10
Parenting, attention and externalizing problems: testing mediation longitudinally, repeatedly and reciprocally Jay Belsky, 1 R.M. Pasco Fearon , 2 and Brian Bell 1 1 Birkbeck University of London, UK; 2 University College London, UK Background: Building on prior work, this paper tests, longitudinally and repeatedly, the proposition that attentional control processes mediate the effect of earlier parenting on later externalizing prob- lems. Methods: Repeated independent measurements of all three constructs – observed parenting, computer-tested attentional control and adult-reported externalizing problems – were subjected to structural equation modeling using data from the large-scale American study of child care and youth development. Results: Structural equation modeling indicated (a) that greater maternal sensitivity at two different ages (54 months, 6 years) predicted better attentional control on the Continuous Performance Test (CPT) of attention regulation two later ages (6/9 years); (2) that better attentional control at three different ages (54 months, 6/9 years) predicted less teacher-reported externalizing problems at three later ages (6/8/10 years); and (3) that attentional control partially mediated the effect of parenting on externalizing problems at two different lags (i.e., 54 months 6 years 8 years; 6 years 9 years 10 years), though somewhat more strongly for the first. Additionally, (4) some evidence of reciprocal effects of attentional processes on parenting emerged (54 months 6 years; 6 years 8 years), but not of problem behavior on attention. Conclusions: Because attention control partially mediates the effects of parenting on externalizing problems, intervention efforts could target both parenting and attentional processes. Keywords: Parenting, sensitivity, attention, inhibition, externalizing problems, mediation. Extensive research reveals linkages between the quality of parenting that children receive and the development of externalizing problems, with work repeatedly documenting adverse effects of harsh, conflicted and coercive parenting and beneficial consequences of warm, supportive and sensitive rearing. This is true of cross-sectional and longit- udinal research during the early-childhood/ elementary-school years (e.g., Pettit et al., 1997; Gadeyne et al., 2004) and the later-childhood/ adolescent years (e.g., Deater-Deckard & Dodge, 1997; Pieser & Heaven, 1996). Given such consistent findings, attention is appropriately turning to the mediating psychological mechanisms that account for associations between parenting and behavior-problems. One potential mediator is attentional or effortful control (NICHD ECCRN, 2003). These processes are defined collec- tively as ‘the ability to inhibit a dominant response and/or to activate a subdominant response, to plan, and to detect errors’ (Eisenberg et al., 2005a; Roth- bart et al., 2000). Effortful control, including emotional control, appears compromised by harsh, negative parenting and fostered by warm, supportive rearing (Eisenberg et al., 2005a). Evidence consistent with this pro- position comes from studies of toddlers (Kochanska et al., 2000), young children/pre-adolescents (Eisenberg et al., 2005a) and adolescents (Gauvain & Huard, 1999) and utilizing diverse definitions and measures of effortful control. Especially notable, perhaps, are findings of an intervention study showing that youth self-control, which is conceptu- ally related to effortful control as it includes the ability to set goals and plan actions while consider- ing their consequences, improved among 332 African-American 11-year-olds when supportive and responsive parenting increased (Brody et al., 2005). The fact that parenting predicts effortful-control attentional processes seems even more significant given evidence that indices of effortful control predict externalizing problems (Caspi et al., 1995). In the previously cited Eisenberg et al. (2005a) study, lower adult ratings of child effortful control at 9 and 11 years forecast higher levels of adult-rated exter- nalizing problems at age 13, extending similar results obtained at earlier ages of measurement (Eisenberg et al., 1997). Importantly, Eisenberg’s (Eisenberg et al., 2005b) work on 4–7-year-olds followed up two years later also reveals a reciprocal process, with changes in externalizing problems fostering changes in regulatory abilities, including effortful control and impulsivity. Such evidence underscores the need to consider bi-directional processes of influence in research on parenting, attentional control and problem behavior. Given the work summarized above, it is note- worthy that several recent longitudinal studies, mostly of young adolescents, provide evidence that effortful-control processes (partially) mediate Conflict of interest statement: No conflicts declared. Journal of Child Psychology and Psychiatry 48:12 (2007), pp 1233–1242 doi:10.1111/j.1469-7610.2007.01807.x Ó 2007 The Authors Journal compilation Ó 2007 Association for Child and Adolescent Mental Health. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

Parenting, attention and externalizing problems: testing mediation longitudinally, repeatedly and reciprocally

  • Upload
    ucl

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Parenting, attention and externalizingproblems: testing mediation longitudinally,

repeatedly and reciprocally

Jay Belsky,1 R.M. Pasco Fearon ,2 and Brian Bell11Birkbeck University of London, UK; 2University College London, UK

Background: Building on prior work, this paper tests, longitudinally and repeatedly, the propositionthat attentional control processes mediate the effect of earlier parenting on later externalizing prob-lems. Methods: Repeated independent measurements of all three constructs – observed parenting,computer-tested attentional control and adult-reported externalizing problems –were subjected tostructural equation modeling using data from the large-scale American study of child care and youthdevelopment. Results: Structural equation modeling indicated (a) that greater maternal sensitivityat two different ages (54 months, �6 years) predicted better attentional control on the ContinuousPerformance Test (CPT) of attention regulation two later ages (�6/9 years); (2) that better attentionalcontrol at three different ages (54 months, �6/9 years) predicted less teacher-reported externalizingproblems at three later ages (�6/8/10 years); and (3) that attentional control partially mediated theeffect of parenting on externalizing problems at two different lags (i.e., 54 months fi �6 years fi �8years; �6 years fi �9 years fi �10 years), though somewhat more strongly for the first. Additionally,(4) some evidence of reciprocal effects of attentional processes on parenting emerged (54 months fi �6years; �6 years fi �8 years), but not of problem behavior on attention. Conclusions: Becauseattention control partially mediates the effects of parenting on externalizing problems, interventionefforts could target both parenting and attentional processes. Keywords: Parenting, sensitivity,attention, inhibition, externalizing problems, mediation.

Extensive research reveals linkages between thequality of parenting that children receive and thedevelopment of externalizing problems, with workrepeatedly documenting adverse effects of harsh,conflicted and coercive parenting and beneficialconsequences of warm, supportive and sensitiverearing. This is true of cross-sectional and longit-udinal research during the early-childhood/elementary-school years (e.g., Pettit et al., 1997;Gadeyne et al., 2004) and the later-childhood/adolescent years (e.g., Deater-Deckard & Dodge,1997; Pieser & Heaven, 1996).

Given such consistent findings, attention isappropriately turning to the mediating psychologicalmechanisms that account for associations betweenparenting and behavior-problems. One potentialmediator is attentional or effortful control (NICHDECCRN, 2003). These processes are defined collec-tively as ‘the ability to inhibit a dominant responseand/or to activate a subdominant response, to plan,and to detect errors’ (Eisenberg et al., 2005a; Roth-bart et al., 2000).

Effortful control, including emotional control,appears compromised by harsh, negative parentingand fostered by warm, supportive rearing (Eisenberget al., 2005a). Evidence consistent with this pro-position comes from studies of toddlers (Kochanskaet al., 2000), young children/pre-adolescents(Eisenberg et al., 2005a) and adolescents (Gauvain

& Huard, 1999) and utilizing diverse definitions andmeasures of effortful control. Especially notable,perhaps, are findings of an intervention studyshowing that youth self-control, which is conceptu-ally related to effortful control as it includes theability to set goals and plan actions while consider-ing their consequences, improved among 332African-American 11-year-olds when supportive andresponsive parenting increased (Brody et al., 2005).

The fact that parenting predicts effortful-controlattentional processes seems even more significantgiven evidence that indices of effortful control predictexternalizing problems (Caspi et al., 1995). In thepreviously cited Eisenberg et al. (2005a) study, loweradult ratings of child effortful control at 9 and11 years forecast higher levels of adult-rated exter-nalizing problems at age 13, extending similarresults obtained at earlier ages of measurement(Eisenberg et al., 1997). Importantly, Eisenberg’s(Eisenberg et al., 2005b) work on 4–7-year-oldsfollowed up two years later also reveals a reciprocalprocess, with changes in externalizing problemsfostering changes in regulatory abilities, includingeffortful control and impulsivity. Such evidenceunderscores the need to consider bi-directionalprocesses of influence in research on parenting,attentional control and problem behavior.

Given the work summarized above, it is note-worthy that several recent longitudinal studies,mostly of young adolescents, provide evidencethat effortful-control processes (partially) mediateConflict of interest statement: No conflicts declared.

Journal of Child Psychology and Psychiatry 48:12 (2007), pp 1233–1242 doi:10.1111/j.1469-7610.2007.01807.x

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

relations between parenting and externalizingproblems. Brody and Ge (2001) found that observedand reported parenting at age 12 were indirectlyassociated with childhood alcohol use at age14 through their effect on self-regulation at age 13(i.e., ability to set goals, make plans, pay attentionto actions). In work on single-parent, African-Amer-ican families, Kim and Brody (2005) documentedsimilar (partial) mediation upon analyzing maternalreports of supportive parenting (age 13), self-regu-lation (age 14) and teacher-rated externalizingproblems (age 15), as did Eisenberg et al. (2005a)who assessed maternal warmth (age 9), effortfulcontrol (age 11) and adult-reported externalizingproblems (age 13).

Despite these encouraging findings, several designfeatures limit the confidence that can be placed inresults reported. Perhaps most important is that thesame respondents –parents or teachers –have oftenprovided information on children’s effortful controland behavior problems (e.g., Eisenberg et al.,2005b). Additionally, multiple sources of informationconcerning effortful control obtained from parents,teachers and/or structured observations or assess-ments are often combined, including cognitive-experimental ones like the Stroop, thereby obscuringthe role that more precisely definedmechanismsmayplay in fostering externalizing problems (e.g.,Kochanska et al., 2000). Studies adopting thisapproach trade off conceptual purity and experi-mental control for breadth of measurement, leavingunclear just how important the capacity to inhibit aresponse is in children’s externalizing behaviorproblems and inmediating at least some of the effectsof social influences.

Nevertheless, some studies do document linkagesbetween experimental measures of cognitive controland behavior problems (e.g. Hughes et al., 1998),with one suggesting that performance-based assess-ments of attentional control at 54 monthsmediate theeffect of observed and reported parenting (54 months)on teacher-reported externalizing problems (1stgrade: �6 years) (NICHD ECCRN, 2003). Morespecifically, errors of commission on a continuousperformance task at 54 months partially mediatedthe effect of the family rearing environment (i.e.,maternal sensitivity, cognitive stimulation)measuredprior to grade school on externalizing behaviorreported by caregivers at 54 months and by 1st-gradeteachers. Errors of commission in continuousperformance tasks appear to provide a good index ofattentional control abilities, particularly thoseimplicated in response inhibition (Overtoomn et al.,1998; Plizska, Liotti, & Woldorff, 2000). What limitsthe aforementioned work (i.e., NICHDECCRN, 2003),however, and stimulates the current inquiry was thefact that the parenting predictor and the attentionmediator were measured contemporaneously,substantially weakening the case for mediation, andfocused on a limited age range.

Here we overcome these limits by taking advantageof repeated observational measurements of maternalbehavior from 54 months through 5th grade(�10 years), of tested attentional abilities from54 months through 4th grade (�9 years), and ofteacher-reported externalizing behavior from54 months through 5th grade. This affords testing,via longitudinal structural equation modeling (Cole& Maxwell, 2003), whether change in attentionalabilities mediates partially or fully the effect ofparenting on change in externalizing behavior.Furthermore, predictor, mediator and outcome areindependently assessed, eliminating common sourcevariance as a contaminating factor. Finally, we testwhether attentional control partially or fully medi-ates the effect of maternal sensitivity on externalizingproblems, as well as any reciprocal processes, at twoseparate lags during early/middle childhood:54 months fi 1st grade fi 3rd grade and 1stgrade fi 4th grade fi 5th grade. Being positioned toevaluate both of these time lags is important becausemajor changes in executive function, a critical com-ponent of which is attentional control, occur withinthe 3–7-year period (which stabilizes significantlythereafter, see Rueda et al., 2004), thereby leadingto the expectation that parenting and mediationaleffects will be more apparent earlier than later dur-ing the middle-childhood years. Given the likelihood,however, that psychological processes other thanattentional control are involved in mediating link-ages between parenting on externalizing problems(e.g., emotion regulation, attributional bias), it isexpected that attentional control will partially ratherthan fully mediate parenting/behavior-problemsrelations.

We rely on a composite parenting construct ofmaternal sensitivity which includes indices ofpositive and negative parenting (e.g., warmth,hostility) because (a) it was repeatedly measured byobservational means, incorporating age-appropriateadjustments, making it ideal for a longitudinalinquiry seeking to determine whether similardevelopmental processes operate across time(NICHD ECCRN, 2005); and (b) it has proven pre-dictive of children’s functioning, including attentionand externalizing problems, when extensive con-trols have been included in short- and long-termprediction models (e.g., NICHD ECCRN, 2003,2005). Due to its documented relation to infantattachment security (NICHD ECCRN, 1997) and thefact that the composite construct has been referredto as ‘maternal sensitivity’ in many reports, weretain that label here. Further, because our per-formance-based attentional-control measurementswere not obtained prior to 54 months, we focusexclusively on parenting from this point forward,though it should be appreciated that maternalsensitivity at this point in time is significantlyassociated with earlier measurements of sensitivity(i.e., 6/15/24/36 months).

1234 Jay Belsky, R.M. Pasco Fearon, and Brian Bell

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

Methods

Participants

Data for this study were collected as part of the firstthree phases (11 years) of the NICHD Study of EarlyChild Care and Youth Development (SECCYD) (NICHDECRN, 2005a). Ten research sites recruited particip-ants from 31 hospitals during the first 11 months of1991. A total of 8,986 mothers were screened in thehospital and 5,416 (60%) were eligible and agreed tobe subsequently contacted. A conditionally randomsample of 3015 mothers (56%) was selected from these5,416 to insure representation (at least 10%marginally)of single parent households, mothers with less than ahigh school education, and ethnic minority mothers.Finally, in 1991, 1,364 (89%) of the 1,526 completedthe 1-month enrollment interview, becoming studyparticipants who provided informed consent repeatedly.These 1,364 participants were reasonably diverse,including 52% male children, 24% ethnic-minoritychildren, 4.5% with mothers who had not completedhigh school and 14% with single parents at the time ofthe child’s birth, but non-random attrition led to moredata being gathered on the more rather than the lessadvantaged children/families (see below). In any event,the current sample includes a relatively small percent-age of children showing clinically meaningful levels ofADHD or more serious forms of disruptive behavior(e.g., conduct disorder). Thus, the generalizability offindings to populations of children with clinicallymeaningful problems remains unclear. Further detailsof sample recruitment and selection are presented inalready-cited publications.

Procedures

Children were followed from 54 months to when theywere in 5th grade (�10 years). Parenting measures usedin this report were obtained at 54 months and at 1stgrade (�6 years), 3rd grade (�8 years) and 5th grade(�10 years); CPT data were obtained at 54 months and1st and 4th grades (�9 years); child behavior problemswere measured at 54 months and 1st, 3rd, and 5thgrades. Procedures for all data collections are docu-mented in the study’s Manuals of Operation (http://public.rti.org/sccc/).

Measures

Parenting: Maternal sensitivity Parenting wasstudied using somewhat different, but related, obser-vational procedures, all of which were videotaped forlater coding, across the period 54 months to 5th grade.Three mother–child activities were videotaped during auniversity-laboratory visit at 54 months (i.e., complet-ing Etch-A-Sketch maze, building block towers, playingwith puppets) and in 1st grade (i.e., Etch-A-Sketchhouse drawing, patterned block building, card game).Videotaping took place at home in 3rd grade whilemother and child planned a trip around a make-believetown to do errands, whereas in 5th grade it occurredwhile the dyad discussed three topics they identified assources of disagreement in their family (e.g., chores,sibling conflict) and built a tower using toothpicks. Also

videotaped were entire-family discussions of how tospend money during a fantasy vacation.

At each age, separate teams of trained observers,working at a single site and blind to all other informa-tion on a family, used age-adjusted 7-point ratingscales to measure three global dimensions of parenting(each operationalized using multiple scales): SupportivePresence (e.g., sensitivity to non-distress signals),Respect for Autonomy (e.g., intrusiveness) and Hostility(e.g., negative regard for child). In previous work on thissample the scales have proven to be substantially inter-correlated, leading to their compositing (i.e., SP+RA-H)to generate an index of sensitivity repeatedly found torelate to other family characteristics (e.g., income,maternal depression) and longitudinally predict childfunctioning in grade school (NICHD ECCRN, 2004,2005). These three scales are used to represent a singleunderlying latent variable broadly representing overallquality of parenting.

Attentional control. The CPT, modeled on a versiondescribed by Mirsky and his colleagues (Mirsky,Anthony, Duncan, Ahearn, & Kellam, 1991; Rosvold,Mirksy, Sarason, Bransome Jr., & Beck, 1956), wasused to measure attentional control at 54 months and1st and 4th grades. The task required children to pressa button whenever a target stimulus appeared on acomputer screen. Ten stimuli were presented in each of22 blocks at 54 months and in each of 30 blocks at 1stgrade; 12 stimuli were presented in each of 45 blocks at1st and 4th grades. Stimulus duration was 500 msec.with an interstimulus interval of 1500 msec. at54 months and, at 1st and 4th grades, 200 msec. withan interstimulus interval of 1500 msec. The targetstimulus was randomly presented twice in each blockwith all response data automatically tallied by thecomputer.

Following the earlier paper by the NICHD ECCRN, weselected commission errors (i.e., responses to non-tar-gets) as a single indicator of attentional control, as thiswas previously found to be associated with children’sbehavioral problems and has been used extensively inother clinical and developmental studies. Commissionerrors in CPT tasks appear to tap aspects of attentionrelated to response inhibition, as indicated by behav-ioral and ERP data (Overtoom et al., 1998; Plizskaet al., 2000). The test–retest reliability of the CPT isgood in the age range studied in this research (Halperinet al., 1991) and commission errors discriminatebetween children with and without attention deficitdisorder with relative consistency (Barkley et al., 1992;Campbell et al., 1994).

Externalizing behavior problems. To insure no con-ceptual overlap between mediational and outcomeconstructs, all subscales (and thus items) having to dowith attentional problems or processes were excludedfrom the well-validated instruments used to assessexternalizing behavior problems. At 54 months care-givers (i.e., not parents) completed the Teacher ReportForm Age 2–5 version (Achenbach, 1991) and anabbreviated version of the Child Behavior Question-naire (Rothbart et al., 2001). These yielded threesubscale scores to serve as latent-variable indicators:TRF, Aggression (e.g., destroys own things, attacks

Parenting, attention and externalizing problems 1235

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

people, disobedient, fights), TRF Immaturity (e.g.,frustrated, feelings hurt, sulks) and CBQ Anger/frus-tration subscale (e.g., has temper tantrums whendoesn’t get what he/she wants, gets angry when can’tfind something, gets quite frustrated when preventedfrom doing something). In 1st, 3rd and 5th grades, threesubscales of the TRF were used as latent-variableindicators: Aggression (e.g., argues, fights, tempertantrums, threatens), Social Problems (e.g., doesn’tget along with other pupils, feels others are out to gethim/her) and Delinquency (e.g., lies and cheats,swears).

Missing data

The longitudinal data contained a significant amount ofmissing data for a variety of reasons, including loss ofparticipants with time. Table 1 shows that poor familiesand those with young and/or single mothers were morelikely to have missing data. Excluding cases withmissing data from analyses can undermine statisticalpower and bias parameter estimates (Allison, 2003).Thus, the full information maximum likelihood (FIML)method, which uses all the available data to estimatethe parameter estimates of a model (by calculating thelog-likelihood of the data for each observational unitseparately), was used to estimate missing data (Allison,2003).

Data analysis

Maximum likelihood structural equation modeling(SEM) techniques were used to test for longitudinalmediation. Because structural equation models allowthe examination of the interrelationships betweenmultiple dependent and independent variables as wellas the impact of measurement error, they are ideallysuited for testing mediational processes. Moreover,longitudinal data makes it possible to examine whethernew variance in a mediator or outcome is predicted fromearlier variance in a predictor or mediator, respectively,after accounting for stability of and prior associationsbetween the variables. Here, we follow Cole and Max-well’s (2003) guidelines for testing mediationalhypotheses with longitudinal data, using the SEMsoftware AMOS 5.0 (Arbuckle, 2003).

Results

We begin by examining structural equation modelsthat test within-construct longitudinal mediation(i.e., stability within constructs over time). There-after, we test for cross-domain associations betweenhypothesized causal factors and hypothesized con-sequences forward in time, directly testing medi-ation. Finally, we examine reciprocal pathways, withconstructs operating in the opposite direction to thatoriginally evaluated.

Within-domain longitudinal models

The first step was to fit longitudinal models forthe domains measured using multiple indicators(parenting sensitivity and externalizing behaviorproblems). These models specified that cross-timestability in the measured variables was carried by asingle latent variable (i.e., an order 1 autoregressivemodel). This also allowed an evaluation of whetherthere existed cross-time correlated errors of meas-urement for each measured variable. As attentionalcontrol was only measured by a single observedvariable, stability could be estimated from the rawcorrelations and therefore latent variable modelingwas not conducted. Attentional control correlatedr ¼ .26 (p < .01) between 54 months and 1st gradeand r ¼ .43 (p < .01) between 1st and 4th grade.

Maternal sensitivity. For observed maternal sensit-ivity, there were significant correlated errors amongthe measures of maternal hostility between all adja-cent time points, as well as between maternal sup-portive presence at 54 months and 1st grade. Withthese included in the model, the overall model wasstill highly significant (v2(47) ¼ 251.4,p < .01), due tothe large sample size, but the RMSEA suggested goodfit (RMSEA ¼ .056). The Normed Fit Index (NFI) was.96. The parameter estimates and standard errors forthe paths between the sensitivity latent variables ateach time point are shown in Table 2, revealing a highdegree of stability in maternal sensitivity.

Table 1 Demographic correlates of missing data

Associated with missing data (p-value)

Maternal ageChild gender Maternal ethnicity Single parent Income-to-needs

Parenting 54 months 4 (.026) 4 (.037) 4 (<.001) 4 (.001) 4 (<.001)Attention 54 months 4 (.038) 4 (.017) 4 (<.001) 4 (<.001) 4 (<.001)Externalizing 54 months 4 (.050) 4 (.003) 4 (<.001) 4 (<.001) 4 (<.001)Parenting Grade 1 4 (.028) 8 4 (.001) 4 (.002) 4 (<.001)Attention Grade 1 4 (.040) 8 4 (.007) 4 (.005) 4 (<.001)Externalizing Grade 1 4 (.026) 4 (.003) 4 (.001) 4 (<.001) 4 (<.001)Parenting Grade 3 8 8 4 (.009) 4 (.013) 4 (<.001)Attention Grade 4 4 (.004) 8 4 (.007) 8 4 (<.001)Externalizing Grade 3 4 (.018) 4 (.027) 4 (.003) 4 (<.001) 4 (<.001)Parenting Grade 5 4 (.010) 8 4 (.007) 4 (.010) 4 (<.001)Externalizing Grade 5 4 (.047) 4 (.002) 4 (.002) 4 (.001) 4 (<.001)

Note: 4 association with presence of missing data significant at p < .05; 8 no significant association with presence of missing data atp < .05.

1236 Jay Belsky, R.M. Pasco Fearon, and Brian Bell

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

Behavior problems. The longitudinal model fittingfor behavior problems revealed correlated errors foreach measure between 1st and 3rd grades andbetween 3rd and 5th grades, though none between54 months and 1st grade. Subsequently, overallmodel fit, while significant given the large samplesize (v2(44) ¼ 223.2, p < .01), was good according tothe RMSEA statistic (.055) and NFI (.96). Table 2indicates that stability of behavior problems in-creased somewhat after 1st grade.

Cross-domain longitudinal models

Parenting, CPT attentional control and behaviorproblems To test the hypothesis that parentinginsensitivity affects behavior problems via its influ-ence on children’s attentional control, a series ofstructural equation models were tested. Initially, abase model was tested in which all three within-do-main longitudinal models were estimated together inthe absence of any cross-domain paths connectingearly and later time points (though initial levels wereallowed to freely correlate). This model fitted the datareasonably well (v2(305) ¼ 1057.7, p < .01;RMSEA ¼ .043; NFI ¼ .92). However, inclusion ofcross-domain paths from sensitivity to attentionalcontrol and from attentional control to behaviorproblems – the predicted contributory pathways –resulted in a highly significant improvement inmodel fit (Model v2(300) ¼ 913.9, p < .01; Changev2(5) ¼ 143.8, p < .01).

Despite the moderate and highly significant asso-ciations between all three latent variables at54 months, Figure 1a shows, as expected, thatgreater sensitivity significantly predicted moreattentional control over time, while more attentionalcontrol significantly predicted less problems overtime (all cross-domain paths are significant atp < .001). When direct paths were added from sen-sitivity to behavior problems (at lag 2), there was ahighly significant improvement in model fit (Modelv2(298) ¼ 871.1, p < .01; Change v2(2) ¼ 42.8,p < .01), clearly suggesting, as expected, that medi-ation was only partial. Mediation tests can only beconducted once these paths have been added. Thestandardized and unstandardized structural pathscoefficients for this model are shown in Table 3(under ‘Model 1’).

The best estimate of mediation in a multi-wavelongitudinal design is the overall indirect effect of

sensitivity at 54 months on behavior problems in 5thgrade (Cole & Maxwell, 2003). This is constituted bythe sum of all mediating pathways, which summedto b ¼ ).22. As no significance test of such an overallindirect effect exists, each contributing pathway wastested separately and found to be significant (54 m-Sensitivity fi G3-BP: Sobel test z ¼ 4.23, p < .01;G1-Sensitivity fi G5-BP: Sobel test z ¼ 2.22, p ¼.03).

To examine the impact that possible time-specificcorrelations between the latent variables might havehad on mediation tests, the model was re-estimatedwhile allowing all disturbance terms in the latentvariables at the same measurement occasion tocorrelate. (In the case of CPT commission errors at4th grade, the disturbance term was correlated withsensitivity and behavior problems at 3rd grade.) Theaddition of correlated disturbances led to a signifi-cant reduction in model v2(Model v2 (291) ¼ 826.3,p < .01; Change v2 (7) ¼ 44.8, p < .01). The stan-dardized parameter estimates for the structuralaspects of this model are shown in Figure 1b andthe unstandardized estimates and standard errors inTable 3. With correlated disturbance terms included,both mediated pathways remained significant (54 m-Sensitivity fi G3-BP: Sobel test z ¼ 4.04, p < .01;G1-Sensitivity fi G5-BP: Sobel test z ¼ 2.16, p ¼.03). The overall indirect effect from sensitivityat 54 months to behavior problems at Grade 5 wasb ¼ )19.

Testing for bidirectional associations. When con-sidering longitudinal mediation, it is advisable toexamine possible causal paths operating in theopposite direction (Cole & Maxwell, 2003). We thusexamined change in model fit when these cross-mediational paths were added to the last modeldescribed (see bottom of Table 3), in which direct,indirect effects and correlated disturbances wereincluded. The addition of the cross-mediationalpaths (from Behavior Problems to Sensitivity viaAttentional Control, and the respective direct paths)led to a further significant improvement in model fit(Model v2(284) ¼ 759.3, p < .01; Change v2 (7) ¼67.0, p < .01). The standardized path coefficientsfor the structural parts of this model are shown inFigure 1c (and Table 3).

As evident in Figure 1c, neither of the reversepathways from behavior problems to attentionalcontrol were significant, though two of three paths

Table 2 Stability coefficients and model fit statistics for within-domain longitudinal analyses

Stability parameters Model fit statistics

B12 B23 B34

v2 RMSEAb B s.e. b B s.e. b B s.e.

Maternal sensitivity (Sens) .58 .62 .038 .59 .50 .030 .53 .49 .032 253.2 .056Behavior Problems (Beh Probs) .46 .47 .040 .55 .66 .041 .57 .54 .034 223.2 .055

Parenting, attention and externalizing problems 1237

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

from children’s attentional control to maternal sen-sitivity were significant (p < .01). With these effectsincluded, the test of mediation for sensitivity at54 months on behavior problems at 3rd grade (via1st-grade CPT attentional control) was significant(Sobel test z ¼ 3.34, p < .01), as was the secondmediational pathway (via 4th-grade CPT attentionalcontrol) (Sobel’s test z ¼ 2.29, p ¼ .02). The total

mediating effect of sensitivity on behavior problems –via attentional control –was estimated to be b ¼ )18.Because neither of the reverse effects of behaviorproblems on attentional control were significant,tests of mediation from behavior problems tomaternal sensitivity were non-significant, thoughboth direct paths from behavior problems to sensit-ivity were significant, suggesting some reciprocal

Sens54 SensG1 SensG5SensG3.59 .59 .53

Prob54 ProbG1 ProbG5ProbG3.44 .52 .55

.23 .40

-.22

.21

-.13

.20 .13

-.26

.26

-.30CPT54 CPTG1 CPTG4

Sens54 SensG1 SensG5SensG3.59 .59 .53

Prob54 ProbG1 ProbG5ProbG3.43 .46 .52

.23 .39

-.19

.20

-.12

.15 .10

-.27

.20

-.27

-.20 -.09ns

CPT54 CPTG1 CPTG4

Sens54 SensG1 SensG5SensG3.55 .54 .48

Prob54 ProbG1 ProbG5ProbG3.45 .47 .53

.22 .38

-.17

.20

-.10

.15 .12

-.26

.21

-.28

-.09ns

.10ns

-.16

.07ns

-.12-.10

-.19

-.05ns

-.10

CPT54 CPTG1 CPTG4

(a)

(b)

(c)

Figure 1 Cross-domain longitudinal associations between maternal sensitivity, children’s CPT Attention andBehavior Problems (all paths are significant p < .01). (1a) basic cross-domain longitudinal model. (1b) Longitudinalmediation model (with correlated disturbances, not shown). (1c) Reverse causation model

1238 Jay Belsky, R.M. Pasco Fearon, and Brian Bell

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

influences not mediated by children’s attentionalcontrol.

Discussion

We sought to advance understanding of the interre-lation of parenting, attentional control abilities andexternalizing problems by specifically testing theproposition that insensitive parenting would under-mine children’s attentional control abilities and,thereby, foster externalizing problems, appreciatingthat such mediational inquiry has focused princi-pally on adolescence to date. We did this (1) byfollowing children from 54 months to the age of 10;(2) by relying on completely independent measure-ments of parenting, attention and problem behavior;(3) by examining two separate parenting fi atten-tion fi externalizing time lags (i.e., 54 months fi 1stgrade fi 3rd grade; 1st grade fi 3rd grade fi 5thgrade); and (4) by exploring reciprocal processeswhereby problem behavior might affect attentionand, thereby, parenting. Results were in line withexpectations in this investigation in which parentingwas measured observationally and repeatedly at fourdifferent time points (i.e., 54 months, 1st/3rd/5thgrades), attentional control was measured via CPTcommission errors at three different time points (i.e.,54 months, 1st/4th grades) and teacher-reportedexternalizing problems were measured repeatedly bydifferent teachers (54 months, 1st/3rd/5th grades).

Lower levels of maternal sensitivity at 54 monthsand in 1st grade predicted, respectively, poorerattention control in 1st and 3rd grades. Poorerattentional performance at 54 months and in 1stand in 4th grade predicted, respectively, moreexternalizing problems in 1st, 3rd and 5th grades.Most significantly, attentional control in 1st gradepartially mediated the effect of insensitive parentingat 54 months on externalizing problems in 3rd

grade, and the effect of sensitivity in 1st grade onproblem behavior in 5th grade. Importantly, thesignificant links just cited emerged with earliermeasurements of the predictors (i.e., parenting),mediators (i.e., attentional control) and outcomes(i.e., externalizing problems) controlled, therebyindicating that changes in antecedent variablespredicted changes in their examined sequelae (i.e.,parenting predicting attentional control, attentionpredicting externalizing problems). Within the con-text of a non-experimental, longitudinal field study,this is perhaps the strongest basis for inferringcausation. Important to note, however, is that inno case did attentional performance fully mediateparenting-behavior-problems linkages, nor was thatexpected. Given that psychological processes likeemotion (dys)regulation and attributional bias areassociated with both unsupportive/harsh parentingand problem behavior, there are clearly additionalmediating mechanisms that merit examination inattempt to account for the linkage between parentingand problem behavior that were not mediated byattentional performance in this inquiry.

Limitations

Despite the strengths of this study, it is risky toembrace causal conclusions without direct experi-mental manipulation, even with longitudinal dataused to illuminate the mediational effects of changein a predictor on change in an outcome. Unmeasuredthird variables, perhaps closely correlated withattentional control, could be responsible for thechanges over time observed in this work, includinggenetic factors. Also limiting the present work is thatthe analytic methods chosen to handle situationswhere data are missing systematically make anumber of assumptions that may not be correct.Furthermore, by aiming the analysis at a large but

Table 3 Longitudinal mediation model statistics

Longitudinal cross-domain model parameter estimates

B12 B23 B34

b B s.e. b B s.e. b B s.e.

Model 1Sens–Attn ).21 ).009 .001 ).13 ).003 .001Attn –Beh Probs .21 9.17 1.44 .15 20.83 4.34 .11 23.18 7.02Sens–Beh Probs ).21 )1.23 .20 ).10 ).53 .18Model 2 With Correlated DisturbancesSens–Attn ).19 ).008 .001 ).12 ).003 .001Attn –Beh Probs .20 8.84 1.44 .15 20.54 4.38 .10 21.82 6.99Sens–Beh Probs ).20 )1.15 .19 ).09 ).45 .18Model 3 With Bidirectional EffectsSens–Attn ).17 ).007 .001 ).10 ).002 .001Attn –Beh Probs .20 8.96 1.44 .15 21.08 4.40 .11 24.36 7.03Sens–Beh Probs ).19 )1.12 .19 ).09 ).44ns .18Beh Probs–Attn .10 .001ns .001 .07 .001ns .001Attn –Sens ).10 ).88 .29 ).05 )1.14ns .69 ).10 )3.14 1.06Beh Probs–Sens ).16 ).029 .008 ).12 ).020 .001

Parenting, attention and externalizing problems 1239

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

relatively low-risk sample, the results of the currentstudy may not generalize to other communities orhigher-risk subgroups, including clinical samples.

Implications for intervention

The findings presented suggest at least two strate-gies for intervening to prevent the developmentof externalizing problems or to remediate problemsalready apparent. These, of course, should be testedvia experimental manipulation before any conclu-sions regarding their efficacy are drawn. First, effortsto enhance parental sensitivity in the late preschoolyears should contribute to enhanced attentionalabilities early in elementary school and, thereby,prevent or reduce externalizing behaviors otherwiseevident in 3rd grade. Such efforts, including thoseundertaken in 1st grade, could also directly con-tribute to the prevention/reduction of problembehavior, given evidence showing direct, unmediatedeffects of insensitive parenting on problem behavior.As it seems doubtful that such effects are trulyunmediated – by anything – future work shouldconsider mediators other than those examined inthis inquiry, as already indicated.

The fact that attentional control processes in 1stgrade mediated some of the effect of earlier parentingon later externalizing problems suggests that directefforts to improve children’s attentional capabilitieswhen they are in 1st grade (and perhaps earlier)should also be considered. Rueda et al. (2005) pro-vide one example of a computer-assisted trainingregimen that proved effective with 4- and 6-year olds.

The role of timing

Of particular interest with respect to such efforts topromote attentional skills directly, or even indirectlyvia parenting, are results showing that the direct aswell as mediational influence of attentional controlon problem behavior decreased somewhat over time(see Figure 1b). Recall that attentional controlprocesses in 4th grade initially failed to mediate theeffect of parenting in 1st grade on behavior problemsin 5th grade, though evidence of such mediation didemerge when reciprocal effects were added to thestructural equation model in the final analytic step(see Figure 1c). The fact that attentional processesappear to affect problem behavior somewhat morestrongly at younger than older ages – at least withinthe confines of the developmental assessmentsundertaken in this inquiry – suggests that efforts toenhance attentional abilities might prove moreeffective, at least in terms of their impact on problembehavior (and especially with respect to mediatingparenting effects), if undertaken earlier rather thanlater in childhood. Such reasoning would seemconsistent with theory and evidence indicating thatmajor changes in executive functioning, a compon-

ent of which is effortful (attentional) control, takeplace within the 3–7-year period (Rueda et al., 2004).That further changes in executive functions occurduring adolescence raises the prospect, in fact, thatefforts to ameliorate attentional deficits might bebest targeted before age eight and during adoles-cence, rather than in between. Before this proposi-tion is embraced, it should be experimentally tested.

Reciprocal effects

Most research on parenting, attention and problembehavior has addressed the influence of parenting onattention or of attention on problem behavior, withless consideration given to reciprocal effects. OnlyEisenberg et al. (2005b) have investigated suchdevelopmental processes, finding that problembehavior at age 4–7 predicted effortful control twoyears later, but failing to discern effects of effortfulcontrol on parenting in another study of 9–13-year-old children (Eisenberg et al., 2005a). The presentresults are entirely inconsistent with these findings.Whereas neither pathway linking earlier behaviorproblems to children’s later attentional controlcapacities proved significant, all three paths linkingearlier attentional control to less sensitive parentinglater in time proved significant. Quite conceivably,differences in the designs of the current research andEisenberg et al.’s (2005a, b) account for the differ-ences in findings. In our inquiry, in contrast toEisenberg et al. (2005a, b), children were tested atthe same age rather than across a range of ages (i.e.,4–7, 9–13) and attention was measured by means ofthe CPT rather than via parent/teacher reports.Irrespective of whether these design differencesaccount for the variation in research results acrossinvestigations, it should be clear that bidirectionalprocesses merit consideration in efforts to under-stand the interrelation of parenting, attention andproblem behavior. Even though no evidence emergedto support the proposition that attentional processesmediate effects of problem behavior on parenting,results clearly indicated that some of the relationsinvestigated were not unidirectional.

Conclusion

The study of attention and of self-regulation moregenerally provides a potentially important modelsystem for investigating the psychological, and pos-sibly biological, mechanisms that mediate betweensocial-contextual forces and children’s problembehavior. Given evidence that attentional controlskills are heritable (e.g., Rueda et al., 2005; Fosellaet al, 2002), the results of our and others’ researchsuggest that attentional control may represent acommon pathway by which social and genetic factorsinfluence children’s emotional and behavioral func-tioning. In light of the relatively well-understood

1240 Jay Belsky, R.M. Pasco Fearon, and Brian Bell

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

neural basis of attentional self-control (Posner &Rothbart, 2000), these interconnections betweenrelationships, cognition and functioning could pro-vide a valuable means of understanding the complexmechanisms by which nature and nurture interactover the course of children’s development.

Acknowledgements

The research described herein was supported by acooperative agreement with the National Instituteof Child Health and Human Development(U10-HD25420). The authors express appreciationto all collaborating NICHD-Study investigators andparticipating children/families.

Correspondence to

Jay Belsky, Institute for the Study of Children,Families and Social Issues, Birkbeck University ofLondon, 7 Bedford Square, London WC1B 3RA, UK;Email: [email protected]

References

Achenbach, T.W. (1991). Integrative guide of the 1991CBCL/4-18, YSR & TRF Profiles. Burlington: Univer-sity of Vermont, Department of Psychiatry.

Allison, P.D. (2003). Missing data techniques forstructural equation modeling. Journal of AbnormalPsychology, 112, 545–557.

Arbuckle, J.L. (2003). Amos 5.0 update to the AmosUser’s Guide. Chicago, IL: Smallwaters Corporation.

Barkley, R.A., Grodzinsky, G.M., & DuPaul, G.J.(1992). Frontal lobe functions in attention deficitdisorder with and without hyperactivity: A review andresearch report. Journal of Abnormal Child Psycho-logy, 20, 163–188.

Brody, G.H., & Ge, X. (2001). Linking parenting pro-cesses and self-regulation to psychological function-ing and alcohol use during early adolescence. Journalof Family Psychology, 15, 82–94.

Brody, G.H., Murry, V.M., McNair, L., & Chen, Y.(2005). Linking changes in parenting to parent–childrelationship quality and youth self-control: Thestrong African-American families program. Journalof Research on Adolescence, 15, 47–69.

Campbell, S.B., Pierce, E.W., March, C.L., Ewing, L.J.,& Szumowski, E.K. (1994). Hard-to-manage pre-school boys: Symptomatic behavior across contextsand time. Child Development, 65, 836–851.

Caspi, A., Henry, B., McGee, R.O., Moffitt, T.E., & Silva,P.A. (1995). Temperamental origins of child andadolescent behavior problems: From age three toage fifteen. Child Development, 66, 55–68.

Cole, D.A, & Maxwell, S.E. (2003). Testing mediationalmodels with longitudinal data: Questions and tips inthe use of structural equation modelling. Journal ofAbnormal Psychology, 112, 558–577.

Deater-Deckard, K., & Dodge, K. (1997). Externalizingproblems and discipline revisited: Nonlinear effects

and variation by culture, context and gender. Psy-chological Inquiry, 8, 161–175.

Eisenberg, N., Fabes, R.A., Shepard, S.A., Murphy,B.C., Guthrie, I.K., Jones, S. et al. (1997). Contem-poraneous and longitudinal prediction of children’ssocial functioning from regulation and emotionality.Child Development, 68, 642–664.

Eisenberg, N., Sadovsky, A., Spinrad, T.L., Fabes, R.A.,Losoya, S.H., Valiente, C. et al. (2005b). The relationsof problem behavior status to children’s negativeemotionality, effortful control, and impulsivity:Concurrent relations and prediction of change.Developmental Psychology, 41, 193–211.

Eisenberg, N., Zhou, Q., Spinrad, T.L., Valiente, C.,Fabes, R.A., & Liew, J. (2005a). Relations amongpositive parenting, children’s effortful control, andexternalizing problems: A three-wave longitudinalstudy. Child Development, 76, 1055–1071.

Fosella, J., Sommer, T., Fan, J., Wu, Y., Swanson, J.M.,Pfaff, D., & Posner, M.I. (2002). Assessing the molec-ular genetics of attention networks. BioMed CentralNeuroscience, 3, 14.

Gadeyne, E., Ghesquiere, P., & Onghena, P. (2004).Longitudinal relations between parenting and childadjustment in young children. Journal of ClinicalChild and Adolescent Psychology, 33, 347–358.

Gauvain, M., & Huard, R.D. (1999). Family interaction,parenting style, and the development of planning:A longitudinal analysis using archival data. Journalof Family Psychology, 13, 75–92.

Halperin, J.M., Sharma, V., Greenblatt, E., & Schwartz,S.T. (1991). Assessment of the Continuous Perfor-mance Test: Reliability and validity in a nonreferredsample. Psychological Assessment. A Journal of Con-sulting and Clinical Psychology, .3, 603–608.

Hughes, C., Dunn, J., & White, A. (1998). Trick or treat?Uneven understanding of mind and emotion andexecutive dysfunction in ‘Hard-to-manage’ preschool-ers. Journal of Child Psychology and Psychiatry, 39,981–994.

Kim, S., & Brody, G.H. (2005). Longitudinal pathwaysto psychological adjustment among black youth livingin single-parent households. Journal of Family Psy-chology, 19, 305–313.

Kochanska, G., Murray, K.T., & Harlan, E.T. (2000).Effortful control in early childhood: Continuity andchange, antecedents, and implications for socialdevelopment. Developmental Psychology, 36, 220–232.

Mirsky, A.F., Anthony, B.J., Duncan, C.C., Ahearn,M.B., & Kellam, S.G. (1991). Analysis of the elementsof attention: A neuropsychological approach. Neuro-psychology Review, 2, 109–145.

NICHD Early Child Care Research Network. (1997). Theeffects of infant child care on infant–mother attach-ment security. Child Development, 68, 860–879.

NICHD Early Child Care Research Network. (2003). Dochildren’s attention processes mediate the linkbetween family predictors and school readiness?Developmental Psychology, 39, 581–593.

NICHD Early Child Care Research Network. (2004).Father’s and mother’s parenting behavior and beliefsas predictors of children’s social adjustment in thetransition to school. Journal of Family Psychology,18, 628–638.

Parenting, attention and externalizing problems 1241

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.

NICHD Early Child Care Research Network. (2005).Predicting individual differences in attention, mem-ory, and planning in first graders from experiences inhome, child care, and school. Developmental Psycho-logy, 41, 99–114.

Overtoom, C.C., Verbaten, M.N., Kemner, C., Kene-mans, J.L., Engeland, H.V. et al. (1998). Associationsbetween event-related potentials and measures ofattention and inhibition in the Continuous Perform-ance Task in children with ADHD and normal con-trols. Journal of the American Academy of Child andAdolescent Psychiatry, 37, 977–985.

Peiser, N.C., & Heaven, P.C.L. (1996). Family influ-ences on self-reported delinquency among highschool students. Journal of Adolescence, 19, 557–568.

Pettit, G.S., Bates, J.E., & Dodge, K.A. (1997). Sup-portive parenting, ecological context, and children’sadjustment: A seven-year longitudinal study. ChildDevelopment, 68, 908–923.

Plizska, S.R., Liotti, M., & Woldorff, M. (2000). Inhibit-ory control in children with attention-deficit/hyper-activity disorder: Event-related potentials identify theprocessing component and timing of an impairedright-frontal response-inhibition mechanism. Biologi-cal Psychiatry, 48, 238–246.

Posner, M.I., & Rothbart, M.K. (2000). Developingmechanisms of self-regulation. Development and Psy-chopathology, 12, 427–441.

Rosvold, H.E., Mirksy, A.F., Sarason, I., Bransome Jr.,E.D., & Beck, L.H. (1956). A continuous performancetest of brain damage. Journal of Consulting Psycho-logy, 20, 343–350.

Rothbart, M.K., Ahadi, S.A., & Evans, D.E. (2000).Temperament and personality: Origins and out-comes. Journal of Personality and Social Psychology,78, 122–135.

Rothbart, M.K., Ahadi, S.A., Hershey, K., & Fisher, P.(2001). Investigations of temperament at three toseven years: The Children’s Behavior Questionnaire.Child Development, 72, 1394–1408.

Rueda, R.M., Fan, J., Candliss, B.D., Halparin, J.D.,Gruber, D.B., Lercani, L.P. & Posner, M.I. (2004).Development of attentional networks in childhood.Neuropsychologia, 42, 1029–1040.

Rueda, M.R., Rothbart, M.K., Saccamanno, L. & Pos-ner, M.I. (2005). Training, maturation and geneticinfluences on the development of executive attention.Proceedings of the US National Academy of Sciences,102, 14931–14936.

Manuscript accepted 18 June 2007

1242 Jay Belsky, R.M. Pasco Fearon, and Brian Bell

� 2007 The AuthorsJournal compilation � 2007 Association for Child and Adolescent Mental Health.