The Relations of Effortful Control and Reactive Control to Children’s Externalizing Problems: A Longitudinal Assessment Carlos Valiente Nancy Eisenberg Cynthia L. Smith Mark Reiser Richard A. Fabes Sandra Losoya Ivanna K. Guthrie Bridget C. Murphy Arizona State University ABSTRACT In this study, we examined the role of negative emotionality as a moderator of the relations of effortful control and overcontrol (versus undercontrol) with children’s externalizing problem behaviors; we also examined the longitudinal relations among these This research was supported by grants from the National Institutes of Mental Health (1 R01 HH55052 and 1 R01 MH 60838) to Nancy Eisenberg and Richard Fabes and a Research Scientist Award from the National Institute of Mental Health (K05 M801321) to Nancy Eisenberg. The authors wish to thank the many students who assisted in this study, the parents and children involved, the principals and teachers in the Tempe, Kyrene, Mesa, Scottsdale, Gilbert, Chandler, Phoenix, Peoria, and Washington School Districts (and other districts with few teachers), especially the numerous teachers at Fuller, Hudson, Meyer, and Rover schools. The authors would also like to thank Stephanie Shepard for her assistance in the data collection. Correspondence concerning this article may be addressed to Nancy Eisenberg, Dept. of Psychology, Arizona State University, Tempe, AZ 85287-1104. Tel.: 480-965- 5217; Fax: 480-965-8544. Email: [email protected]. Journal of Personality 71:6, December 2003. Blackwell Publishing 2003
Transcript
The Relations of Effortful Control and Reactive
Control to Children’s Externalizing Problems:
A Longitudinal Assessment
Carlos Valiente
Nancy Eisenberg
Cynthia L. Smith
Mark Reiser
Richard A. Fabes
Sandra Losoya
Ivanna K. Guthrie
Bridget C. Murphy
Arizona State University
ABSTRACT In this study, we examined the role of negativeemotionality as a moderator of the relations of effortful control andovercontrol (versus undercontrol) with children’s externalizing problembehaviors; we also examined the longitudinal relations among these
This research was supported by grants from the National Institutes of Mental Health
(1 R01 HH55052 and 1 R01 MH 60838) to Nancy Eisenberg and Richard Fabes and a
Research Scientist Award from the National Institute of Mental Health (K05
M801321) to Nancy Eisenberg. The authors wish to thank the many students who
assisted in this study, the parents and children involved, the principals and teachers in
the Tempe, Kyrene, Mesa, Scottsdale, Gilbert, Chandler, Phoenix, Peoria, and
Washington School Districts (and other districts with few teachers), especially the
numerous teachers at Fuller, Hudson, Meyer, and Rover schools. The authors would
also like to thank Stephanie Shepard for her assistance in the data collection.
Correspondence concerning this article may be addressed to Nancy Eisenberg,
Dept. of Psychology, Arizona State University, Tempe, AZ 85287-1104. Tel.: 480-965-
variables. Teachers’ and parents’ reports of children’s negative emotion-ality, effortful control, overcontrol and externalizing problem behaviorswere obtained at T1 (N5 199;M age5 89.51 months) and again 2 (T2) and4 years (T3) later. In addition, children’s effortful control was assessed withan observed measure of persistence. In a T3 concurrent structural equationmodel, effortful control, but not overcontrol, was negatively related tochildren’s T3 externalizing problem behaviors. In regression analyses, thenegative relation between T3 effortful control and externalizing problembehaviors was strongest at high levels of T3 negative emotionality. In thebest-fitting longitudinal structural equation model, both T1 effortful controland T1 overcontrol negatively predicted externalizing problems at T1,whereas T3 effortful control (but not T3 overcontrol) was significantlynegatively related to T3 externalizing problem behaviors when controllingfor T1 externalizing problem behaviors.
Developmental theorists frequently have suggested that aspects of
children’s temperamental emotionality and regulation/control arerelated to the development of their externalizing problem behaviors
(Eisenberg, Guthrie, et al., 2000; Rothbart & Bates, 1998). However,there have been relatively few attempts to examine the uniquecontributions of different types of regulation or control and of
emotionality to externalizing problem behaviors (Eisenberg et al.,1997; Eisenberg, Guthrie, et al., 2000; White et al., 1994) or to
examine how these relations change over time. In addition, despiteRothbart and Bates’ (1998) discussion of the importance of
examining interactions of different temperamental characteristics(e.g., emotionality and regulation), few investigators have done so.
The purpose of the present study was to address these issues.
Types of Control
Despite the fact that there are numerous definitions of emotion-
related regulation, there is some consistency in these definitions. Forexample, Thompson (1994) proposed that emotion regulation is the‘‘extrinsic and intrinsic processes responsible for monitoring,
evaluating, and modifying emotional reactions, especially theirintensive and temporal features, to achieve one’s goals’’ (pp. 27–28).
Similarly, Cicchetti, Ganban, and Barnett (1991) defined emotionregulation as ‘‘the intra- and extraorganismic factors by which
emotional arousal is redirected, controlled, modulated, and modifiedto enable an individual to function adaptively in emotionally
1172 Valiente et al.
arousing situations’’ (p. 15). Stemming from this work, and the work
of Cole and colleagues (Cole, Michel, & Teti, 1994; Campos,Mumme, Kermoian, & Campos, 1994), we define emotion-related
regulation as the process of initiating, avoiding, inhibiting, main-taining, or modulating the occurrence, form, intensity, or duration
of emotion, generally in the service of accomplishing one’s goals(Eisenberg & Morris, 2002).
In thinking about the nature of emotion-related regulation, wefind it useful to differentiate between two related constructs, controland regulation. Control pertains to constraint or inhibition or the
lack thereof (e.g., strong reactive approach tendencies as reflected inimpulsivity). The term ‘‘control’’ does not indicate the nature of the
constraint (or impulsivity)—that is, whether it is based on voluntaryor a less voluntary, reactive processes. Self-regulation includes
control, but only control that is effortfully modulated and managedso it can be used in a flexible, adaptive way to meet one’s goals
(Eisenberg & Morris, 2002).
Effortful control. Self-regulation (but not necessarily regulationbased on extraorganismic factors), as we view it, involves effortfulcontrol. Effortful control is defined by Rothbart as ‘‘efficiency of
executive attention, including the ability to inhibit a dominantresponse and/or to activate a subdominant response, to plan, and to
detect errors" (personal communication, Rothbart, 1/26/02;Rothbart & Bates, 1998). Effortful control includes the abilities
to voluntarily focus and shift attention (e.g., shift attention froma distressing event and/or focus it on a different task or object to
avoid or limit one’s experience and/or expression of emotion) andinhibitory and activational control (i.e., the abilities to effortfully
stop or initiate behaviors when it is adaptive to do so, even if theindividual does not really feel like doing so). Eisenberg, Fabes,Guthrie, and Reiser (2000) and others (Rothbart & Bates, 1998)
have hypothesized that children who are high in effortful control arelikely to be prosocial, socially competent, and relatively low in
problem behaviors, especially externalizing problem behaviors.Individual differences in effortful control appear to predict develop-
mental outcomes from relatively early in development. For example,observations and parental reports of young children’s inhibitory control
Regulation/Control and Problem Behaviors 1173
have been related to internalized compliance, rule-abiding behavior, and
Vandegeest, 1996). Moreover, effortful attentional control (e.g., usingdistraction when confronted with frustration or temptation) appears to
be linked to delay of gratification (e.g., Peake, Hebl, & Mischel, 2002;Sethi, Mischel, Aber, Shoda, & Rodriguez, 2000). The relation of early
effortful control to positive developmental outcomes appears to bemaintained over time. For example, in longitudinal research, Mischel
and colleagues (Ayduk et al., 2000; Mischel, 2000; Mischel, Shoda, &Peake, 1988; Mischel, Shoda, & Rodriquez, 1989) found that youngchildren’s delay of gratification (especially as reflected in their ability to
spontaneously use attentional strategies) was related to a number ofpositive developmental outcomes one to two decades later (also see
Rothbart, Ahadi, & Hershey, 1994). Similarly, Eisenberg, Fabes, et al.(2000) found that effortful attentional control predicted quality of
children’s social competence over time.Recent research supports the premise that effortful attentional
control is negatively related to children’s externalizing problembehaviors (Gilliom, Shaw, Beck, Schonberg, & Lukon, 2002). Forexample, in a normative sample, high levels of parent- and teacher-
reported attentional control were negatively related to their reports ofchildren’s externalizing problem behaviors, both concurrently and
often longitudinally, even when controlling for consistency inexternalizing problems (and attentional control) over time (Eisenberg,
Guthrie, et al., 2000). Moreover, behavioral indices of regulation (e.g.,persistence on tasks or the ability to inhibit behavior) often relate
negatively to externalizing problem behaviors (Eisenberg et al., 1996;Eisenberg, Guthrie, et al., 2000; Olson, Schilling, & Bates, 1999;
Oosterlaan & Sergeant, 1996). However, in some of this research,effortful inhibitory control measures were not clearly differentiatedfrom measures of reactive control (e.g., Eisenberg, Guthrie, et al.,
2000), and in few studies have investigators assessed the prediction ofexternalizing problems from effortful control while controlling for
initial levels of effortful control.
Overcontrol. In contrast to effortful control, some aspects ofcontrol or lack thereof, appear to be so automatic that they often are
not under voluntary control. Thus, it is important to differentiatereactive over- and undercontrol from effortful control. The
1174 Valiente et al.
construct of overcontrol is similar to Derryberry and Rothbart’s
(1997) construct of passive or reactive control and Block and Block’s(1980) construct of ego control. According to the Blocks, ego
undercontrol involves insufficient modulation of impulses, theinability to delay gratification, immediate and direct expression of
motivations and affects, and vulnerability to environmentaldistractors. In contrast, ego overcontrol refers to the containment
of impulses, delay of gratification, inhibition of actions and affect,and insulation from environmental distractors. Although the Blocks
did not differentiate between effortful and reactive control,Derryberry and Rothbart (1997) suggested that their construct ofego control taps primarily reactive control.
Consistent with Block and Block (1980), we believe that eitherovercontrolled children (e.g., those who exhibit highly inhibited and
rigid behavior that is not fully under voluntary control) orundercontrolled children (e.g., children pulled by impulsive,
reward-driven approach tendencies with insufficient modulation ofimpulses) are at risk for the development of problem behaviors. This
is because they lack the ability to modulate their emotion andemotion-related behavioral responses in accordance with therequirements of social situations.
The Unique Linear Effects of Effortful and Reactive Control
The unique effects of effortful control and reactive control in the
prediction of externalizing problems seldom have been examined,especially as children age. There is evidence that effortful control
increases with age in childhood and early adolescence (e.g., Murphy,Eisenberg, Fabes, Shepard, & Guthrie, 1999; Reed, Pien, &
Rothbart, 1984; Williams, Ponesse, Schachar, Logan, & Tannock,1999). Moreover, Asendorpf (1994) found that behavioral inhibi-
tion—often considered a behavioral manifestation of overcontrol(Derryberry & Rothbart, 1997)—is less evident in middle than inearly childhood for children high in social competence or
intelligence. Because effortful control probably contributes to socialcompetence (see Eisenberg, Fabes, et al., 2000, for a review) and is
used to inhibit or activate behavior, it is quite possible that age-related increases in effortful control can be used by children to
modulate their overt expression of overcontrolled or undercontrolledreactive tendencies. Although overcontrolled individuals may con-
Regulation/Control and Problem Behaviors 1175
tinue to experience reactive inhibition with age, and undercontrolled
children may continue to be pulled impulsively by potentiallyrewarding stimuli and situations, the expression of these tendencies
may be increasingly influenced by age-related increases in the abilitiesto effortfully manage their emotions and behavior. Thus, it is
reasonable to hypothesize that the unique prediction of externalizingproblems by overcontrol versus undercontrol declines with age.
The Moderating Effects of Negative Emotionality
Often examined as a temperament or personality variable, emotion-
ality, especially the frequency and/or intensity of negative emotion-ality, has been viewed as a key construct predicting adjustment (e.g.,Eisenberg, Fabes, et al., 2000). Consistent with this view, a number of
investigators have reported that children high in anger, frustration,and/or hostility are also high in externalizing problem behaviors
Although the direct effects of emotionality on externalizing problembehaviors are of interest, its potential role as a moderator of the relations
of effortful or reactive control to externalizing problems is also ofconceptual import. Eisenberg and Fabes (1992; Eisenberg, Fabes, et al.,2000) presented a heuristic model in which the additive andmultiplicative
effects of negative emotionality and regulation were hypothesized topredict externalizing problem behaviors. According to their model, the
relation between regulation and externalizing problem behaviors isexpected to be stronger for children prone to experience intense negative
emotions than for children who are not. Further, children high innegative emotionality and low in regulation are predicted to be most
susceptible to externalizing problems and their precursors (e.g., defiancein young children). For children who are not prone to frequent and
intense negative emotions, there may be less of a need to effortfullymanage their attention and the behavioral concomitants of emotionbecause they are unlikely to become overaroused.
Eisenberg and Fabes’ (1992; Eisenberg, Fabes, et al., 2000) modelsuggests that the relation of undercontrol versus overcontrol with
externalizing problems also may be moderated by negativeemotionality. Specifically, they suggested that the combination of
low effortful control and low reactive control, especially when highnegative emotionality is related to high levels of reactive externaliz-
1176 Valiente et al.
ing problems. Children prone to negative emotions may be
especially likely to experience fear or other negative emotions thatresult in behavioral inhibition, whereas a proneness to anger may
heighten the tendency to respond in impulsive ways.Existing data generally are consistent with the premise that
individual differences in children’s negative emotionality andregulation interact when predicting externalizing problem behaviors,
even among young children (Stifter, Spinrad, & Braungart-Ricker,1999; also see Belsky, Friedman, & Hsieh, 2001). For example, using
hierarchical regressions, Eisenberg et al. (1996) found that teachers’and parents’ reports of externalizing problem behaviors were moststrongly related to children’s control (a composite of both effortful
control and ego control) if teachers reported that children also werehigh on negative emotionality.
In a 2-year follow-up of the same sample, Eisenberg, Guthrie,et al. (2000), using structural equation modeling, found that the
negative relation between behavioral control and externalizingproblems was not moderated by level of negative emotionality. In
contrast, prediction of externalizing problem behaviors fromattentional control (i.e., the average of parents’ and teachers’ reportsof attention shifting and attention focusing) was positive and
significant only for those prone to negative emotionality. However,in the aforementioned studies, the unique linear and multiplicative
(moderated) contributions of effortful control and reactive over/undercontrol were not tested (e.g., in Eisenberg’s studies, measures of
primarily effortful inhibitory control and over/undercontrol werecombined). To our knowledge, moderation of the relation of
overcontrol versus undercontrol to social functioning has been testedonly once. In that study, the moderating effects of over/undercontrol
(as well as effortful control) were not significant (Cumberland,Eisenberg, & Reiser, in press). However, the sample size in that studywas small so the study did not provide a powerful test of moderation.
THE PRESENT STUDY
This study is a 4-year follow-up of the sample in Eisenberg et al.(1996); 2-year findings were presented in Eisenberg, Guthrie, et al.
(2000). The primary focus in the present article is on the findings in thethird follow-up (henceforth labeled T3), and prediction of T3 outcomes
Regulation/Control and Problem Behaviors 1177
from T1 and T2 measures. Our goals in the present study were
threefold: (a) to test the hypothesis that effortful control andovercontrol uniquely predicted T3 externalizing problem behaviors,
(b) to examine if concurrent effortful control predicted T3 externalizingproblem behaviors when taking into account consistency of problem
behaviors over time (essentially controlling for the prior level ofproblem behaviors and the prior effects of effortful control and
overcontrol on problem behaviors), and (c) to examine whethernegative emotionality moderated the relations of effortful control or
overcontrol to externalizing problem behaviors. Consistent with thefindings reported by Eisenberg, Guthrie, et al. (2000), we expectedeffortful control to predict unique variance in externalizing problem
behaviors. In addition, we expected overcontrol (versus undercontrol),measured at T1, T2, or T3, to correlate with externalizing problems at
T3 but were unsure if overcontrol would predict unique variance inexternalizing problems once the effects of effortful control were taken
into account. We further predicted that the magnitude of the concurrentrelations between these constructs would be similar over the four years
(see Eisenberg, Guthrie et al., 2000), with the possible exception of thestrength of the path between overcontrol and externalizing problems.As noted previously, as children’s effortful control develops, and they
are able to effortfully reduce their emotional arousal and inhibitemotion-related behavior, the association between the outer manifesta-
tions of reactive undercontrol and externalizing problem behaviors maybe diminished. Furthermore, we expected the relation of effortful
control, and perhaps overcontrol, to externalizing problems to bestrongest for children prone to experience negative emotions.
Finally, we examined the role of children’s sex, age at the time ofassessment, and socioeconomic status as potential moderators of the
pattern of relations. However, because these variables generally didnot moderate the focal relations at younger ages, and because weexpected the proposed relations to be fairly robust, we did not
expect such moderation at T3.
METHOD
Participants
The sample was children participating in an ongoing longitudinal studyin a large southwestern urban area. Children and parents who completed
1178 Valiente et al.
a first-time laboratory visit (T1) and a follow-up visit (or responded toquestionnaires by mail) either 2 years (T2) or 4 years later (T3), or both,were included in this study. The sample at T1 was 199 children (Mage5 89.51 months, SD5 13.94), including 97 girls (M age5 88.19,SD5 13.63) and 102 boys (M age5 90.76, SD5 14.19). At T1, 20% ofthe sample was in kindergarten (20 girls, 20 boys), 31% was in first grade(32 girls, 30 boys), 24% was in second grade (25 girls, 23 boys), and 25%was in third grade (20 girls, 29 boys). Participants at T1 were 76% Euro-American, 4% African American, 12% Hispanic, 0.1% Asian, 2%American Indian, and 5% other. The mean family income at T1 was$46,500 (SD5 24,000). Mean years of education were 14.60 (SD5 2.00)for mothers and 14.99 (SD5 2.55) for fathers. At T2, the sample included86 girls and 81 boys, and 79% was Euro-American (with 11% HispanicAmerican, 1% Asian American, 3% American Indian, 1% AfricanAmerican, and 5% other groups). At T3, the sample included 84 girls and85 boys, and 80% was Euro-American (with 11% Hispanic American,1% Asian American, 2% American Indian, 1% African American, and5% other groups).
The participants who completed the visit at T1 were compared withthe participants who discontinued participation at the T3 assessment onthe demographic variables as well as the T1 variables used in this study.Families who discontinued participation from T1 to T3 (n5 30) weredisproportionately from minority families, w2(5)5 28.34, po.001, hadlower incomes, t(186)5 3.15, p5 .002, and lower levels of both maternaland paternal education, ts(195 and 180)5 3.01 and 2.48, ps5 .003 and.014. The participants who discontinued participation also received lowerteacher ratings on effortful control and overcontrol, ts(196 and197)5 2.00 and 2.34, ps5 .047 and .02. (see Eisenberg, Guthrie, et al.,2000, for a comparison of T1 and T2 samples).
Procedure
Generally children and their parent (usually the mother) came to auniversity laboratory at T1, T2, and T3. While the parent completedquestionnaire measures, children were administered a puzzle taskdesigned to assess their regulation. At each assessment, children’s currentteachers reported on children’s effortful control, overcontrol, negativeemotionality, and externalizing problem behaviors. Questionnaires weresent to parents when they could not come to the laboratory.
Measures
Children’s regulationEffortful control. At each assessment, parents and teachers completedthe attention shifting and attention focusing subscales for the Child
Regulation/Control and Problem Behaviors 1179
Behavior Questionnaire (Goldsmith & Rothbart, 1991). Both attentionshifting (e.g., ‘‘Has an easy time leaving play to come inside for schoolwork’’) and attention focusing (e.g., ‘‘When picking up toys or othertasks, usually keeps at the task until it’s done’’) were rated on 7-point(15 extremely untrue to 75 extremely true) scales, and final scores werethe average of 11 items each. Alphas for parent-reported attentionshifting at T1, T2, and T3 were .66, .77, and .82 and for attention focusingalphas at T1, T2, and T3 were .80, .82, and .86. Alphas at T1, T2, and T3were .84, .87, and .88 for teacher-reported attention shifting and .89, .89,and .86 for teacher-reported attention focusing. Parents’ reports ofattention shifting and attention focusing were positively correlated at T1,T2, and T3, rs(196,156,160)5 .25, .33, and .41, ps o.001, as wereteachers’ reports, rs(196,151,143)5 .47, .53, and .66, pso.001. Therefore,scales were averaged within reporter at each assessment to form separatecomposite scores for parents’ and teachers’ reports of effortful control.
Observed effortful control. Children’s persistence also was observedwhile they completed a puzzle. The experimenter told the child to assemblethe puzzle without looking at it and that if the puzzle were finished withinthe allotted time (5 minutes), the child would receive an attractive prize.The box was constructed with a clear Plexiglas back (so that children’shand movements could be observed) and a cloth covering the front thathad sleeves through which the children placed their arms. Although thecloth at the front of the box blocked the child’s view of the puzzle, it wasnot attached at the bottom and could be easily lifted so the child couldcheat by looking at the puzzle. The experimenter left the room during thepuzzle task until the child called him or her back by ringing a bell or untilthe timer went off. The amount of time children persisted on the puzzletask rather than being off task or cheating was coded by an observer, aswell as by a reliability coder (interrater reliabilities, computed for 90, 76,and 77 children at T1, T2, and T3, respectively, were .99 at T1, .93 at T2,and .98 at T3). The time a child spent persisting was divided by the totaltime he or she spent on the puzzle task. Because the puzzle task involved areward, it is possible that this measure taped reactive impulsivity (e.g.,reward dominance) for some children in addition to effortful control.
Overcontrol. At T1, parents and teachers sorted cards from the Blockand Block Q-sort (Block & Block, 1969, 1980) into 9 unequal piles (with anormal distribution) on a 9-point scale (15most undescriptive to 95mostdescriptive). To construct the overcontrol scale, we used the list of itemsidentified by Block and Block (1969; personal communication, 1992) asbeing very representative of overcontrol or undercontrol (i.e., were rated byexperts with a value of 7.7 or higher, or 2.3 or lower, on a 9-point scale).
1180 Valiente et al.
Then, items that three experts on the topic deemed by consensus asreflecting specific social skills, moral behaviors, problem behaviors, or overtemotional responding were eliminated. This procedure resulted in a 19-itemovercontrol scale (e.g., ‘‘Is inhibited and constricted’’; see Eisenberg et al.,1996, for item numbers). To reduce the length of time it took to completethis measure at T2 and T3, a questionnaire containing the 19 items wasadministered (with the exception that 18 items were used for mothers’reports at T3; one item was dropped because it had a negative item-scalecorrelation). Some items were simplified by adding phrases from Caspi et al.(1992). Cronbach’s alphas at T1, T2, and T3 were .75, .80, and .82 forparents and .85, .84, and .89 for teachers, respectively. One item (‘‘isattentive and able to concentrate’’) could refer to both attentionalregulation and the ability to sit still and work. This scale likely reflectsprimarily reactive overcontrol versus undercontrol (Derryberry & Roth-bart, 1997), although it also may tap effortful control to some degree(Eisenberg & Morris, 2002).
Negative emotionality. Children’s negative emotionality was assessedwith two measures. At each assessment, parents and teachers rated(15 never to 75 always) children’s negative emotional intensity (e.g.,‘‘When my child experiences anxiety, it normally is very strong’’) on anadapted version of Larsen and Diener’s (1987) Affect Intensity Scale(Eisenberg et al., 1997). Alphas at T1, T2, and T3 were .72, .75, and .86for parents and .85, .86, .87 for teachers, respectively. Parents andteachers also rated children on 11 Q-sort items (e.g., ‘‘Is fearful andanxious’’; 15most undescriptive to 95most descriptive) pertaining tochildren’s negative emotions (Block & Block’s, 1969, 1980); Alphas at T1,T2, and T3 were .77, .84, and .83 for parents and .86, .86, and .89 forteachers, respectively. Parents’ reports of negative emotional intensityand the Q-sort items were significantly related, T1, T2, and T3, rs(197,164, and 164)5 .52, .65, and .66 pso.001, respectively, as were teachers’reports, rs(197, 150, and 146)5 .70, .80, and .75 pso.001, respectively.Therefore, the scales were standardized and then averaged withinreporter to form separate measures of parents’ and teachers’ reports ofnegative emotionality at T1, T2, and T3.
Problem behaviors. At T1, T2, and T3, parents and teachers ratedchildren’s externalizing problem behaviors using the Lochman andConduct Problems Prevention Research Group’s (1995) 24-item (15 neverto 45 often) screening measure. All items (e.g., argues, lies, physicallyharms other children) were included except for ‘‘sets fires,’’ which webelieved to be infrequent and possibly offensive to parents. All alphas ateach time period, and for both parents and teachers, were above .90.
Regulation/Control and Problem Behaviors 1181
RESULTS
Table 1 contains the means and standard deviations. Correlations werecomputed between the T3 variables of interest in the study and both ageof the participants and SES. Two significant correlations were found forage. Teachers rated older children as higher in effortful control,r(150)o.16, po.047, and older children were higher on persistenceduring the puzzle task, r(151)o.25, po.002. Participants’ SES wascomputed by calculating the mean of the standardized score of parents’education (i.e., the average of maternal and paternal education) and thestandardized score of family income. Higher SES was associated withhigher ratings of effortful control and overcontrol, rs(149 and 152)o.29and .16, p5 .001 and .043, and lower externalizing problem behaviors,r(152)o� .24, po.003.
To examine possible sex differences in the measures, two MANOVAswere computed, one for the T3 parent-report measures and one for the T3teacher-report measures. The MANOVA for parent-reported measureswas not significant, but there was a significant multivariate effect(Hotellings’s T) for the teacher-report measures, F(3,146)5 7.97, po.001.According to univariate analyses, teachers rated girls higher on effortfulcontrol and overcontrol, Fs(1,148)5 21.04 and 10.71, pso.001 and .001,and lower on behavior problems F(1,148)5 20.08, po.001. Finally,according to a one-way ANOVA, girls and boys did not differ inobserved persistence on the puzzle.
Relations Among Constructs
Correlations were computed to examine the relations among effortfulcontrol, overcontrol, and externalizing problem behaviors. Table 2presents the correlations among the measures at T3. Only one correlation,between observed effortful control and teacher-reported overcontrol,was nonsignificant. The various indexes of effortful control andovercontrol were positively interrelated. Moreover, higher externaliz-ing scores, as reported by both parents and teachers, were associatedwith lower effortful control and lower overcontrol (or higher under-control).
Correlations were also computed among the variables at T1, T2, andT3 (see Table 3; correlations among T1 and T2 variables are presented inEisenberg, Guthrie et al., 2000). All variables were consistent (i.e., weresignificantly related with the same variable at a later time) over two andfour years. Moreover, relations among the different variables wereconsistent over time, with positive relations among various indexes ofeffortful control and between measures of effortful control and
1182 Valiente et al.
overcontrol, and negative relations between measures of effortful controlor overcontrol and externalizing problems over 2 or 4 years.
The intercorrelations also were computed after transforming thosevariables that were skewed. The pattern of relations was highly similar;no correlations changed in their value more than .05 (such changes weregenerally for highly significant correlations) and none of the levels ofsignificance for the correlations differed across transformed andnontransformed variables.
Table 1Means and Standard Deviations of the Major Variables
Time 1 Time 2 Time 3
M SD M SD M SD
Variables
1. Parent: Effortful Control 4.43 0.67 4.58 0.73 4.63 0.79
2. Teacher: Effortful Control 4.58 0.89 4.72 0.98 5.07 1.06
Note. Ns range from 150 to 168. 1po.10; npo.05; nnpo.01; nnnpo.001.
Regulation/Control and Problem Behaviors 1183
Table
3CorrelationsFrom
T1andT2to
T3
T3Variables
12
34
56
7
T1Variables
1.Parent:EffortfulControl
.61nnn
.29nnn
.06
.37nnn
.25nnn
�.34nnn
�.32nnn
2.Teacher:EffortfulControl
.30nnn
.39nnn
.07
.29nnn
.30nnn
�.13+
�.40nnn
3.Puzzle
Box
.12
.32nnn
.44nnn
.27nnn
.29nnn
�.31nnn
�.37nnn
4.Parent:Overcontrol
.53nnn
.36nnn
.14+
.68nnn
.49nnn
�.44nnn
�.42nnn
5.Teacher:Overcontrol
.36nnn
.44nnn
.15+
.40nnn
.50nnn
�.28nnn
�.54nnn
6.Parent:ExternalizingBehaviors
�.37nnn
�.19n
�.27nnn
�.44nnn
�.21nn
.72nnn
.34nnn
7.Teacher:ExternalizingBehaviors
�.32nnn
�.39nnn
�.15+
�.32nnn
�.48nnn
.32nnn
.59nnn
T2Variables
8.Parent:EffortfulControl
.73nnn
.36nnn
.24nn
.52nnn
.18n
�.54nnn
�.31nnn
9.Teacher:EffortfulControl
.46nnn
.66nnn
.23nn
.42nnn
.35nnn
�.26nnn
�.54nnn
10.Puzzle
Box
.22nn
.32nnn
.58nnn
.26nn
.25nn
�.29nnn
�.28nn
11.Parent:Overcontrol
.57nnn
.46nnn
.33nnn
.79nnn
.49nnn
�.53nnn
�.43nnn
12.Teacher:Overcontrol
.38nnn
.56nnn
.14+
.56nnn
.62nnn
�.36nnn
�.64nnn
13.Parent:ExternalizingBehaviors
�.42nnn
�.32nnn
�.36nnn
�.47nnn
�.24nn
.78nnn
.37nnn
14.Teacher:ExternalizingBehaviors
�.39nnn
�.56nnn
�.20n
�.41nnn
�.57nnn
.35nnn
.62nnn
Note.Nsrangefrom
150to
199.+po.10;npo.05;nnpo.01;nnnpo.001.
Concurrent T3 Structural Equation Model and
Tests of Moderation
A primary aim of the present study was to examine the uniquelongitudinal relations of children’s effortful control and overcontrol tochildren’s externalizing problem behaviors when taking into accountconsistency of the various variables over time. Such an analysis assessesthe likelihood that change in the unique prediction from effortful controlor overcontrol over the years is related to change in externalizing problembehaviors over the same period (in a sense, controlling for levels of thevariables four years prior). To examine these relations, we used Mplus(Muthen & Muthen, 1998), a structural equation modeling program thatcan be used with missing data. First, a model was computed for the T3data; then the T1 to T3 longitudinal model was tested (the T1 to T2 modelis presented in Eisenberg, Guthrie, et al., 2000). The sample size was notsufficient to justify computing a model with data from T1, T2, and T3.
T3 Model. When using the same reporters for latent variables, therelations among the latent variables may be positively biased. To reducethe potential bias, the within-reporter covariances among the error termsof indicators were estimated (i.e., the error term for parents’ report onone variable was correlated with error terms for their reports on othermeasures; covariances between errors for teachers’ and parents’ measures(or the observed measure of effortful control) were not estimated; Kenny& Kashy, 1992). The measurement-error covariances are not displayed inthe figure to simplify presentation.
The T3 model fit the data fairly well, w2(5; N5 169)5 8.00, ns; CFI(Comparative Fit Index)5 .994; RMSEA5 .06 (confidence intervals (CI)for the RMSEA5 .00 to .133). The nonstandardized and standardizedparameter estimates are presented in Figure 1. As predicted, effortfulcontrol was negatively related to externalizing problem behaviors;however, overcontrol was not significantly related to externalizing problembehaviors once the effects of effortful control were taken into account.
Tests of moderation by negative emotionality. Next, we tested thehypothesis that negative emotionality moderated the relations of effortfulcontrol and overcontrol to externalizing problem behaviors. Weconstructed low and high negative emotionality groups (using theaverage of parents’ and teachers’ reports), based on a median split, andthen we used Box’s M to test if the covariance matrix differed based onnegative emotionality. Although results from Box’s M suggested thatmoderation was present, F(28,57729)5 1.56, p5 .031, we did not obtainevidence of moderation using a multiple group structural model.
Regulation/Control and Problem Behaviors 1185
Given the relatively small sample size, it is not surprising thatmoderation was not found in the structural model. Moreover, in SEMmultiple-group moderational analyses, the moderator is split into twogroups at the median, which provides less power than using a continuousmeasure of the moderator. Thus, because Box’s M was significant, we usedregressions to further examine the hypothesis that negative emotionalitymoderates the relations of effortful control and overcontrol to externaliz-ing problem behaviors. To limit the number of regressions, we formedthree composite variables. Because parents’ and teachers’ reports ofeffortful control and the puzzle box were all significantly related (see Table2), we standardized and averaged these three measures to form a singlemeasure of effortful control. In addition, we standardized and thenaveraged parents’ and teachers’ reports of overcontrol to form a singlemeasure of overcontrol, and we standardized and then averaged parents’and teachers’ reports of externalizing problem behaviors to form a singlemeasure of externalizing problem behaviors.
When predicting externalizing problem behaviors, we entered the maineffects on the first step (effortful control or overcontrol and negativeemotionality), and the multiplicative interaction term was entered on thesecond step. As suggested by Aiken and West (1991), predictors werecentered prior to computing the regressions.
As predicted, the main effects of effortful control (and negativeemotionality) and the interaction between effortful control and negativeemotionality were significant, Fs(2,166 and 1,165) for R2 change for steps 1and 25 155.48 and 4.74 (R2 change5 .65 and .01, respectively), pso.001and .04. Betas for effortful control, negative emotionality, and the
Figure1The T3 model with unstandardized estimates and standardizedestimates (the latter in parentheses). Solid lines represent significantpaths and the dotted line represents a hypothesized but nonsignifi-
cant path. nnpo.01.
1186 Valiente et al.
interaction term were � .35, .57, and � .14, pso.001, .001, and .04. Usingprocedures outlined by Aiken and West (1991), we plotted the interactionpresented in Figure 2. As can be seen in Figure 2, the negative relationbetween effortful control and externalizing problem behaviors wasstrongest at high levels of negative emotionality (slopes for low, medium,and high negative emotionality were � .24, � .35, and � .46,ts(165)5 � 2.78, � 5.64, and –6.34, pso.001, respectively). There wasalso a marginally significant interaction between overcontrol and negativeemotionality, Fs(2,166 and 1,165) for the change in R2 at steps 1 and 2were 170.26 and 2.65, pso.001 and .105, R2 change5 .67 and .01, betasfor the main effects of overcontrol, negative emotionality, and theinteraction were � .35, .58, and � .09, ps o.001, .001, and .105. Again,the inverse relation between overcontrol and externalizing problembehaviors was most pronounced at high levels of negative emotionality(slopes for low, medium, and high negative emotionality were � .28,� .35, and � .42, ts(165)5 � 3.94, � 6.78, and � 6.69, ps o.001).1
Figure2The interaction of effortful control and negative emotionality on
externalizing problem behaviors. npo.01.
1. We also computed interactions without averaging across reporter or the puzzle
box. Most of the significant findings for effortful control were for the puzzle box.
The plots of these interactions (when significant or marginally significant) were
very similar to the interaction presented in Figure 2.
Regulation/Control and Problem Behaviors 1187
Tests of moderation by sex or age of the child and SES of the family. UsingBox’s M, we tested whether the covariance matrices differed for boys andgirls, older and younger children, or for low and high SES families(Winer, 1971, based on a median split). Results from Box’s M indicatedthat children’s gender and family SES did not moderate the T1 or T3models. Box’s M suggested that children’s age may moderate the T1 andT3 models, Fs(28,122131 and 28,57275)5 43.76 and 44.24, ps o.04 and.05. However, using structural equation modeling (or regressions inwhich we used composite measures of the various constructs), we couldnot obtain evidence that children’s age moderated the pattern of findings.
Longitudinal SEM model. Using a panel model, we estimated thelongitudinal relations among the variables. In the longitudinal model, thesame within-time relations were estimated, as well as the autoregressivepaths (i.e., paths across time for a given variable, see Table 3 forcorrelations between T1 and T3 variables). In addition, covariancesbetween the error terms for observed variables within a given reporterwere included (e.g., the error term for the observed construct of parent-reported effortful control at T1 was correlated with the error term forparent-reported effortful control at T3); error terms for the puzzle taskalso were correlated across T1 and T3. We did not covary teachers’reports across time because different teachers were used at T1 and T3.
Because we were interested in comparing relations across time, we setthe loadings on T1 latent variables equal to the equivalent loadings on theT3 latent variables (this was not done for the puzzle box because only twoof the three indicators of a latent construct need to be equal to eachother). In addition, we tested whether the magnitudes of the paths fromT1 effortful control and overcontrol to T1 externalizing problembehaviors were equal to the same paths at T3. Therefore, in model 1,we constrained these paths to be equal across time (e.g., the path fromeffortful control to externalizing problem behaviors at T1 was set to beequal to the path from effortful control to externalizing problembehaviors at T3). This model did not fit the data well, w2(51;N5 199)5 105.058, po.001 and a CFI5 .956 and an RMSEA5 .073(CI for the RMSEA5 .053 to .093; Akaike Informational Criterionor AIC5 3900.003, a lower AIC when comparing models indicates abetter fit).
In model 2, we allowed the paths from overcontrol to externalizingproblem behaviors (at both time periods) to be freely estimated. Thisresulted in a significant improvement in model fit, w2(50)5 84.953,po.01 and a CFI5 .972 and an RMSEA5 .059 (CI for theRMSEA5 .037 to .080; AIC5 3881.898), w2D(1)5 20.105, po .001.However, the measurement error variance of externalizing problem
1188 Valiente et al.
behaviors (the latent variable) at T3 was estimated to be negative.Therefore, in model 3, we set this parameter equal to zero (its lowestpossible value), per standard convention (thus, the R2 for this variable isnot meaningful). For this model, w2(51)5 85.198, po.0019, CFI5 .972,RMSEA5 .058 (CI for the RMSEA5 .035 to .079; AIC5 3880.144).Compared to model 2, setting this parameter to zero did not change thefit of the model, w2D(1)5 .245, ns. In model 4, we allowed the paths fromboth T1 overcontrol and effortful control to T1 externalizing problembehaviors to be different than the analogous paths from T3 overcontroland effortful control to T3 externalizing problem behaviors: w2(50)585.08, po.002, CFI5 .972, RMSEA5 .059 (CI for the RMSEA5 .037to .081; AIC5 3882.025. On one degree of freedom, the improvement inw2 between model 4 and model 3 was not significant, w2D(1)5 .12, ns.Because (a) a model is more parsimonious if more paths are restricted tobe equal across time, (b) model 3 was an improvement over the fullyconstrained model, and (c) model 3 had a lower AIC than model 4, wedecided that model 3 (with paths from effortful control to externalizingset to be equal) was the best model (see Figure 3). As expected, low levelsof T1 externalizing behaviors were significantly predicted by botheffortful control and overcontrol. When controlling for T1 externalizingbehaviors, the path from T3 effortful control to T3 externalizingbehaviors was negative and significant and the path from overcontrolto externalizing problem behaviors was nonsignificant. In addition, thethree autogressive (across time) paths were all significant.2
DISCUSSION
The goal of the present study was to examine the relation of children’sdispositional emotionality, effortful control, and over- versus under-control to their externalizing problem behavior. The associations amongregulation/control, negative emotional intensity, and externalizing
2. Based on the consensus of the second author and another expert, five items
that most clearly might tap attentional or inhibitory effortful control (especially
the voluntary element of it; that is, ‘‘Is attentive and able to concentrate on
things,’’ ‘‘Is planful; thinks ahead,’’ ‘‘Is reflective; thinks and deliberates before
speaking or acting,’’ ‘‘Is restless and fidgety, has a hard time sitting still,’’ and ‘‘Is
persistent or determined in what he/she does, does not give up easily’’) were
dropped from both the parent- and teacher-reports of overcontrol. Alphas for the
adapted 14-item teacher-report and parent-report scales were .82 and .70 for T1,
.83 and .72 for T2, and .86 and .73 for T3. When we recomputed the SEM models
using these scales, the findings were similar. However, the marginal interaction
between overcontrol and negative emotionality became nonsignificant.
Regulation/Control and Problem Behaviors 1189
behavior problems were examined in a longitudinal sample of childrenseen over a 6-year period.
One of the questions addressed was the role that effortful control andovercontrol play in predicting externalizing behavior problems. Becausewe view effortful control and overcontrol as distinct constructs related toemotional-related control, we hypothesized that both could predict someunique variance in externalizing behavior problems, especially at youngerages. At all assessments, both effortful control and over- versusundercontrol generally were negatively correlated with low levels ofexternalizing problem behaviors. Moreover, at T1, both effortful controland over/undercontrol provided some unique prediction of externalizingproblems (i.e., both were uniquely, negatively related to externalizingproblems). However, at T3, effortful control was uniquely, negativelyassociated with externalizing problem behaviors, even when controllingfor the effects of overcontrol, whereas overcontrol did not relate toexternalizing problems once the effects of effortful control were takeninto account in the structural equation model.
Figure3The longitudinal model with unstandardized estimates and stan-dardized estimates (in parentheses). Solid lines represent significantpaths and the dotted line represents a hypothesized but nonsignifi-
cant path. nnpo.01.
1190 Valiente et al.
Thus, at all three assessments, the findings are consistent with the viewthat children who were relatively high in effortful control were able tomodulate their behavior and, consequently, engaged in relatively lowlevels of externalizing behaviors. Of course, the data were correlational sowe cannot ascertain causal relations. In addition, based on thecorrelations, and consistent with expectations, children high on over-control displayed fewer externalizing problems than their peers, probablybecause of the association between impulsive (i.e., undercontrolled)tendencies and externalizing problems (e.g., Eisenberg et al., 2001).
In addition to exploring the relation between concurrent effortfulcontrol or overcontrol and externalizing problems, these relations wereexamined across time. We hypothesized that T3 effortful control and T3overcontrol (especially the former) would predict T3 externalizingproblem behavior, even when controlling for T1 levels of externalizingproblem behaviors. Findings in the longitudinal model, as well as for theacross-time correlations, generally were consistent with expectations.Individual differences in effortful control did correlate with externalizingproblems 2 or 4 years later. Moreover, in the SEM model, effortfulcontrol at T3 negative predicted T3 problem behaviors when controllingfor T1 problem behaviors. In contrast, T3 overcontrol did not predict T3externalizing problems after controlling for consistencies in problembehaviors and the variance accounted for by T3 effortful control.
The difference in the pattern of unique relations of externalizingproblems with effortful control and overcontrol over time suggests that aschildren are able to use effortful control more efficiently, overcontrol is lessof a factor in the display of externalizing problems. As children gain morecontrol over voluntary regulatory processes, they may develop the meansto modulate the expression (albeit perhaps less so the experience) of moreinvoluntary, reactive types of responding (e.g., impulsive behaviors that arelikely to contribute to externalizing problems). Stated differently, withdevelopment, children’s overt behavior likely is based less on their reactive(more automatic) behavioral tendencies because they have learned tomodulate their expression with attentional strategies or effortful inhibition(or activation) of behavior. Due to young children’s limited effortfulcontrol (Kochanska et al., 1997; Murphy et al., 1999; Reed et al., 1984),they often may lack the ability to flexibly and voluntarily modulateimpulsive tendencies in an adaptive manner.
The fact that overcontrol and effortful control loaded on differentlatent constructs and both constructs uniquely predicted externalizingproblems provides support for differentiating between the two constructs.Nonetheless, measures of the two constructs were substantially corre-lated. This is not surprising given that the measure of overcontrolcontained some items that could tap effortful control. Moreover, the
Regulation/Control and Problem Behaviors 1191
measure of persistence may have partly tapped reactive impulsivity forchildren who were pulled by the promise of a reward and, as aconsequence, cheated rather than persisted. Important tasks for thefuture are to develop purer measures of reactive over- and undercontroland to assess the degree to which various behavioral measures ofregulation (e.g., delay of gratification tasks) tap effortful control versusreactive over/undercontrol.
The results from the present study support Eisenberg, Fabes,et al.’s (2000) heuristic model in which negative emotionality is apotential moderator of the relation between externalizing behaviorproblems and both effortful control and overcontrol. In the regressionanalyses, the negative relation between effortful control and children’sexternalizing problems was strongest for those children prone to negativeemotions. Moreover, there was a similar, albeit marginally significant,pattern of moderation when overcontrol was used as a predictor ofbehavior problems. Effortful control (or over/undercontrol) may be lesspredictive of externalizing problems for children who are not prone tonegative emotions because they are less likely to engage (or motivated toengage) in emotion-based externalizing behaviors. We were not ableto find moderation in the SEM models, but this was most likely dueto limitations in power, based on the number of parameters, and thefact that negative emotionality had to be dichotomized (rather thantreated as a continuous variable) in the SEM moderational analyses.
In both the correlational analyses and the SEMs, there was evidence ofconsiderable consistency in effortful control, overcontrol, and externaliz-ing problems. These findings are consistent with prior data on aggression(Coie & Dodge, 1998), as well as with the limited research on consistencyin effortful control and reactive control (e.g., behavioral inhibition) overtime (Kagan, 1998; Kochanska et al., 1997; Murphy et al., 1999).Nonetheless, the findings suggest that changes in effortful control dooccur and are linked to changes in the level of externalizing problems.It is likely that experiential factors such as socialization in the home andby peers contribute to the intra-individual changes that occur over time inthese aspects of functioning.
In summary, the results from the current study support the predictedrelation between externalizing problems and both low effortful controland overcontrol. Of particular interest, the unique relation betweenreactive overcontrol and behavior problems was stronger at the earlierassessment than when the children were older. In addition, there wasevidence that negative emotionality moderated the association betweenexternalizing behavior problems and both effortful control and over-control. Because of the longitudinal nature of the study and the fact thatinitial levels of variables were controlled when assessing later relations,
1192 Valiente et al.
we can have more confidence in the causal nature of the relations than ifthe relations were found solely in correlational analyses. However, evenin a prospective design such as that used in this study, the direction ofrelations cannot be proven. In addition, although there was somediversity in the sample, the majority of the participants were Euro-American and from middle-class homes. Even though we did not findmoderation by SES, it is unclear if the findings would generalize to asample comprised of children from lower SES homes or other cultures.
REFERENCES
Aiken, L., & West, S. G. (1991). Multiple regression: Testing and interpreting
interactions. Newbury Park, CA: Sage.
Asendorpf, J. B. (1994). The malleability of behavioral inhibition: A study of
