Childhood Trajectories of Inattention and Hyperactivity and Prediction of Educational Attainment in Early Adulthood: A 16-Year Longitudinal Population-Based Study

In 1986–1987, a representative sample (N=6,397) of kindergarten children in the French-speaking schools of the province of Québec was selected. Both teachers and mothers completed the Social Behavior Questionnaire for 3,715 of these children (23). Of these 3, 715 children, 2,000 (1,001 boys) were randomly selected for participation in the present study and followed longitudinally. Information on high school graduation was obtained from the Québec Ministry of Education official records and was available for the whole sample. Among the 2,000 study participants, 32.1% did not graduate from high school relative to 34.1% among the initial representative sample. Although statistically significant (χ²=5.36, df=1, p=0.02), this difference is clearly low in magnitude. Additional information regarding attrition and missing data for the 2,000 participants followed longitudinally is presented in this report. After complete description of the study, written consent was obtained at each wave of data collection from the mothers (including consent regarding teachers' reports).

Teachers rated children with the Social Behavior Questionnaire each year from kindergarten to sixth grade, providing seven assessment points from the age of 6 to 12 years. (In Québec, within this age range a teacher teaches only at one level, and thus the assessments were made by a different teacher each year.) The Social Behavior Questionnaire is based on the Children's Behavior Questionnaire (24) and the Preschool Behavior Questionnaire (25), which both demonstrate good psychometric properties, particularly test-retest reliability. These results for test-retest reliability have been replicated with the Social Behavior Questionnaire (23). Furthermore, the Social Behavior Questionnaire was used in several studies of large sample cohorts that documented its predictive validity on a range of adolescent and adult outcomes, particularly for hyperactivity and inattention (12, 21, 22, 26 [also see the data supplement accompanying the online version of this article]). We assessed hyperactivity with the following two items from the scale: 1) restless, runs about, or jumps up and down, does not keep still and 2) squirmy, fidgety child (Cronbach's alpha for the seven assessments: 0.84–0.89). The following four items were used to assess inattention: 1) weak capacity for concentration, cannot maintain his or her attention for a long time on the same task; 2) easily distracted; 3) absentmindedness; and 4) gives up easily (Cronbach's alpha for the seven assessments: 0.85–0.90). Each item for both dimensions was scored from 0 to 2.

The measure of educational attainment differentiated participants who had a high school diploma at 22–23 years of age (coded as 0) and those who did not (coded as 1). The latter category represented 32.1% of the participants (and included individuals who dropped out of school [17.2%] and individuals receiving vocational and adult education [14.9%]).

Initial individual-level confounding variables were sex (coded as 0 for girls and 1 for boys) and birth weight (coded as 1 for low birth weight [under 2,500 g, N=6.7%] and 0 otherwise). Socioeconomic variables were as follows: years of schooling of each parent, assessed when the child was 6 years old (mother: mean=11.97 years [SD=2.56], father: mean=12.17 years [SD=3.42]); occupational socioeconomic index of each parent, assessed when the child was 6 years old and based on criteria by Blishen et al. (27) (mother: mean rating=44.01 [SD=13.03], father: mean rating=44.00 [SD=14.88]); and annual income of the whole family, assessed when the child was 7 years old and divided into categories (from 1 to 13) of $5,000 (Canadian), with category 1 being <$5,000 and category 13 being >$60,000 (median income score=7/13; interquartile range=5/13). Family structure was assessed when the child was 6 years old; a score of 1 was given if the child was living with both biological parents (N=82.7%) and a score of 0 otherwise.

We included the following two additional control variables reflecting changes in the family between the ages of 6 and 12 years of age, since these changes can influence educational success (28): 1) a documented divorce or a separation between the two biological parents (N=13.9%) and 2) a residential move (N=40.5%). These two variables were derived from annual questions asking whether the target event had occurred within the previous year. A score of 1 indicated that the informant reported the occurrence of the event during this period.

We also assessed dimensions that co-occur with inattention and hyperactivity. The three behavioral dimensions (anxiety/depressive symptoms, physical aggression, and opposition) were assessed from age 6 through age 12 years, as rated by the child's teachers with the Social Behavior Questionnaire. Each item was scored from 0 to 2, depending on how frequently the child manifested the behavior. The score for each dimension was averaged across the 7 years of assessment. We used the following five items to assess anxiety/depressive symptoms: 1) is worried, worries about many things; 2) tends to do things on his own, rather solitary; 3) appears miserable, unhappy, tearful, or distressed; 4) tends to be fearful or afraid of new things or new situations; and 5) cries easily (mean score=2.00 [SD=1.33]; Cronbach's alpha: 0.72–0.77). Assessment of physical aggression encompassed the following three items: 1) fights with other children; 2) bullies other children; and 3) kicks, bites, or hits other children (mean score=0.47 [SD=0.87]; Cronbach's alpha: 0.81–0.88). Assessment of opposition encompassed the following five items: 1) irritable, quick to “fly off the handle;” 2) is disobedient; 3) does not share toys; 4) blames others; and 5) inconsiderate of others (mean score=1.32 [SD=1.48]; Cronbach's alpha: 0.79–0.83). Scores of verbal intelligence were obtained at age 15 years using the Sentence Completion Test (mean score=9.87 [SD=1.52]; range: 0–13), which demonstrated a high correlation with other verbal and nonverbal measures of intelligence and with educational achievement, irrespective of population characteristics (29).

First, we estimated trajectories of inattention and hyperactivity during elementary school using k-means for longitudinal data (30). In this procedure, each observation is first assigned arbitrarily to one group. Then, the mean of each group is calculated and each observation is reassigned to the group with the closest mean. The operation is repeated until convergence. The estimations are repeated a large number of times (10,000 times in the present study) to obtain the best solution as assessed by a criterion that maximizes a ratio computed by dividing the trace of the between-variance by the trace of the within-variance.

Next, we used a binary logistic regression to examine the predictive links between the trajectories and later high school graduation, controlling for confounders. Many of the confounding variables were strongly intercorrelated, and thus we adopted a stepwise procedure. We entered these variables one at a time and did or did not keep them in the model according to a criterion dependent on the imputation procedure. At the final stage, we entered the trajectories of hyperactivity and inattention together. We tested for a two-way interaction between inattention and hyperactivity as well as two-way interactions between either inattention or hyperactivity and each confounding variable.

Information on high school graduation was available for all participants. All also had at least one assessment of inattention and hyperactivity over the seven time points. Participants with at least one assessment were included in the estimation procedure for the trajectories (30) (98.0% had three or more assessments for both dimensions).

For the confounding variables, 0%–20% of the data were missing except for birth weight (24.4%), family income (43.5%), and verbal IQ (38.4%). Even though the percentage of missing data were low for most variables, deletion of cases with missing data for one variable would have resulted in a strong power decrease and a possible bias as a result of differences in the characteristics of nonrespondents. Consequently, we performed multivariate imputation by chained equations to generate 50 complete data sets, a number considered to be sufficient (31, 32). We then fitted the logistic model to each of the 50 imputed data sets and pooled the resulting estimates. This procedure also calculated the fraction of missing information for each estimate, which was attributable to the uncertainty caused by the missing data. We could then directly compare different statistical models fitted to the same imputed data sets. In the aforementioned stepwise procedure for the selection of the confounding variables, the decision to retain or eliminate a given variable was based on a Wald test assessing the information added to the model by the variable.

The following four developmental trajectories of inattention were identified from the analysis (Figure 1): one stable low trajectory (46.3% of the study sample); one stable high trajectory (16.8%); and two crossing trajectories, with one rising (17.6%) and one declining (19.3%). The following four trajectories of hyperactivity were found (Figure 1): one stable low trajectory (59.4%); one high declining trajectory (10.3%); and two crossing trajectories, with one sharply declining (16.0%) and one slightly increasing (14.3%). Biederman (9) emphasized that symptoms of hyperactivity tend to wane, whereas symptoms of inattention tend to persist over time. In the present study, two hyperactivity trajectories were declining, and only one was slightly increasing, whereas for inattention only one trajectory was declining, while the others were stable or sharply increasing.

A total of 67.9% of the participants in the whole sample had a high school diploma at the age of 22–23 years. Table 1 shows that those in the high inattention trajectory were the least likely to have a high school diploma at the age of 22–23 years (29.2%), while those in the low inattention trajectory were the most likely to have completed high school (88.5%). The difference between the high and low hyperactivity groups was smaller (40.1% versus 77.1%, respectively).

Bivariate associations are presented in Table 1 of the data supplement. All variables significantly predicted high school graduation (p<0.001) except birth weight. Socioeconomic variables, anxiety, and verbal IQ correlated more with inattention than hyperactivity. Hyperactivity was strongly correlated with physical aggression and opposition. Parental separation between the ages of 6 and 12 years increased the risk of being in a higher inattention trajectory, particularly the rising trajectory, suggesting the importance of time-varying covariates of these trajectories.

To build the model predicting high school graduation, we used a stepwise procedure for the confounding variables and then entered the trajectories for hyperactivity and inattention. When contrasted with the low inattention trajectory, the declining inattention trajectory (odds ratio=2.67, 95% confidence interval [CI]=1.90–3.75), rising inattention trajectory (odds ratio=3.87, 95% CI=2.75–5.45), and high inattention trajectory (odds ratio=7.66, 95% CI=5.06–11.58) all indicated an increased risk for not graduating from high school (Table 2). In contrast, the hyperactivity trajectories were not significantly associated with high school graduation after accounting for inattention trajectories. We compared the fit of a model, including all the confounding variables as well as inattention, to a model with hyperactivity added and found that hyperactivity did not add significant information to the model.

The interaction between inattention and hyperactivity was not significant. We tested the two-way interactions between inattention and each confounding variable and between hyperactivity and each confounding variable. These interactions were tested for the following two reasons: to determine whether the role of inattention was conditioned by personal or sociofamilial moderators and to determine whether the role of hyperactivity was significant for subgroups of children. No term of interaction added to the model fit except between hyperactivity and residential move (Wald test: p=0.05), but this result should be interpreted with caution given the number of tested interactions.

To test for the possibility that results varied depending on the statistical procedure used, we employed complete data sets without imputation alongside joint trajectories of inattention and hyperactivity and averaged scores instead of trajectories. Lastly, only the first wave of data on a larger sample was used. In all cases, inattention was highly significant but not hyperactivity. Further details regarding these analyses are presented in the data supplement.

This study is the first to use data on developmental trajectories (and joint trajectories) of both inattention and hyperactivity in children from kindergarten to the end of elementary school. The use of a population sample, official records of high school graduation available for all participants, and a range of meaningful confounding variables provide additional strengths.

Although teachers seem more reliable than parents in the diagnosis of ADHD and its subtypes (36, 37), their assessment could be biased by a general negative halo effect. However, the fact that both dimensions had a differentiated effect on educational attainment suggests that any such halo effect was limited. Teachers may have rated behaviors for several children, yielding a lack of independence of assessments. However, given the study design based on sampling of individual children, only a few children may have shared the same teacher, with a change in clustering each year. We were unable to model this complex effect. Additionally, the instrument used includes few items for the key predictors. However, it has proven to be predictive of a range of adult and adolescent outcomes in numerous studies, and our predictive results were based on seven repeated assessments over a 7-year period according to different raters at each time point. Lastly, an assessor bias is possible because teachers who know that their students have poorer academic performance may tend to give these students higher ratings in inattention symptoms, and school performance is highly correlated with educational attainment. Clinical and/or laboratory assessments of inattention by assessors who are blind to school performance are needed.