0.02 1.13 0.26 0.019 (0.003) Fussy–difficult 0.05 2.68 0.008 0.027 (0.001)
Caregiver-reported child anxiety
8 to 9 (739 800) (Constant) 1.21 12.34 <0.001 (Constant) 1.24 10.37 <0.001
Child sex 0.00 0.03 0.97 Child sex 0.00 0.03 0.97
Child age 0.05 1.29 0.20 Child age 0.03 0.86 0.39
SES 0.01 0.63 0.53 SES 0.01 0.42 0.68
Unadaptable 0.05 2.21 0.03 0.015 (0.05) Fussy–difficult 0.05 2.20 0.03 0.010 (0.26)
Caregiver-reported child anxiety
10 to 11 (732 900) (Constant) 1.53 12.75 <0.001 (Constant) 1.46 12.51 <0.001
Child sex 0.02 0.65 0.52 Child sex 0.02 0.68 0.50
Child age –0.01 0.35 0.73 Child age –0.02 0.43 0.67
SES –0.02 1.34 0.18 SES –0.02 1.35 0.18
Unadaptable –0.01 0.59 0.55 0.004 (0.59) Fussy–difficult 0.02 0.62 0.54 0.004 (0.50)
continued
report) of objectivity of measurement.38 However, the direct
observation approach generally relies on just a single occa-
sion of assessment. Conversely, caregiver-reported informa-
tion most likely reflects child temperament across a broad
range of situations and occasions.38
The primary strength of the current investigation was the use
of a large, nationally representative, longitudinal sample.
The inclusion of basic demographic statistical control
subjects to ensure that the effects of temperament on later
anxiety were not entirely due to demographic variables com-
monly associated with anxiety was another strong point.
Moreover, 2 potential temperamental predictors of anxiety
were examined herein, in contrast to most past research that
has focused on only one type of temperament (for example,
Schwartz et al13).
The results of our investigation are important in that they are
from a large, nationally representative, longitudinal sample
and support other findings implicating early childhood tem-
perament as a risk factor for anxiety disorders in later child-
hood (for example, Schwartz et al13). Although our analyses
yielded statistically significant associations between early
temperament and later anxiety, the relations were modest in
magnitude and present in only 3 of the 10 instances tested,
precluding us from drawing strong conclusions regarding the
study’s clinical implications. Nonetheless, the current results
are consistent with calls for increased efforts to intervene
directly with temperament in the prevention of childhood
anxiety pathology.38
Funding and Support
Ms Grant has been supported by a Nova Scotia Health Research
Foundation Student Award, a Social Sciences and Humanities
Research Council (SSHRC) Doctoral Fellowship, Killam
Predoctoral Scholarships, and funds from Raising and Levelling
the Bar, awarded by SSHRC to the Canadian Research Institute
for Social Policy during this research. Dr Bagnell is supported by
Raising and Levelling the Bar New Investigators Network. Dr
Chambers is supported by a Canada Research Chair. Dr Stewart
was supported through a Canadian Institutes of Health Research
Investigator Award at the time this research was conducted and is
currently supported through a Faculty of Science Killam
Research Professorship at Dalhousie University. The analyses are
based on data from Statistics Canada, but the opinions expressed
do not represent the views of Statistics Canada. Specific names of
NLSCY variables used can be provided by the authors on request.
References
1. Kendall PC, Ronan KR. Assessment of children’s anxieties, fears, and phobias:
cognitive-behavioral models and methods. In: Reynolds CR, Kamphaus RW,
editors. Handbook of psychological and educational assessment of children. Vol
2. New York (NY): Guilford Press; 1990. p 223–244.
2. Cartwright-Hatton S, McNicol K, Doubleday E. Anxiety in a neglected
population: prevalence of anxiety disorders in pre-adolescent children. Clin
Psychol Rev. 2006;26:817–833.
3. Ialongo N, Edelsohn G, Werthamer-Larsson L, et al. Social and cognitive
impairment in first-grade children with anxious and depressive symptoms. J
Clin Child Psychol. 1996;25:15–24.
WLa Revue canadienne de psychiatrie, vol 54, no 5, mai 2009328
Original Research
Table 3 continued
With unadaptable temperament at ages 2 to 3 years as a predictor With fussy–difficult temperament at ages 2 to 3 years as a predictor
Outcome variable, age in years
(Weighted na)
Predictor
variable B tPR
2(P)
Predictor
variables B tPR
2(P)
Child-reported anxiety
10 to 11 (579 200) (Constant) 1.36 12.62 <0.001 (Constant) 1.35 11.35 <0.001
Child sex 0.07 1.73 0.08 Child sex 0.08 1.74 0.08
Child age 0.09 2.19 0.03 Child age 0.09 2.18 0.03
SES –0.02 0.82 0.41 SES –0.02 0.81 0.42
Unadaptable 0.00 0.20 0.84 0.014 (0.05) Fussy–difficult 0.00 0.04 0.97 0 .014 (0.05)
All tand R2values are derived from calculations using bootstrap weights. All predictor variable data were collected at Cycle 1, when children were aged 2 to 3 years.
aWeighted using Cycle 5 longitudinal funnel weights
bBoys = 0; Girls = 1
cChild age in years (either 2 years or 3 years)
4. Kessler RC, Berglund P, Demler O, et al. Lifetime prevalence and age-of-onset
distributions of DSM-IV disorders in the National Comorbidity Survey
Replication. Arch Gen Psychiatry. 2005;62:593– 602.
5. Costello EJ, Mustillo S, Erkanli A, et al. Prevalence and development of
psychiatric disorders in childhood and adolescence. Arch Gen Psychiatry.
2003;60:837–844.
6. Kendall PC. Treating anxiety disorders in children: results of a randomized
clinical trial. J Consult Clin Psych.1994;62:100 –110.
7. Spence SH. Prevention strategies. In: Vasey MW, Dadds MR, editors. The
developmental psychopathology of anxiety. New York (NY): Oxford University
Press; 2001. p 325–351.
8. Donovan CL, Spence SH. Prevention of childhood anxiety disorders. Clin
Psychol Rev. 2000;20:509–531.
9. Keiley MK, Bates JE, Dodge KA, et al. Effects of temperament on the
development of externalizing and internalizing behaviors over 9 years.
In: Columbus F, editor. Advances in psychology research. Vol 6. Huntington
(NY): Nova Science Publishers; 2001. p 255–288.
10. Rothbart MK, Bates JE. Temperament. In: Damon W, series editor, Eisenberg N,
volume editor. Handbook of child psychology: social, emotional, and personality
development. Vol 3. 5th ed. New York (NY): John Wiley & Sons, Inc; 1998.
p 105–176.
11. Bates JE, Bennett Freeland CA, Lounsbury ML. Measurement of infant
difficultness. Child Dev. 1979;50:794–803.
12. Kagan J, Snidman N. Infant predictors of inhibited and uninhibited profiles.
Psychol Sci. 1991;2:40–44.
13. Schwartz CE, Snidman N, Kagan J. Adolescent social anxiety as an outcome of
inhibited temperament in childhood. J Am Acad Child Adolesc Psychiatry.
1999;38:1008–1015.
14. Hirshfeld DR, Rosenbaum JF, Biederman J, et al. Stable behavioral inhibition
and its association with anxiety disorder. J Am Acad Child Adolesc Psychiatry.
1992;31:103–111.
15. Muris P, Meesters C, Spinder M. Relationships between child- and
parent-reported behavioural inhibition and symptoms of anxiety and depression
in normal adolescents. Pers Individ Dif. 2003;34:759–771.
16. Biederman J, Hirshfeld-Becker DR, Rosenbaum JF, et al. Further evidence of
association between behavioral inhibition and social anxiety in children. Am J
Psychiatry. 2001;158:1673–1679.
17. Rothbart MK, Ahadi SA, Hershey KL. Temperament and social behavior in
childhood. Merrill Palmer Q. 1994;40:21–39.
18. Keiley MK, Lofthouse N, Bates JE, et al. Differential risks of covarying and pure
components in mother and teacher reports of externalizing and internalizing
behavior across ages 5 to 14. J Abnorm Child Psychol. 2003;31:267–283.
19. Hagekull B. Infant temperament and early childhood functioning: possible
relations to the five-factor model. In: Halverson CF Jr, Kohnstamm GA, Martin
RP, editors. The developing structure of temperament and personality from
infancy to adulthood. Hillsdale (NJ): Erlbaum; 1994. p 227–240.
20. Prior M, Smart D, Sanson A, et al. Does shy-inhibited temperament in childhood
lead to anxiety problems in adolescence? J Am Acad Child Adolesc Psychiatry.
2000;39:461–468.
21. Caspi A, Henry B, McGee RO, et al. Temperamental origins of child and
adolescent behavior problems: from age three to age fifteen. Child Dev.
1995;66:55–68.
22. Statistics Canada. National Longitudinal Survey of Children and Youth, Cycle
5—Microdata User Guide. Ottawa (ON): Statistics Canada, Special Surveys
Division; 2003.
23. Keiley MK, Bates JE, Dodge KA, et al. A cross-domain growth analysis:
externalizing and internalizing behaviors during 8 years of childhood. J Abnorm
Child Psychol. 2000;28:161–179.
24. Ford T, Goodman R, Meltzer H. The British Child and Adolescent Mental
Health Survey 1999: the prevalence of DSM-IV disorders. J Am Acad Child
Adolesc Psychiatry. 2003;42:1203–1211.
25. Statistics Canada. National Longitudinal Survey of Children and Youth, Cycle
1—Codebook. Release 3. Ottawa (ON): Statistics Canada, Special Surveys
Division; 1998.
26. Phillips O. Using bootstrap weights with Wes Var and SUDAAN. Res Data
Cent Inf Tech Bull. 2004;1:6–15.
27. Willms JD, Fedick CB. The CRISP-NLSCY files: SPSS data and syntax files
for use with the National Longitudinal Survey of Children and Youth.
Fredericton (NB): Canadian Research Institute for Social Policy/University of
New Brunswick [distributor]; 2003.
28. Loewenthal KM. An introduction to psychological tests and scales. London
(GB): UCL Press; 1996.
29. Boyle MH, Offord DR, Racine Y, et al. Evaluation of the Original Ontario
Child Health Study scales. Can J Psychiatry. 1993;38:397–405.
30. Boyle MH, Offord DR, Hofmann HG, et al. Ontario Child Health Study:
I. Methodology. Arch Gen Psychiatry. 1987;44:826–831.
31. Boyle MH, Offord DR, Racine YA, et al. Adequacy of interviews vs checklists
for classifying childhood psychiatric disorder based on parent reports. Arch Gen
Psychiatry. 1997;54:793–799.
32. Boyle MH, Offord DR, Racine YA, et al. Interviews versus checklists:
adequacy for classifying childhood psychiatric disorder based on adolescent
reports. Int J Methods Psychiatr Res. 1996;6:309–319.
33. StataCorp. Stata Statistical Software: Release 9. College Station (TX):
StataCorp LP; 2005.
34. Sribney B. Estimating correlations with survey data. Stata FAQs [Internet].
College Station (TX): StataCorp LP; 2001 [updated 2005; cited 2007 May 4].
Available from: http://www.stata.com/support/faqs/stat/survey.html.
35. Morris TL, Hirshfeld-Becker DR, Henin A, et al. Developmentally sensitive
assessment of social anxiety. Cogn Behav Pract. 2004;11:13–28.
36. Warren SL. Anxiety disorders. In: DelCarmen-Wiggins R, Carter A, editors.
Handbook of infant, toddler, and preschool mental health assessment. New
York (NY): Oxford University Press; 2004. p 355–375.
37. Ferdinand RF, van Lang NDJ, Ormel J, et al. No distinctions between different
types of anxiety symptoms in pre-adolescents from the general population.
J Anxiety Disord. 2006;20:207–221.
38. Rapee RM. The development and modification of temperamental risk for
anxiety disorders: prevention of a lifetime of anxiety? Biol Psychiatry.
2002;52:947–957.
Manuscript received May 2008, revised, and accepted August 2008.
Portions of this research were presented at the 68th Annual Convention of
the Canadian Psychological Association, 2007 June, in Ottawa, Ontario.
1Student, Department of Psychology, Dalhousie University, Halifax, Nova
Scotia.
2Assistant Professor, Department of Psychiatry, Dalhousie University,
Halifax, Nova Scotia; Staff Psychiatrist, IWK Health Centre, Halifax,
Nova Scotia.
3Associate Professor, Departments of Pediatrics and Psychology,
Dalhousie University and IWK Health Centre, Halifax, Nova Scotia.
4Professor, Departments of Psychiatry and Psychology, Dalhousie
University, Halifax, Nova Scotia.
Address for correspondence: Ms V Grant, Department of Psychology,
Life Sciences Centre, Dalhousie University, Halifax, Nova Scotia
B3H 4J1; Valerie.Grant@dal.ca
Early Temperament Prospectively Predicts Anxiety in Later Childhood
The Canadian Journal of Psychiatry, Vol 54, No 5, May 2009 W329
WLa Revue canadienne de psychiatrie, vol 54, no 5, mai 2009330
Original Research
Résumé : Le tempérament précoce prédit prospectivement l’anxiété ultérieure dans
l’enfance
Objectif : Investiguer la contribution des constructs du tempérament de la petite enfance
correspondant à 2 sous-types d’émotionnalité négative générale — la détresse angoissée
(tempérament inadaptable) et la détresse irritable (tempérament tatillon-difficile) — à l’anxiété
ultérieure dans un échantillon nationalement représentatif.
Méthode : À l’aide de multiples analyses de régression linéaire, nous avons vérifié l’hypothèse que
les échelles de tempérament inadaptable et de tempérament tatillon-difficile déclarées par un
soignant pour des enfants de2à3ans(en1995) prédiraient prospectivement les symptômes
d’anxiété déclarés par un soignant chez l’enfant de4à5ans, de6à7ans, de8à9ans, et de 10 à
11 ans, et l’anxiété déclarée par l’enfant de 10 à 11 ans (contrôle selon le sexe, l’âge, et le statut
socioéconomique) dans un échantillon nationalement représentatif de l’Enquête longitudinale
nationale sur les enfants et les jeunes de Statistique Canada (ninitial pondéré = 768 600).
Résultats : Seul le tempérament tatillon-difficile prédisait l’anxiété chez les enfants de6à7ans.
Dans des régressions distinctes, le tempérament inadaptable et le tempérament tatillon-difficile
prédisaient chacun l’anxiété de8à9ans, mais quand les deux étaient entrés simultanément, seul le
tempérament inadaptable demeurait un prédicteur marginal. Le tempérament ne prédisait pas
significativement l’anxiété déclarée par le soignant ou l’enfant de 10 à 11 ans, ce qui suggère que
lorsque les enfants grandissent, les facteurs environnementaux peuvent devenir des contributeurs à
l’anxiété plus importants que le tempérament précoce.
Conclusion : Les résultats actuels offrent la première démonstration que le tempérament précoce
est lié à l’anxiété ultérieure dans l’enfance, dans un échantillon nationalement représentatif.