From Early Attachment to Engagement With Learning in School: The Role of Self-Regulation and Persistence PDF

Developmental Psychology © 2013 American Psychological Association...

Developmental Psychology © 2013 American Psychological Association 2014, Vol. 50, No. 5, 1350 –1361 0012-1649/14/$12.00 DOI: 10.1037/a0032779 From Early Attachment to Engagement With Learning in School: The Role of Self-Regulation and Persistence Kim Drake Jay Belsky University of West London University of California, Davis; King Abdulaziz University; and Birkbeck University of London R. M. Pasco Fearon University College London This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article presents theoretical arguments and supporting empirical evidence suggesting that attachment experiences in early life may be important in the later development of self-regulation and conscientious behavior. Analyses of data from the National Institute of Child Health and Human Development (NICHD) Study of Early Child Care and Youth Development (NICHD Early Child Care Research Network, 2005; N ⫽ 1,149) were conducted to test the association between attachment, measured at 15 and 36 months, and 3 measures of self-regulation (social self-control rated by teachers, task persistence as measured by observers in a series of lab tasks, and a continuous performance test) between Grades 1 and 5. Mediational analyses were also conducted to test whether self-regulation mediates the effect of attachment on children’s engagement with learning in the classroom, as measured by direct observation. The results confirmed the hypothesis that attachment would be related to later self-regulation, but only for social self-control, and attentional impulsivity, not task persistence. Furthermore, social self-control at Grade 1 mediated the effect of attachment (at both 15 and 36 months) on school engagement at Grade 5, even when Grade 1 school engagement was statistically controlled. The discussion focuses on the potential importance of early attachment experiences for the development and maintenance of conscientiousness across the lifespan. Keywords: attachment, conscientiousness, self-regulation, learning, school Supplemental materials: http://dx.doi.org/10.1037/a0032779.supp Conscientiousness, broadly defined, refers to the tendency to be it sizeable benefits for health and wellbeing across the lifespan. organized, careful, persistent in one’s endeavours, responsible, Therefore, the causal factors that give rise to individual differences trustworthy, cooperative and self-possessed in one’s dealings with in conscientiousness are an important focus for prevention re- others (Roberts, Chernyshenko, Stark, & Goldberg, 2005). Con- search. Despite its status as a primary dimension of adult person- scientiousness is thus considered to be closely related to the ality being well established (Goldberg, 1990), much less is known concepts of self-management or self-regulation (Hoyle, 2006), about the developmental antecedents of conscientiousness. This with both intrapersonal aspects (planning, impulse control) and includes, most notably for the purposes of this report, the role of interpersonal ones (considering others, trustworthiness, compli- social experience in shaping its emergence across development. ance with norms and expectations). It is also clear, from the several Behavior genetic studies after all consistently point to the potential articles in this special section, that conscientiousness carries with importance of both environmental, social experiences, in shaping individual differences in conscientiousness (Bergeman et al., 1993). Furthermore, studies have indicated that variance in con- scientiousness in adulthood may be relatively predictable from This article was published Online First May 6, 2013. measurements of child behavior taken in very early childhood Kim Drake, School of Psychology, Social Work & Human Sciences, (Caspi et al., 2003). Such temperamental variables likely represent University of West London, London, England. Jay Belsky, Human and the joint effect of both genetic and environmental influences Community Development, University of California, Davis; Department of (Saudino, 2005). There are thus grounds to investigate the devel- Psychology, King Abdulaziz University, Jeddah, Saudi Arabi; and Depart- opmental processes that contribute to conscientiousness, and its ment of Psychological Sciences, Birkbeck University of London, London, constituent behaviors, in early life. England. R. M. Pasco Fearon, Research Department of Clinical, Educa- As reviewed by Eisenberg, Duckworth, Spinrad, & Valiente tional and Health Psychology, University College London, London, Eng- land. (2014) in this issue, there is clear evidence that the broad constel- Correspondence concerning this article should be addressed to R. M. lation of behaviors and attitudes that typically fall under the Pasco Fearon, Research Department of Clinical, Educational and Health common latent factor of conscientiousness in adults can also be Psychology, University College London, Gower Street, London WC1E identified at least at school-age if not earlier (Goldberg, 2001; 6BT, England. E-mail: [email protected] Halverson et al., 2003). Furthermore, correlational studies indicate 1350  ATTACHMENT, SELF-REGULATION, AND SCHOOL ENGAGEMENT 1351 that there are consistent and coherent links between aspects of shaped by, primary attachment relationships. It is also necessary to temperament—particularly effortful control and positive affectiv- explore the mechanisms via which these relational processes sup- ity—and conscientiousness in childhood (Halverson et al., 2003). port the child’s broader adjustment. In this article we concentrate While one might consider childhood personality and temperament on the potentially fruitful intersections between the field of attach- to be partially overlapping sets of indicators best conceptualized at ment and children’s self-regulation. In particular, we explore the the same level of analysis, a causal-developmental perspective notion that, via several plausible mechanisms, secure attachment encourages a consideration of more complex inter-relationships, may promote aspects of self-regulation, which in turn may bolster beyond the statistical correlations among trait indicators at a given adaptation in a range of contexts. Through this process, we sug- point in time. For example, this perspective highlights the possi- gest, secure attachment could also support the emergence of con- bility that different traits emerge at different developmental scientiousness as a stable trait, with potentially long-term impli- phases, causally influence later-emerging traits over time, and may cations for achievement and health. Having outlined this represent the manifestation of behavior under the control of dis- theoretical background we then present data from the National tinct mechanisms, with distinct developmental antecedents. Roth- Institute of Child Health and Human Development (NICHD) This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. bart, Ellis, Rueda, and Posner (2003), for example, have argued Study of Early Child Care and Youth Development (SECCYD; This document is copyrighted by the American Psychological Association or one of its allied publishers. that temperament represents basic and early-emerging traits rooted NICHD Early Child Care Research Network, 2005) to test the in neurobiological systems that are influenced by genes and expe- contribution of attachment to the growth of self regulation across rience. Temperamental dispositions then causally influence the the primary school years and explore the impact of both on a key emergence of more generalized beliefs, attitudes and habits that example of immediate significance for understanding the contri- progressively cohere into dimensions of personality. Self- bution of self-regulation to children’s emergent conscientiousness, regulatory mechanisms are considered especially relevant in this and ultimately their long-term achievement: engagement with emergent process, particularly for conscientiousness, because they learning in the classroom. are viewed as modulating reactive tendencies (e.g., impulsive approach, distractibility, fearfulness) that, unchecked, might un- Attachment and Self-Regulation dermine successful learning and adaptation in a range of develop- mentally important contexts. This approach thus focuses attention Attachment theory provides a relatively well-articulated model on the early causal influences on self-regulation, and the manner in for understanding the ways in which selected aspects of the which self-regulation then supports conscientious behavior over parent– child relationship may affect a child’s emotional and social time and across contexts. In the current article, we therefore behavior in the short and long-term, including self regulation and examine the role of early social experiences in the growth of thus conscientiousness, and places special emphasis on the child’s self-regulation across childhood and investigate its unfolding con- use of the parent as a secure base from which to explore the tribution to conscientious behavior in a particularly important environment (Bowlby, 1988). In Bowlby’s (1969) original work learning context—the school classroom. These inter-related devel- on attachment, he proposed that the attachment behavioral system opmental processes are an important subject for research in their had been shaped by natural selection to support the survival of own right but also serve to illustrate a developmentally focused immature offspring, by actively monitoring the whereabouts and approach to understanding the emergence and maintenance of availability of attachment figures and signaling to or actively conscientiousness that may have broader application across the seeking out those individuals when there were “natural clues” to lifespan. threat in the environment. According to this view, establishment of The quality of the parent– child attachment relationship is one contact with an attachment figure leads to termination of attach- potential source of influence on the development of self-regulation ment behavior, reduces anxiety and thereby engenders feelings of and its subcomponents. The attachment relationship between a comfort. The attachment system was thus conceptualized as a young child and their primary caregivers has been the focus of homeostatic regulatory system, switching on when threat was extensive investigation in the fields of developmental, clinical and present, and switching off when a perception of safety was applied psychology (Cassidy & Shaver, 2008). This focus is mo- achieved. tivated by the assumption that secure primary attachment relation- Crucially, Bowlby (1969) proposed that the attachment system ships provide the child with developmental advantages that are operates in dynamic interplay with other behavioral systems whose relatively stable over time. A host of studies provide evidence of behavioral ‘outputs’ might be incompatible with that of the attach- such developmental benefits in domains that are arguably relevant ment system. The prime example of this described in Bowlby’s to the subject of conscientiousness. Research indicates, for exam- (1969) early work, and taken forward by many others in the field ple, that children with secure attachments, relative to their insecure since, was exploration (see, e.g., Waters & Cummings, 2000). counterparts, tend to have fewer behavioral problems (Fearon, Clearly, exploration of the environment is incompatible with si- Bakermans-Kranenburg, van IJzendoorn, Lapsley, & Roisman, multaneously seeking comfort from a primary caregiver, and thus 2010), are more resilient when faced with challenge (Matas, Ar- the attachment system when activated must in some sense inhibit end, & Sroufe, 1978; Murray et al., 2011), are more socially exploration. When the comfort-providing function of the attach- competent (Cohn, 1990; Elicker, Englund, & Sroufe, 1992; Wa- ment system is operating successfully however, achievement of ters, Wippman, & Sroufe, 1979), and may show more advanced contact should terminate attachment behavior and “release” the cognitive development (particularly in language, see van IJzen- exploration system from inhibition, thereby promoting the child’s doorn, Dijkstra, & Bus, 1995). return to exploration. This body of research suggests that it may be important to It was this dynamic interplay between attachment and explora- consider the ways in which self-regulation is embedded in, and tion that gave rise to the term “secure base,” which neatly captures  1352 DRAKE, BELSKY, AND FEARON the notion that by consistently providing a haven of safety in times histories. More recently, Murray et al. (2011) found that the of threat, a secure attachment provides support for the child’s anticipated adverse effect of maternal postnatal depression on effective exploration of the environment. This central element of children’s ego-resiliency at 5 and 9 years of age—that is, their attachment theory makes understandable the once seemingly par- capacity to cope and maintain regulatory competence in the face of adoxical prediction that a secure attachment relationship promotes stress resulting from losing a rigged game—was mediated by the separation-individuation process and the growth of autonomy earlier attachment history. Of interest, too, are Gilliom, Shaw, (Matas et al., 1978). This model of attachment and exploration Beck, Schonberg, and Lukon’s (2002) data showing that secure helps conceptualize at least one of the ways in which the parental attachment in infancy forecasts increased capacities for delay of relationship might support the child’s productive engagement with gratification, often considered a component of effortful control, the environment, by providing the child with confidence that he or which is itself regarded as a feature of self-regulation and thus she can fall back on a supportive figure in times of need. would seem pertinent to this report on attachment and conscien- Theorists further assume that this process gradually becomes tiousness. Further and perhaps more direct evidence of a link something that is less immediately dependent on the parent’s between attachment and effortful control was provided by Fearon This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. presence—that is, becomes internalized—thereby enabling it to and Belsky (2004), who analyzed data from the continuous per- This document is copyrighted by the American Psychological Association or one of its allied publishers. generalize across time and context (see Arend, Gove, & Sroufe, formance test—which required children to respond to an infre- 1979; Sroufe, 1988). This process of generalization may be viewed quently appearing target presented on a computer screen, and as emerging as a result of more than one (self-regulatory) process. withhold a response to more frequently occurring nontargets— First, the child becomes increasingly able to cognitively represent administered at age 54 months to children enrolled in the parental figures and their likely supportive responses in their NICHD Study of Early Child Care and Youth Development. absence, such that the knowledge that the parent will be supportive Children judged secure in the Strange Situation at 15 months continues to influence effective exploration outside of the confines were less likely to miss targets (an index of inattention) than of the parent– child relationship (e.g., at nursery or school, see insecure children (avoidant and disorganized), an effect that Suess, Grossmann, & Sroufe, 1992). Second, through repeated was more pronounced for children raised in more adverse social experiences of supportive care, the child acquires more sophisti- circumstances. Similarly, insecure children, particularly boys cated abilities to regulate their emotions, which may, in turn, from more stressful social circumstances, tended to be more enhance the child’s broader capacities for coping and adaptation impulsive, as they more frequently responded to nontargets than (Calkins & Leerkes, 2004; Cassidy, 1994). Third, the repeated secure children. experience of successful, and supported, exploration of the envi- ronment provides the child with greater opportunities to learn, take Attachment, Self-Regulation, and Engagement With initiative, cope with challenge independently, and internalize skills Learning for self-regulation (Riksen-Walraven, Meij, van Roozendaal, & Koks, 1993). If it is indeed the case that attachment promotes self-regulation one would expect that such effects would be apparent in a range of Empirical Support for the Role of Attachment in circumstances where emotional and behavioral self-control, per- sistence, attentional focus, and planning would be important. Ar- Self-Regulation guably the most significant of these in childhood is the school The theoretical proposition that secure attachment facilitates the classroom, where normative expectations for children’s behavior young child’s effective exploration of the environment and the strongly emphasize these behaviors. There is extensive evidence internalization of skills for mastery and self-regulation has been that self-regulatory ability is related to scholastic achievement and the basis of much research on the developmental consequences of engagement with learning (Rothbart & Jones, 1998; Sektnan, attachment. In their classic investigation, Matas, Arend, and McClelland, Acock, & Morrison, 2010; Valiente, Lemery- Sroufe (1978) predicted and observed that 24-month-old children Chalfant, Swanson, & Reiser, 2008). Rothbart and Jones (1998) who had earlier been classified as secure in the Strange Situation have argued that a key contribution of self-regulatory ability to Procedure engaged more enthusiastically with challenging explor- school achievement is via its effects on children’s capacities to atory tasks, spent more time on-task, and proved less likely to derive intrinsic reward from engaging in challenging tasks (“mas- become emotionally frustrated than their insecure counterparts; tery motivation”). By being able to resist distraction, manage similar results emerged at age 4 –5 years when teachers rated these frustration, focus attention, and persist at a difficult task, children same children on their capacity to cope with emotional and social with better self-regulatory abilities will likely experience a greater challenges (i.e., ego-resiliency; Arend et al., 1979). Such work sense of efficacy in dealing with challenge, gain greater intrinsic stimulated Belsky, Garduque, and Hrncir (1984) to test the hy- (satisfaction) and extrinsic reward (e.g., praise) from their suc- pothesis—and find support for the notion—that secure infants cesses and manage the emotions associated with setbacks more would manifest a smaller competence-performance gap when the effectively. Repeated experiences of these kinds would be ex- cognitive sophistication of their exploratory play was monitored. pected to increase motivation to engage in learning, stabilise this Further evidence of greater competence in exploration and engage- tendency over time, and promote its generalization to other con- ment among secure children was reported by Riksen-Walraven et texts (Rothbart & Jones, 1998). The notion that these processes al. (1993), who observed, like Hazen and Durrett (1982) had may be influenced by the quality of early primary caregiving before them, that secure infants later explored a novel toy with relationships has been strongly advocated by Pianta, Smith, and greater engagement, energy and enthusiasm when they were 30- Reeve (1991). Although limited research has directly explored the month-old toddlers than did children with insecure attachment role of early attachment security in engagement with learning at  ATTACHMENT, SELF-REGULATION, AND SCHOOL ENGAGEMENT 1353 school, there is good evidence that the quality of parent– child families completed a home interview when the infant was 1 month interaction, itself a reliable if modest predictor of attachment old and became the study participants. Details of the sampling plan security (De Wolff & van IJzendoorn, 1997), is related to child can be found in NICHD Early Child Care Research Network, competence and behavior problems in the classroom (Pianta & 2005). In terms of demographic characteristics, 26% of the moth- Harbers, 1996; Pianta et al., 1991). ers had no more than a high school education at time of enrollment, 21% had incomes no greater than 200% of the poverty level at The Current Study sixth grade, and 22% were minority (i.e., not non-Hispanic Euro- pean American). In this article we report on analyses based on As reviewed up to this point, there are good theoretical reasons 1,149 children who had data on attachment at 15 months and 1,140 for expecting secure attachment to be associated with greater children who had data on attachment at 36 months. The cases self-regulatory capacity over the course of development, and the included in the analyses were different to those not included on modest body of existing evidence suggests that this may be the several demographic variables, with the included families drawn case. Furthermore, there is good evidence that self-regulatory from generally less disadvantaged circumstances. Families in the This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. ability is a determinant of school engagement and achievement, current analyses were less likely to report during the study period This document is copyrighted by the American Psychological Association or one of its allied publishers. which is arguably a critical context in which emergent conscien- that the child’s father did not live in the family home (t1362 ⫽ 2.70, tiousness develops. However, most of the work on attachment and p ⬍.001), had a higher average family income (t1362 ⫽ 3.43, p ⫽ self-regulation has been limited in statistical power and restricted.001), and had greater maternal education (t1361 ⫽ 5.18, p ⬍.001), to single points in time, which precludes a developmental analysis and the mothers tended to be older (t1362 ⫽ 4.38, p ⬍.001). of the growth of self-regulatory skills. Furthermore, few if any studies have examined the seemingly plausible hypothesis that self-regulatory capacities may mediate the effect of attachment on Measures later developmental outcomes and on school engagement in par- Measurements used in the research reported herein were se- ticular. lected from the NICHD SECCYD archive to provide information In the current article, we examine children’s behavior in the regarding (a) attachment, (b) self-regulation, and (c) school en- classroom, and particularly their engagement with learning, be- gagement. Measures of these constructs were chosen deliberately cause this should be a highly discriminating context for observing to minimize the possibility that shared method variance could the effects of individual differences in self-regulatory capacity, is influence the longitudinal associations. Thus, attachment was mea- an important context in which conscientiousness is manifest and sured in relation to the child’s mother in all cases, whereas three develops and because of its broader relevance for subsequent distinct indices of self-regulation (see below) were based on achievement (e.g., see Wentzel, 1999; Wolters, 1999). A further teacher reports, observer ratings of child behavior in the presence reason for choosing this outcome is that it represents a more novel of the father (or other adult–and not the mother) or in the child’s outcome from the point of view of existing attachment research, objective performance. School engagement was assessed by and hence affords the opportunity for a more exacting test of the trained research assistants, directly observing the children in the theory outlined above. In other words, while attachment effects classroom. may be expected, given past research, in relation to more com- Information about this public data set can be found at http:// monly studied outcomes such as behavioral problems, a model secc.rti.org/. Ethical approval was granted by each of the 10 based on self-regulation as a primary mediating process would data-collecting universities before any data included in this re- make the more novel prediction that attachment effects should be ported were collected, and at each age of measurement, informed evident in the classroom, where the requirements for self- consent was secured from parents and/or teachers. regulation are demanding. The current article therefore (a) examines the role of attachment security, as measured at 15 and 36 months using standardized Attachment observational instruments, in the growth of self-regulatory behav- Infant–mother attachment security was assessed at both 15 iors across the early school years (Grades 1 to 5), (b) tests the role months and 36 months. The 36-month attachment assessment was of self-regulatory processes in change in children’s engagement included in light of some evidence that later attachment assess- with learning in the classroom from Grade 1 to 5, and (c) tests ments may carry greater predictive strength than earlier ones whether self-regulatory processes mediate the effects of attach- (Fearon et al., 2010; McCartney, Owen, Booth, Clarke-Stewart, & ment on change in school engagement. Vandell, 2004). At 15 months, Ainsworth and Wittig’s (1969) standard Strange Situation procedure was used. At 36 months, a Method modified Strange Situation procedure (SSP) based on recommen- dations by Cassidy and Marvin (1992) and the MacArthur Work- Participants ing Group on Attachment was used. Each SSP videotape was coded twice at a central location by two of three coders blind to all The NICHD SECCYD recruited 1,364 families through hospital information on the children, using the Ainsworth et al. (1978) visits shortly after the birth of a child in 1991 at 10 U.S. locations standard criteria of secure (B), insecure-avoidant (A), insecure- (for detailed description of recruitment procedures and sample resistant (C), disorganized (D), and unclassifiable (U) at age 15 characteristics, see NICHD Early Child Care Research Network, months and the system developed by the MacArthur Working 2001). During selected 24-hr intervals, all women giving birth Group on Attachment (Cassidy & Marvin, 1992) at 36 months. (N ⫽ 8,986) were screened for eligibility. From that group, 1,364 Even though several years separated coding of 15- and 36-month  1354 DRAKE, BELSKY, AND FEARON attachments, every effort was made to insure that the same coders of EC (see Mirsky, Anthony, Duncan, Ahearn, & Kellam, 1991; did not code the same children at both times of measurement. Rosvold, Mirksy, Sarason, Bransome, & Beck, 1956). The task Reliability at 15 months was calculated based on 867 randomly required children to press a button whenever a target stimulus paired cases. Inter-coder agreement on the four-category A, B, C, appeared on a computer screen. At 54 months, children were and D classifications was 75.7% (kappa ⫽.58). At 36 months, presented with 22 blocks of 10 stimuli; the stimulus duration, at agreement with the five-category classification system was 83% this age, was 500 ms, with an inter-stimulus interval of 1,500 ms. (kappa ⫽.69; NICHD ECCRN, 1997). Only cases classified A, B, At Grade 1, 30 blocks of 10 stimuli were presented; each stimulus C, or D are included in this study (N ⫽ 1,149). lasting 200 ms, with an inter-stimulus interval of 1,500 ms. The target stimulus was randomly presented twice in each block with all response data automatically tallied by the computer. Self-Regulation Following the earlier article by the NICHD ECCRN (see Bel- Three distinct aspects of self-regulation that may be differen- sky, Fearon, & Bell, 2007) we used children’s commission errors tially related to attachment history are included in this report. (i.e., responses to nontargets) as a single indicator of EC, as these This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Social self-control reflects the child’s ability to regulate behavior were previously found to be associated with children’s behavioral This document is copyrighted by the American Psychological Association or one of its allied publishers. and affect in the context of social relationships, as reported by problems and have been used extensively in other clinical and teachers. Task persistence reflects the child’s capacity to engage developmental studies. The test–retest reliability of the CPT is consistently in a challenging task without becoming distracted or good in the age range studied in this research (Belsky et al., 2007; irritable. Effortful control (EC) represents the capacity to inhibit Halperin, Wolf, Greenblatt, & Young, 1991). pre-potent responses. Social self-control (SSC). Ten items from the self-control School-Engagement subscale of the Social Skills Rating System (Gresham & Elliott, 1990) were used, repeatedly across the primary school years, to School-engagement was measured using direct in-school obser- measure behavioral and emotional regulation in social contexts, vations, obtained using the Classroom Observation System at with items reflecting the extent to which teachers judged children Grades 1, 3, and 5 (NICHD Early Child Care Research Network, to be able to control temper in conflict situation with peers; 2005, 2006). The focus of the observations was the child’s behav- compromise in conflict situations; respond appropriately to peer ior and experiences within the classroom. Observations were re- pressure; respond appropriately to teasing by peers; control temper corded on a 7-point scale regarding the degree to which the child when in conflict situation with adults; receive criticism well; was engaged in classroom activities and teacher– child interactions accept peer ideas during group activities; cooperate with peers during several observational cycles by trained observers (see sup- without prompting; respond appropriately when pushed or hit. plemental online materials for further information). Inter-rater When these 10 items were subject to principle component analysis reliability was ⬎.80 at each age. and oblimin rotation, separately at each of five ages of measure- ment, to determine whether these items did in fact load strongly Missing Data onto a common factor across grades, evidence emerged that two The numbers of cases with valid data varied significantly across items needed to be dropped at Grade 1 (only) to generate highly time and across measures (see Table 1 for a summary). To avoid internally consistent measures: “accepts peer ideas for group ac- excluding cases with missing data, which can undermine statistical tivities” and “gets along with people who are different.” Chron- bach’s alpha ranged from.91 to.94 across grades. Task-persistence (TP). TP was measured during the Adult– Table 1 Child Interaction Task (ACIT) designed for the NICHD SECCYD Covariances and Correlations Between Self-Regulation study to assess the quality of father/partner/other adult and child Measures by Grade behavior during a semistructured teaching and play situation which presented behavioral and emotional challenges when children were Variable Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 in Grades 1, 3, and 5. These interactions were videotaped and Social self-control rating scales used to measure parenting dimensions and child Grade 1 7.822 0.440 0.410 0.419 0.406 behaviors. Further details of the interaction tasks at each age are Grade 2 3.636 8.734 0.457 0.485 0.486 presented in the supplemental online materials. Grade 3 3.447 4.062 9.053 0.492 0.454 Children were rated on a 7-point scale ranging from very high to Grade 4 3.449 4.220 4.360 8.673 0.512 very low (levels of TP). TP is a measure of the child’s involvement Grade 5 3.478 4.396 4.180 4.614 9.369 M 10.810 11.011 10.675 10.786 10.798 in the tasks irrespective of the degree to which the parent is SD 2.797 2.955 3.009 2.945 3.061 instrumental in creating the persistence. Children scoring low in Task persistence TP show no involvement in the discussion or the tasks, and either Grade 1 0.960 0.130 0.086 refuse to get involved or spend the time off-task, easily distracted, Grade 2 Grade 3 0.128 1.012 0.271 inattentive, bored and/or fidgety. Children scoring very high on TP Grade 4 are actively engaged with all of the discussion and tasks and works Grade 5 0.080 0.259 0.902 persistently at the tasks (either independently or with the help of M 6.089 5.627 5.257 the father/partner/other adult), offering ideas and asking questions. SD 0.980 1.006 0.950 Effortful control (EC). Performance at 54 months and Grade Note. Covariances appear below the diagonal, correlations above the 1 on the Continuous Performance Task (CPT) served as indicators diagonal. Variances are shown on the diagonal.  ATTACHMENT, SELF-REGULATION, AND SCHOOL ENGAGEMENT 1355 power and bias parameter estimates (Allison, 2003), we employed ment security) and mediator (self-regulation), we also controlled the full-information maximum likelihood (FIML) method, which for Grade 1 school engagement. uses all the available data to estimate the parameter estimates of a Due to the high level of nonlinearity associated with the CPT model (by calculating the log-likelihood of the data for each scores from 54 months to Grade 4, LGM models were not esti- observational unit separately). This approach is clearly superior to mated and only the CPT data at 54 months and Grade 1 were used mean substitution and list-wise deletion, and of comparable per- in the mediation analysis. In this case, EC at both 54 months and formance to multiple-imputation (Allison, 2003; Schafer & Gra- Grade 1 were treated as potential mediators between early attach- ham, 2002). ment and school engagement. All analyses were conducted using Mplus Version 4.21 (Muthén & Muthén, 2006). Data Analysis Model CFI Derivation The longitudinal data on social-self control (SSC) and task Owing to the sensitivity of CFIs to misspecification, both in the persistence (TP) were analyzed using latent growth curve (LGC) This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. marginal mean and covariance structure, a relatively novel ap- models. These offer a multivariate approach to longitudinal data This document is copyrighted by the American Psychological Association or one of its allied publishers. proach was utilized to obtain the CFI for each LGM. For each analysis based on the general structural equation modeling frame- LGM we computed two different CFIs, each based on a compar- work. The advantage of this approach is that it allows for (a) the ison of a null model against a hypothesized model in which either estimation of the overall growth, (b) the estimation of the hetero- the mean structure or the covariance structure was saturated. In geneity in the growth curve at the individual level, and (c) the doing so, these adjusted CFIs can more precisely capture lack of fit flexible analysis of the invariance of growth parameters across in the mean structure or covariance structure of a LGM than groups, as well as the influence of constant and time-varying traditional CFIs (see Wu & West, 2010). covariates, and the efficient estimation of effects in the presence of missing data (Curran & Willoughby, 2003). The models were first constructed to estimate the initial level at Results Grade 1 (the intercept parameter) and growth from Grades 1 to 5. The results are presented in four subsections: (a) descriptive The overall means and variances in these parameters were also statistics; (b) the latent growth curve fittings (LGCs) of self- estimated using maximum-likelihood estimation. The quality of regulation and task persistence, respectively, over Grades 1 to 5, the model fit was simultaneously tested using the maximum- and Grades 1, 3, and 5; (c) the attachment-group effects on both likelihood ratio-test statistic and indices of model fit (i.e., the LGCs; and (d) the mediating effects of SSC, TP, and EC (inves- comparative fit index [CFI], and the root-mean-square error of tigated separately as a function of attachment group) on change in approximation [RMSEA]). CFI of values of.90 or above and school-engagement from Grade 1 to 5. RMSEA values of.08 or lower are considered to indicate an acceptable model fit (Bentler, 1990). With the growth parameters of SSC and TP in hand, we inves- Descriptive Statistics and LGC Modeling of Social tigated their relation to attachment security, comparing secure and Self-Control and Task-Persistence (all) insecure children. We did this by comparing models where Table 1 presents the descriptive statistics for teacher-reported both the intercept and growth parameters were allowed to vary social self-control and task-persistence from Grades 1 to 5; recall across the secure and insecure attachment groups with alternative that for task-persistence data were only obtained at Grade 1, 3, and models where either (a) the intercept parameter was constrained or 5. The covariance/correlations amongst these variables are also (b) the growth parameter was constrained. Again, the quality of the included. The data show that, overall, children’s self-control levels model-fit was ascertained by inspecting the CFI and RMSEA peaked at Grade 2, declined at Grade 3, and then remained rela- values. tively stable throughout Grades 4 and 5; this suggested some Finally the models of SR and TP across Grades 1 through 5 as general decline and the possibility of a degree of curvilinearity to a function of attachment were used to predict school-engagement the data. A latent growth curve model comprising a random at Grade 5, controlling for school-engagement at Grade 1, thereby intercept and both a linear and quadratic slope proved a good fit to resulting in the prediction of change in school engagement. This the data, ␹2(6) ⫽ 11.36, p ⫽.08; CFI ⫽.99; RMSEA ⫽.028. was done to test the notion that these self-regulatory measures However, constraining the quadratic term to zero did not signifi- (SSC and TP) mediate the effect of attachment on change over cantly reduce overall model fit, ⌬␹2(4) ⫽ 2.68, p ⫽.62. Con- time in school engagement, as observed in the classroom. In these straining the linear slope to zero did reduce overall model fit final models we treated attachment as a binary independent vari- (⌬␹2 ⫽ 11.01, p ⫽.01). A model with intercept and linear slope able (as opposed to two groups in which separate growth models was thus chosen as the best fitting model. This model fit the data were run, as in the first and second models), and regressed the well (CFI ⫽.99, RMSEA ⫽.019; CFI ⫽.99, p ⫽.33 [covariance intercepts of SSC on attachment security, and Grade 5 school saturated]; CFI ⫽ 1.00, p ⫽.64 [mean saturated]). The intercept engagement (net of Grade 1 engagement) on the intercept, so that showed significant variance (variance ⫽ 3.809, SE ⫽ 0.33); the indirect paths (attachment ⬎ self-regulation ⬎ school engage- variance of the slope was marginally significant (variance ⫽ ment) could be estimated. Tests of mediation were conducted on 0.065, SE ⫽ 0.034). the indirect paths. By controlling for Grade 1 school engagement Overall task-persistence levels showed a gradual decline from when predicting school engagement 4 years later, we strengthened Grade 1 to 3 and then to 5. This suggests a linear pattern to the data the basis for causal interpretations regarding the predictor (attach- and therefore a linear growth curve would best describe the pattern  1356 DRAKE, BELSKY, AND FEARON of growth. A latent growth curve model comprising a random Table 3 intercept and random linear slope fit the data well, ␹2(1) ⫽ 1.266, Intercepts and Slopes for Overall LGC Models, and Separately p ⬎.05; CFI ⫽.995; RMSEA ⫽.017; significant individual by Attachment Group at 15 and 36 Months variation occurred in the intercept and slope variance (intercept variance ⫽ 0.173, SE ⫽ 0.083; linear slope variance ⫽ 0.112, Intercept Linear slope SE ⫽ 0.043). Variable B p B p Base LGC models Attachment Security and the Growth of Social Self- Social self-control 10.787 ⬍.001 ⫺0.009 0.969 Control, Persistence, and Effortful Control Task persistence 5.999 ⬍.001 ⫺0.407 ⬍.001 Table 2 presents both the sample means and standard deviations of Attachment LGC models teacher-ratings of social self-control and task-persistence levels from Social self-control 15 m Insecure 10.507 ⬍.001 0.041 0.157 Grades 1 to 5 by attachment group (secure vs. insecure attachment) as 15 m Secure 10.993 ⬍.001 ⫺0.028 0.184 This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. measured at both 15 and 36 months of age. Table 3 shows the 36 m Insecure 10.340 ⬍.001 0.045 0.143 This document is copyrighted by the American Psychological Association or one of its allied publishers. intercept and growth parameter estimates of teacher-reported self- 36 m Secure 11.055 ⬍.001 ⫺0.016 0.290 control and task-persistence. The LGC models were estimated in Task persistence order to investigate the main effects of attachment group at both 15 15 m Insecure 5.999 ⬍.001 ⫺0.407 ⬍.001 15 m Secure 6.113 ⬍.001 ⫺0.405 ⬍.001 and 36 months of age on both the intercept and change parameters for 36 m Insecure 5.992 ⬍.001 ⫺0.419 ⬍.001 children’s self-control and task-persistence levels. 36 m Secure 6.115 ⬍.001 ⫺0.406 ⬍.001 When we examined the growth curves of social self-control and Note. LGC ⫽ latent growth curve; m ⫽ months. persistence as a function of attachment security we found significantly better social self-control in secure children at Grade 1 (the intercept), for both 15-month and 36-month attachment security, but no effects of attachment on persistence. To show this we proceeded as follows: dren who were secure at 15 months scored more highly overall on (a) first we tested a latent growth curve model for teacher-reports of social self-control at baseline (Grade 1) than insecure children both children’s social self-control and task-persistence levels and (secure intercept ⫽ 10.99; insecure intercept 10.51; see Table 3). checked whether the variances of our self-regulation measures could The effect size was modest, with Cohen’s d ⫽ 0.20. Turning to be assumed to be equal across attachment groups; (b) we then ran 36-month attachment, in the case of social self-control there was a