As have others, we conceptualise classroom engagement as being driven by individual differences in cognitive control (Li-Grining et al. 2010; Ponitz et al. 2009; Sasser, Bierman & Heinrichs 2015). Working memory is a component of cognitive control, which is akin to one’s mental workspace or desktop where learners can manipulate and update information that they will use in the short term (Bull et al. 2012). Research suggests that working memory contributes to short- and long-term academic success (Fitzpatrick & Pagani 2012; Fitzpatrick et al. 2015). As such, it is possible that individual difference in preschool working memory predicts the nature and course of classroom engagement throughout elementary school.

Preschool-aged children who arrive at school better able to control and regulate their attentional resources have an easier time-sustaining engaged learning in the classroom (Duncan et al. 2007; Pagani et al. 2010; Razza, Martin & Brooks-Gunn 2010). In contrast, students who have difficulty inhibiting inappropriate behaviour and who are more distractible may be at greater risk of disengaging from classroom activities. Consequently, it is likely that preschool-aged children who demonstrate attention problems in the form of impulsivity (i.e. not being able to wait their turn, not being able to settle down) will also be less engaged in the elementary school classroom.

Young children’s family context and experiences can also influence their capacity for engaged learning. Early life exposure to family adversity and toxic stress has been linked to excessive activation in children’s stress response systems (Lupien et al. 2007; Robles et al. 2016; Rutter & O’Connor 2004). Such chronic activation can then lead to a series of structural and functional alternations in the brain circuitry of the amygdala and hippocampus and prefrontal cortex. Eventually, these alterations are believed to contribute to increased anxiety and poor stress regulation during learning activities and ultimately impair engagement (Arnsten 2009; Shonkoff 2011).

Both positive and adverse rearing environments are likely to shape learning outcomes. Experimental animal research provides compelling evidence that high-quality parenting in terms of maternal responsiveness and warmth can exercise long-term influences on the cognitive development of offspring through epigenetic processes (Champagne & Curley 2009). Longitudinal research in humans has generated similar findings. For example, maternal warmth in early childhood has been shown to positively influence social and cognitive outcomes in children prior to school entry (Guinosso, Johnson & Riley 2015; Wade et al. 2015).

Finally, children’s broader ecology can also influence their healthy development (Bronfenbrenner 1986; Sameroff 2010). For one, enduring stress that leads to emotional, physical and mental exhaustion in parents may impair effective parenting practices (Schieve et al. 2007). In contrast, parental social support may have a positive effect on child outcomes by increasing the number of resources available to parents (Burchinal, Follmer & Bryant 1996).

Little research has employed a person-centred approach to examine how child and family characteristics influence the course of children’s learning-related skills across elementary school. Person-centred approaches allow us to identify discrete subgroups of people who follow distinct developmental trajectories over time. Thus, the identification of non-normative risk pathways is particularly germane to intervention and prevention research.

The Quebec Longitudinal Study of Child Development (QLSCD 1998–2010) represents a birth cohort randomly selected from a stratified sample of 2837 children born in the province of Quebec, Canada, between 1997 and 1998. The sample was stratified to ensure a proportional representation across rural and urban areas, geographical region, socio-economic status and gender. At the inception of the longitudinal component of this public data set coordinated by the Institut de la Statistique du Québec, 93 children were deemed ineligible and 172 were untraceable because of incorrect coordinates. Of the 2572 remaining children, 14 were unreachable and 438 refused to participate. Thus, for its early childhood phase, 2120 five-month-old infants were deemed eligible for follow-up, representing 82% of the eligible target population defined by the birth cohort. Of these, 39% were firstborn. Participants were included in this ethically approved study if they had least one valid classroom engagement score between kindergarten and fourth grade at 74 and 122 months, respectively (N = 1589). From the second grade onwards, data collection occurred on a bi-annual basis; therefore, no data are available for the third grade. This subsample represents 79% from the original sample. Compared to non-retained participants, participants included in the analytic sample were from higher socio-economic status. There were no sample differences in family configuration, child preschool impulsivity, working memory, number knowledge and receptive vocabulary.

For the preschool collection waves, data were collected from children and their families around the date of the child’s birthday. The questionnaires were administered predominantly in French, reflecting the linguistic distribution of the province of Quebec, Canada. During the school-aged waves, data collection occurred in the spring. Families, teachers and school principals received informed consent forms by mail at each data collection wave. Teachers and families also received and returned questionnaires by mail. Teachers were informed that each target child had been followed from birth as a part of an ongoing study of development. Teachers were also informed that participation would involve completing a questionnaire concerning the behaviour, academic performance and social skills of the child at school. Parents were also interviewed by phone or in person by a research assistant trained by the government operated Direction of Social Services. Parents were compensated $25 and were informed that their child would receive a small present.

When children were 53 and 65 months, mothers provided information about three dimensions of parenting practices (hostility, warmth and discipline) using a version of the Parenting Perceptions and Behaviors Questionnaire adapted for 4- and 5-year-olds (Parental Cognitions and Conduct Toward the Infant Scale; Boivin et al. 2005). Items were rated on a five-point Likert scale, ranging from 1 (never) to 5 (always). The Cronbach’s alphas for the five parenting factors were calculated based on the present sample.

Our interest was first to detect trajectories of classroom engagement from kindergarten through grade 4. We were also interested in examining how early childhood characteristics, parenting and family context contribute to these trajectories. We used growth mixture modelling to estimate distinct intercepts and slopes for groups with unique patterns of the variable of interest (Jung & Wickrama 2008).We estimated unconditional trajectories because our interest was in first identifying naturally occurring patterns of engagement and then estimating whether these trajectories can be estimated by meaningful predictors at the child and family level. Model selection was based on fit indices, prior research and the ability to account for meaningful diversity in the data (Richters 1997). We first generated a one-group linear model to which a two-group model was compared. Models with additional groups were then tested sequentially against two-, three- and four-group models. We repeated the same process with higher-order growth terms. Based on our criteria, a three-group quadratic growth model was favoured. Participants were assigned to groups based on their actual most likely classification (Ram & Grimm 2009). We next examined prospective associations between early childhood cognitive control, family adversity and control variables on classroom engagement trajectories using polynomial linear regression.

This study was approved by the Ethics Review Board of Université de Montréal’s Faculty of Arts and Science (reference 2008-179, 2696).

To ensure that our classroom engagement scale provided an equivalent assessment of the construct at the different time points, we began by testing longitudinal measurement invariance from Grades 1 to 4 using the steps outlined by Byrne and Stewart (2006). Invariance was reached for all steps. To address our first objective, we used growth mixture modelling to estimate classroom engagement trajectories between kindergarten and fourth grade through Mplus software. Parameters were obtained by maximum likelihood estimation using an expectation maximisation algorithm (Muthén et al. 1987). As reported in Table 1, the Akaike information criterion (AIC), Bayesian information criterion and sample-size-adjusted Bayesian information criterion (SSBIC) favoured a three-class quadratic solution. The fit criteria for the three-group quadratic model were good (AIC = 10 823.94 and SSBIC = 10 870.01), corresponding to a low (9%, of which 26% = girls and 74% = boys), medium (19%, of which 44% = girls and 56% = boys) and high trajectory (72%, of which 57% = girls and 43% = boys). As an indicator of model fit, we also examined entropy, which was 0.71 for our selected model and which was slightly higher than the entropy observed for the other models (Ramaswamy et al. 1993). Next, we examined the bootstrap likelihood ratio test because it has been shown to provide reliable estimates of model fit. The observed p-value comparing a three- to two-class model quadratic model was smaller than 0.0001, suggesting a significantly better fit for the three-class quadratic model. There was no significant improvement in fit from a three- to four-class model. As a result, we selected a three-class quadratic model for its statistical fit and because it was consistent with prior research examining trajectories of classroom engagement across elementary school (Pagani, Fitzpatrick & Parent 2012). The average posterior class probabilities for the low, medium and high trajectories were 0.90, 0.74 and 0.83, respectively. The posterior mean fitted and observed values for the trajectories in primary school are illustrated in Figure 1. Descriptive statistics according to group membership are presented in Table 2.

Bivariate correlations between the main predictors and classroom engagement scores are presented in Table 3. In terms of regression analyses, Table 4 reports the relationship between child and family predictors and profiles of classroom engagement during primary school. All models control for child sex, prosocial skills, emotional distress and math and reading skills, as well as family configuration and socio-economic status. All odds ratios are interpreted using ‘low engagement’ as the reference group. The Negelkerke r-squared was 0.22, suggesting that our model accounted for a modest amount of the variance in the observed trajectories. In terms of the second objective, a one-point increase in the working memory scores corresponded to 35% greater odds of belonging to the high engagement group. Conversely, a one-point increase on the impulsivity scale was associated with a 43% reduction in the probability of belonging to the high engagement group. Scores of impulsivity were also associated with a 14% decrease in the probability of membership to the moderate engagement group.

In terms of objective 3, maternal hostility was independently associated with a 27% and 22% reduction in the odds of belonging to the high and moderate engagement groups, respectively. Mother self-reported social support was associated with 20% greater odds of belonging to the high engagement group. Finally, maternal stress was associated with a 30% and 44% decrease in the probability of being in the high and moderate engagement groups, respectively.