Abstract
Epidemiological data indicate that 15–20% of school age children suffer relatively serious behavioural and emotional problems which significantly compromise their everyday functioning in multiple domains [1, 2]. Development of any disorder will be through a combination of intrinsic child factors such as intelligence, temperament and biological risk, and family/environmental factors such as socioeconomic status (SES), family adjustment factors, environmental stressors, and protective influences [3–6]. A number of risk factors need to accumulate to affect the development of a disorder [7]. Patterson and his coworkers [8] in studying the genesis of antisocial behaviours or conduct disorder in children, have tested social learning theories and models for the development of such behaviours and concluded that coercive interchanges between child and family members, parental discipline and monitoring practices, and associated factors including school achievement and peer relationships were important.
A variable which is acknowledged, but remains untested in the Patterson et al. [8] models, is child temperament. Researchers such as Thomas and Chess [9], Bates et al. [10] and Maziade [11] have demonstrated the influence of temperament characteristics on behavioural dysfunction. Early studies emphasized the risk factor of ‘difficult temperament’ (a combination of negative mood, non-adaptability, social withdrawal and high intensity) [9, 12] in psychosocial development, in interaction with poor ‘fit’ between the characteristics of the child and the characteristics of the family and the social environment. Studies in the last decade have focused on specific temperament features such as ‘inhibited temperament’, or ‘negative emotionality’ and their associations with particular types of disorder [13, 14], such as internalizing or externalizing problems. Our longitudinal research [12, 15] in the Australian Temperament Project (ATP) has focused particularly on child temperament as a major influence on behavioural adjustment, both as an intrinsic and relatively stable child characteristic and through its bidirectional influence on interpersonal interaction across developments [16]. In addition, we have examined parental characteristics that are hypothesized to enhance or inhibit healthy development in interaction with child attributes [17, 18].
The Australian Temperament Project
The ATP is a prospective, longitudinal study of the psychosocial adjustment of Australian children, involving a large representative cohort of Victorian children from infancy to adolescence. Follow up of these children and their families to the pre-adolescent stage in the present study has allowed the investigation of predictors of adjustment, using a consistent and relevant set of temperament, behavioural, family, and school variables across the first 11–12 years of life. In a previous study we reported on types of psychological disorders based on individual interviews using the Child Assessment Schedule [19] in ATP children aged 11–12 years who had been screened via parent, teacher, and child questionnaire as at-risk for behaviour disorder. Results indicated that almost half of these at-risk children met criteria for DSM-IIII-R diagnoses of various kinds, with a substantial degree of comorbidity. We compared the characteristics of at-risk children who did, or did not, qualify for a diagnosis and found minimal differences between at-risk children with and without a diagnosis, on any concurrent intrinsic or extrinsic variables included in the study. However, the at-risk children differed significantly on almost every variable from control (no problem) children, including lower SES, less optimal parental–child rearing factors, lower social competence, poorer academic success, and poorer peer and family relationships [18]. This study also showed that temperament factors such as negative emotionality, high activity, poor persistence and low approach/sociability were risk factors for maladjustment. A notable proportion of the at-risk children had a history of difficult temperament dating back to early childhood [20].
Our major aim here was to move on from the crosssectional comparison study to investigate the predictors of psychological disorder in this community sample of 11–12-year-old children, and to assess the stability of adjustment status across development. Taking as our basic outcome variable adjustment status at 11–12 years, we report early contributions to current dysfunction, which we have defined operationally via scores on behaviour problem scales. Our at-risk (clinical) group was compared with a group who did not have behaviour problems. Since psychiatric status based on diagnostic interviews added minimally to group discrimination, our longitudinal comparisons were based on the entire at-risk group, rather than only the formally diagnosed subgroup [18].
On the basis of the literature, especially that emphasizing the influence of temperamental attributes as both ‘protective’ and ‘risk’ factors [6, 21], and on the findings of our previous analyses with ATP children at 7–8 years of age [17] when the temperamental factor of ‘inflexibility’ was the best predictor of maladjustment, we hypothesized that temperament would be a powerful longitudinal predictor of adjustment at the pre-adolescent stage.
Our specific aims were:
1. To examine the stability of behavioural adjustment status from infancy onwards, by looking back across the years at behavioural adjustment as measured at approximately 18 month intervals;
2. To identify predictors of psychosocial maladjustment including intrinsic child characteristics such as temperament, behaviour, early school functioning and interpersonal skills, as well as family factors such as SES, life stresses and adversity, and mother–child relationship;
3. To determine the age from which it is possible to predict dysfunction, by examining the predictive power of data that we have collected from infancy onwards;
4. To investigate gender differences in the weighting and nature of the factors which contribute to dysfunction. Based on our earlier work [17] we hypothesized that there would be sex differences in the strength, rather than the nature, of the various behavioural, temperamental, social and school related indices. Temperament characteristics such as inflexibility and low persistence were predicted to be more powerful for boys than girls, and early academic problems would be more salient for boys. We expected family factors to be more important for girls than boys.
Method
Subjects
Details of the subject population for this study are summarized in Prior et al. [18]. In brief, the subjects for the analyses reported here were selected from the ATP, a large and representative sample of 2443 children drawn from urban and rural areas in the state of Victoria, Australia. Families have been followed up in surveys at approximately 18 month intervals since the first year of life. At the time of this study the children were aged between 11 and 12 years, and most were in the final year of primary school. Around 70% (n = 1724) of the original cohort were still enrolled in the study at this stage. Although there was a slight overrepresentation of low SES and ethnic families among those lost, there were no significant differences between the retained and lost/withdrawn samples on any child characteristic measured at infancy, and the SES profiles of the original and retained samples were very similar. Budget constraints and the dispersion of the sample across urban and rural areas (as well as interstate and overseas), led to a decision to assess a subsample of approximately 300 subjects, two-thirds of whom had adjustment difficulties. All children in this study had IQs in the normal range, as assessed with a short form of the Weschler Intelligence Scale for Children – Revised (comparison group mean = 111.95, SD = 13.84; at-risk group mean = 105.76, SD = 12.08). Participants were selected via the following criteria, based on parent, teacher and child ratings of behaviour problems:
At-risk group
Subjects for the at-risk group were selected if they scored 1 SD or more above the mean on one or more of the hostile-aggressive (HA), hyperactive-distractible (HD) or anxious-fearful (AF) subscales of the Rutter Child Behaviour Questionnaire [22] parallel forms, by two or more informants (parent, teacher or child self-report). Slightly different cut-offs were used for girls to allow for the fact that girls’ externalizing behavioural problems tend to differ from those of boys [23]. For externalizing types of behaviour problems (HA or HD), a parent report of significant problems was required as an essential criterion; for internalizing problems (AF), child report was required as an essential source. The following criteria were also applied: (i) all children reported to have behavioural problems by all three sources were selected; and (ii) the most serious cases who were above the cut-off by two sources were selected. A total of 186 clinically at-risk children (103 boys, 83 girls) were selected.
Comparison group
The comparison group participants were selected if they were no more than 0.5 SD above the mean on any behavioural problem scale by any of the three sources; and were matched with clinical subjects for sex and residence locality (metropolitan, outer-urban, large provincial city, or rural area). This latter matching criterion allowed some control of SES and school experience factors. Ninety-six subjects from a pool of 323 meeting matching criteria, all with IQ estimates of 80 or above, were selected (48 boys and 48 girls).
Details of types and rates of diagnoses in this sample can be found in Prior et al. [18]. In summary, 47.3% of clinically at-risk children (60% of boys and 35% of girls), and 9.3% of comparison group children received a DSM III-R diagnosis based on child interview. Internalizing diagnoses were most common, and 44% of diagnosed children had two or more disorders.
Measures
Outcome measure of adjustment at age 12
At age 12, outcome measures of adjustment used were the Rutter Child Behaviour Questionnaires – Scales A and B (Parent and Teacher Forms) (PBQ) [22]; plus a child report form developed by the authors by adaptation of the items in the parent questionnaire. The PBQ provides HA, AF and HD subscale scores as well as a total behaviour problem score. Acceptable reliability and validity data have been reported by Elander and Rutter [24].
Prediction measures
Prediction measures had been collected from the ATP cohort via regular surveys of temperament, behaviour, academic progress and family factors from infancy onwards [20] and are summarized in Table 1.
Summary of measures
Results
Stability of behaviour problems
Stability of behaviour problems was assessed by summing the number of times behavioural problem ratings fell in the clinical range (defined as a score of 1 SD above the mean or greater) on the parentreported behaviour problem measures for each child, from infancy to 9–10 years of age. This entailed giving each child a score of 0 to indicate absence of problems or 1 to indicate presence of problems, at each time point, with a possible range of summed scores of 0–6. Where there were several behaviour problem measures at the one time point (e.g. the HA, HD or AF scales at 3–4, 5–6, 7–8 and 9–10 years), a score of 1 was given if one or more of the three scores was in the clinical range.
As can be seen from Table 2, over 80% of the at-risk children in this study had been in the clinical range of scores on at least one occasion (other than at selection), and over 50% were in the clinical range on half the surveys or more. Conversely, almost two-thirds of comparison children were never in the clinical range, and less than 10% were in the clinical range on three or more occasions. At-risk boys were more likely to be in the clinical range than at-risk girls (χ2 = 14.11, df = 6, p < 0.03). There were no differences between the at-risk children who had received a DSM-III-R diagnosis on the Child Assessment Schedule [18, 19] and those who did not qualify for a diagnosis, in the stability of behaviour problems from infancy to 9–10 years (χ2 = 6.2, df = 6, NS).
Frequency of placement in clinical range from infancy to 9–10 years of age
Calculation of odds ratios across the years (Table 3) indicated that, as expected, the closer in age the predictive variables were to concurrent status, the higher the chance of persistent disorder. For example, the presence of disorder at 7–8 years gives a 10-fold chance that disorder will be present at 11–12 years, and at 9–10 years gives a 15-fold risk. By comparison, the prediction from infancy data is much weaker. The stability of risk, with a consistent fivefold increase in risk from early toddlerhood onwards, is clearly substantial.
Odds ratios and confidence limits for the association between behavioural problems from infancy to 9–10 years, and behavioural problems at 11–12 years
Comparison across time
To detect the timing and nature of any emerging differences related to the development of adjustment problems, and bearing in mind the need to limit the number of statistical analyses performed (i.e. to guard against type 1 errors), a series of MANOVA analyses was conducted, examining the data year by year, and separately by source of report. Following a significant MANOVA result, univariate F-tests were computed to assess the contribution of the individual variables. We then calculated effect sizes, as described by Cohen [32], for each significant variable as a strategy of assessing their relative contribution, and to allow for the analysis of sex differences. With regard to analysis of variance tests, following Cohen [32], an effect size of 0.10 is defined as small, 0.25 as medium, and 0.40 as a large effect.
Significant results are broadly grouped into the preschool period (summarized in Table 4), parent-report of the primary school years (Table 5) and teacher-report of the primary school years (Table 6), with details of overall group differences, and differences separately by sex. Significant differences were consistently in the direction of more difficulties and problems for the at-risk children. Variables for which significant differences were not found are listed beneath the tables. Univariate differences were not reported if the overall MANOVA was not significant.
MANOVA (bold), univariate results and effect sizes comparing at-risk and comparison groups from infancy to 3–4 years of age
MANOVA (bold), univariate results and effect sizes comparing at-risk and comparison groups at age 5 to 10 years — parent report
MANOVA (bold), univariate results and effect sizes (ES) comparing At-Risk and Comparison groups at 5-6 and 7-8 years of age – Teacher Report
Infancy to 3–4 years of age
Significant differences between the at-risk and comparison groups overall were consistently apparent (Table 4). Medium effect sizes were first evident at 1–3 years and were persistent from 3 to 4 years of age. At 1–3 years, the presence of behavioural problems and specific temperament measures differentiated the groups, and the significant temperament factors tended to be particularly relevant for interpersonal and family relationships (e.g. high irritability, high reactivity). At 3–4 years, group differences were found on both externalizing (hyperactivity, aggression) and internalizing (anxious, withdrawn) types of behaviour problems, and the temperament factor of inflexibility.
At-risk and comparison girls differed significantly from 1 to 3 years onwards, with consistent medium effect sizes being obtained. At 1–3 years, univariate differences were noted on behaviour problem scales and on the temperament factors of irritability and reactivity. At 3–4 years, medium effect sizes were found for all types of behavioural problems and the temperament factor of persistence. Significant differences between at-risk and comparison boys were not apparent until 3–4 years of age. Medium effect size differences were found on HA and HD problems and the temperament factor of inflexibility. There were weaker differences on AF problems and on persistence. These internalizing behavioural problems were more powerful differentiators for girls than for boys (effect sizes 0.35 vs 0.21).
Parent-reported data at 5–6, 7–8 and 9–10 years of age
By school age, significant differences between overall groups were evident in a range of areas (Table 5), with large effect sizes for externalizing behavioural problems at the three time points, and medium to large effect sizes for internalizing behavioural problems. Differences were consistently found on mother's overall rating of the child, on the temperament factors of inflexibility and persistence at 5–6 and 7–8 years, and emotionality at 9–10 years. These were predominantly in the medium effect size range, with some large effect sizes evident also. Ratings of the child's confidence in social situations differentiated at-risk and comparison groups at 9–10 years (medium effect size). There were significant, but relatively weak, differences on parental perceptions of life difficulty and coping abilities when the child was 7–8 years of age (effect sizes 0.19 and 0.18). Parent-reported family stress, measured as the total number of negative life stresses, did not significantly differentiate between groups.
These results were generally consistent across the sexes, although as hypothesized, there were effect size differences for girls and boys. Anxious-fearful behavioural problems consistently differentiated at-risk and comparison girls more strongly than boys (effect sizes 0.40 vs 0.25 at 5–6 years, 0.44 vs 0.34 at 7–8 years, and 0.56 vs 0.30 at 9–10 years). Inspection of the various group means suggests that this sex difference may be a reflection of the relatively low level of AF behavioural problems among comparison girls, rather than an elevated level of behavioural problems among at-risk girls, since at-risk boys and girls consistently had similar levels of AF behavioural problems.
Greater temperamental reactivity/emotionality differences between at-risk and comparison girls compared with boys were evident at 9–10 years (effect sizes 0.40 vs 0.27; also noted at 13 years). Social skills at 9–10 years were stronger differentiators for girls, with medium effect sizes on social confidence and empathic skills only for girls.
Teacher-report data at 5–6 and 7–8 years of age
Teacher data supported the picture of widespread and enduring differences between the at-risk and comparison groups, with all variables being significantly different. Differences were generally in the small and medium effect size range. Unlike the corresponding parent data, there was a trend for consistently stronger differences on temperament (the majority with medium effect sizes) than behavioural problems (all with small effect sizes). A teacher's overall rating of the child's temperament, similarly to mother's overall rating, revealed substantial differences at both time points (effect sizes 0.31 and 0.30). Ratings of school readiness at 5–6 years and academic skills at 7–8 years differentiated the groups, both with medium effect sizes. There were indications of more limited interpersonal skills among the at-risk group, according to teacher ratings at 7–8 years (a trend also noted in parent-reported data at 9–10 years), although the effect size was small (0.22).
Teacher ratings of boys at 5–6 years revealed significant group differences in behaviour in the school environment, such as temperamental task orientation, teacher's overall rating of temperamental difficulty and school readiness. By 7–8 years, significant differences between at-risk and comparison boys were found on all measures except AF problems. Medium effect sizes were found at both time points on HA and HD behaviour problems, the temperament factor of task orientation, and academic factors (school readiness, reading). Teacher ratings of girls at 5–6 years revealed significant differences on all measures, and likewise at 7–8 years, with the exception of HD problems. There were stronger differences for girls than boys on measures assessing internalizing aspects of behaviour (e.g. on AF behavioural problems) at 7–8 years, and the temperament factors of reactivity (7–8 years) and flexibility (both time points), with medium effect sizes for girls but not boys. Group differences were stronger for boys than girls on HD problems (medium effect size) and task orientation at 7–8 years (effect sizes 0.47 vs 0.27).
It could be argued that the substantial and consistent differences reported above are a result of comparing extreme groups of poorly and highly functioning children. To clarify this point, a further set of MANOVAs was conducted in which a third group consisting of a randomly drawn, gender-balanced sample of children from the ATP cohort was included. Three planned contrasts, random versus comparison, random versus at-risk, and random and comparison groups combined, versus the at-risk group, were conducted. Significant differences between the selected comparison group and the random ATP group were not found at any time point. Thus the at-risk group clearly exhibited persistently more maladaptive behaviour patterns over a range of domains and environments from the preschool period onwards and our findings could not be attributed to extreme group comparisons.
Group classification analyses
Discriminant function analyses were used to investigate the degree of classification accuracy achievable using the ATP longitudinal data set in distinguishing between at-risk and comparison group children. We did this separately for parent-derived and teacher-derived variables. As for the MANOVAs, analyses were conducted comparing the overall groups and the sexes separately, using variables from earlier years. Parent-reported total behavioural problems, temperament factors, mother's overall rating from 5 to 6 years onwards, and social skills at 9–10 years, were selected as the independent variables for the first set of discriminant function analyses (DFAs). Sixty-three subjects (28 boys and 35 girls) had some missing data and were not included; t-tests indicated no significant differences between included and excluded children on temperament, behavioural problems, IQ, social skills and family background at 11–12 years. Classification accuracy was around 80% for all three analyses. This is a significant improvement over chance, according to Press's Q statistic [33]. The at-risk group was always more accurately classified than the comparison group, with a range of 82.1% to 87.3% correct classifications. Just over two-thirds of comparison group boys, and of children overall, were correctly classified, while almost 80% of comparison girls were accurately classified. Total behaviour problem scores at the various ages, followed by mother's overall rating, were the most powerful discriminators. Temperament factors, especially inflexibility, also contributed significantly.
Teacher-reported total behavioural problems, temperament factors and teacher's overall rating of child difficulty at 5–6 and 7–8 years, school readiness at 5–6 years and academic and social skills at 7–8 years, were entered as the independent variables for the second set of DFAs. Just over 40% of subjects (66 boys and 62 girls) had some missing data and were excluded. However t-test analyses comparing included and excluded children revealed no significant differences on a range of child and family characteristics at 11–12 years.
Classification rates ranged from 70% for the overall groups to 79% for at-risk and comparison girls. As was found for the parent DFAs, higher rates of classification accuracy were noted for the at-risk groups (83% to 86%) than for the comparison groups (a range of 46% to 65%). The much higher misclassification rate for comparison group children overall (46%) than for comparison boys or girls separately (both about 65%) suggests that the classification of comparison group children may have been limited by combining the data across the sexes. All variables discriminated significantly, with the exception of the temperament factors of reactivity and flexibility for boys. Task orientation and academic factors contributed most strongly for boys, while flexibility, reactivity and total behavioural problems contributed most strongly for girls. Teacher's overall rating of the child's temperament also loaded substantially, while the social skills measure was a significant but weaker contributor for both sexes.
Discussion
Budget constraints in this project, which has a very large sample dispersed across urban and rural areas as well as interstate and overseas, did not permit an indepth study of the entire cohort. We adopted the strategy of selecting the most troubled and troubling children in the sample at the age of 11–12 years for further study, using behavioural adjustment ratings provided by three sources: parent, teacher and child self-report. We then selected a non-problem group, matching on child sex and urban–rural location which we showed did not differ from a random sample of ATP children.
Stability
For most children who were rated as having behavioural or emotional dysfunction at the pre-adolescent stage, their difficulties had clearly been present for many years. Although this group did move in and out of the clinical range in behavioural adjustment ratings across the years of measurement, most showed signs of significant dysfunction at some periods during childhood. Lahey et al. [34] also found similar consistency within fluctuating levels across time in conduct-disordered boys. Only 17% in our sample were behaviour disordered for the first time in their history at 11–12 years of age. Hence the degree of stability of dysfunction was substantial. The data also highlight the desirability of incorporating several measurement points in assessing adjustment status [35], to provide a more reliable estimate and to guard against inclusion of children with transitory difficulties, while maximizing the identification of those with persistent problems. Relying on only one point would have resulted in a considerable false positive and false negative rate of risk categorization.
Predictors of adjustment
The major predictor of current dysfunction is previous dysfunction; hence in most of our analyses the most powerful differences between the clinical and comparison groups across time were various kinds of behavioural problems. Group differences were apparent in the first 3 to 4 years of life, especially in the reported presence of behavioural problems of all types. Socioeconomic status was related to dysfunction, with lower SES characteristic of the behaviour problem group. Nevertheless, previous analyses [18] controlling for SES showed that the substantial clinical differences were robust. Temperamental differences, although significant, were relatively weaker differentiators between the groups. Differences between clinical and comparison groups increased substantially as the children grew older, encompassing behavioural adjustment status, temperamental inflexibility and low persistence, and mother's overall child difficulty rating. Those temperament dimensions which exerted persistent effects on adjustment across the school age period (i.e. inflexibility, poor task persistence) may be conceived of as indicators of poor self-regulation [14] which put the child at risk of dysfunction in a range of milieus which are important at this developmental stage. Emotionality, assessed at 9–10 years and representing an index of emotional selfregulation, was also a significant differentiator between clinical and comparison group children. The importance of temperament as a contributor to behavioural maladjustment was confirmed for both sexes.
The only family factors which featured in the group comparisons were parental life difficulty and perceived coping skills at 7–8 years, although the effect sizes were weak (0.19 and 0.18). This may relate to the fact that although there are a number of stressed and disadvantaged families in this community, overall the cohort is relatively well functioning, and few suffer the extremes of disadvantage which are reported in some of the USA studies [8, 36].
One of the criticisms of this kind of research is that the stability of clinical status may be a factor of a stable mother perception of her child as a problem, hence diminishing the validity of such findings. Multiple-source data are essential to address this problem. We highlight the fact that teacher data provided converging results with those derived from mother report, even though children were rated by different teachers at each time point. Teacher's overall rating of the temperamental difficulty of the child featured quite strongly at each age for both sexes, and the teacher temperament factor of task orientation (which has strong conceptual links with the parentrated factor of persistence) was predictive, especially for boys. Three school related variables, ‘readiness’ for school in the first year, and reading and academic skills in the third year, were predictive particularly for boys. These findings very much confirm those reported previously [17] from the ATP, with the data suggesting that teachers perceived more negative characteristics in problem boys by comparison with girls (except for teachers’ rating of child difficulty), and that their evaluations are in part related to their appraisal of academic progress.
Sex differences were in the strength rather than the nature of effects across the years. Internalizing behavioural problems and social skills, consistently over both parent and teacher report, were stronger differentiators for girls than boys; while hyperactive behavioural problems, temperamental characteristics such as inflexibility and persistence, and academic factors were more powerful discriminators for boys. Family factors played a weak predictive role for both sexes.
Discrimination between at-risk and comparison groups
Using parent-derived data, the classification accuracy of group status at 11–12 years of age of around 80% was substantial, particularly when the sexes were considered separately, and particularly for the at-risk group. Previous behavioural problems were consistently the best discriminators, with mother's rating of child difficulty making a consistent but weaker contribution. The temperament dimensions reflecting self-regulation capacities also contributed, confirming the importance of these factors as longitudinal predictors of disorder [14].
The classification accuracy of around 70% using teacher-derived data was somewhat lower, but still substantial. Previous behavioural problem status was a less salient predictor by comparison with the parent data classification. The classroom-related factors of task orientation, academic skills and school readiness featured in these equations, particularly for boys, and social skills also made some contribution.
There is a paucity of long-term data concerning the influence of temperament on the development of dysfunction in large community samples. The New Zealand and Canadian studies [23, 37, 38], which are perhaps our closest comparison, have not specifically considered temperament, although recent reports from these groups using retrospectively constructed temperament factors from earlier measures have confirmed their influence [35, 39]. Maziade et al. [40] have directly assessed the association of difficult temperament (based on the Thomas and Chess model), with behavioural adjustment in a longitudinal study of a French-Canadian sample of children. They report clear associations between non-adaptable, negative, emotionally intense and withdrawing temperamental attributes in early life, and the presence of clinically diagnosed disorders at 12 years of age. Maziade suggested that difficult temperament leads to behavioural disorder in association with numerous environmental risk factors such as the dysfunctional parental control factor which has also been highlighted by the research of Patterson et al. [8] and Bates [41]. In regard to Patterson et al.'s acknowledgement of the likely importance of temperament factors in the development of externalizing disorders, our research provides empirical support for their suggestions.
Conclusion
Behaviour problems are established at an early age, and appear relatively persistent over the elementary school years for a substantial proportion of children. Few argue these days about the importance of early identification and intervention for at-risk children, although many are sceptical about the accuracy of prediction from early to later disorder. The data presented here show that there are factors which are apparent early and are clearly predictive. Moreover, the number of false positives in identifying risk status is sufficiently small, especially if the sexes are considered separately, and if ratings are obtained at several points in time, to allow confidence in identifying children and families in need of assistance at an early stage of development.
Footnotes
Acknowledgements
This research was supported by grants from the National Health and Medical Research Council of Australia. Many thanks to the participating children, their parents and teachers for their loyal participation in the project over the years.