Temperamental, parental and contextual contributors to early ...
Transcript of Temperamental, parental and contextual contributors to early ...
Temperamental, Parental, and ContextualContributors to Early-emergingInternalizing Problems: A New IntegrativeAnalysis Approachsode_629 229..253
Rosemary S. L. Mills, University of Manitoba, Paul D. Hastings andJonathan Helm, University of California Davis and Lisa A. Serbin,Jamshid Etezadi, Dale M. Stack, Alex E. Schwartzman andHai Hong Li, Concordia University
Abstract
This study evaluated a comprehensive model of factors associated with internalizingproblems (IP) in early childhood, hypothesizing direct, mediated, and moderatedpathways linking child temperamental inhibition, maternal overcontrol and rejection,and contextual stressors to IP. In a novel approach, three samples were integrated toform a large sample (N = 500) of Canadian children (2–6 years; M = 3.95 years;SD = .80). Items tapping into the same constructs across samples were used to createparallel measures of inhibited temperament, maternal positive, critical, and punitiveparenting, maternal negative emotionality, family socioeconomic and structural stres-sors, and child’s IP. Multiple-groups structural equation modeling indicated that asso-ciations were invariant across samples and did not differ for boys and girls. Childinhibition, less positive and more critical parenting, maternal negative emotionality,and family socioeconomic disadvantage were found to have direct associations with IP.In addition, maternal negative emotionality was associated with IP through morecritical parenting, and both maternal negative emotionality and socioeconomic stresswere associated with IP through less positive parenting. Results highlight the multiple
This research was supported by Canadian Institutes of Health Research grant MOP-74642 awardedto the team, and by grants from Health Canada (Child and Youth Division), awarded to Lisa Serbinand colleagues from the Concordia Longitudinal Risk Project, from Social Sciences and Humani-ties Research Council of Canada and the Fonds de la Recherche en Santé du Québec grantsawarded to Paul Hastings for the Daycare and Preschool Adjustment Study, and from CanadianInstitutes of Health Research grant MOP-57670 to Rosemary Mills for the Shame in ChildhoodStudy. We thank the children and parents who made this research possible and gratefully acknowl-edge the assistance of Farriola Ladha, Claude Senneville, Nadine Girouard, and Bobbi Walling. Wealso thank Keith Widaman for statistical consultation.Correspondence should be addressed to Rosemary S. L. Mills, Department of Family Social Sci-ences, University of Manitoba, 35 Chancellor’s Circle, Winnipeg, Manitoba R3T 2N2, Canada.Email: [email protected]
doi: 10.1111/j.1467-9507.2011.00629.x
© Blackwell Publishing Ltd. 2011. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street,Malden, MA 02148, USA.
independent and cumulative risk factors for early IP and demonstrate the power ofintegrating data across developmental studies.
Keywords: internalizing; temperament; parenting; early childhood
Introduction
Internalizing problems (IP) are a common type of maladjustment in childhood, affecting14–18 percent of young people (Zahn-Waxler, Klimes-Dougan, & Slattery, 2000). Earlychildhood is a common period of onset, with anxiety problems in particular becomingevident during the transition to childcare or school. Early-emerging IP are moderatelystable from childhood onward (Majcher & Pollack, 1996), and are likely to be detri-mental to development (Hammen & Rudolph, 2003), underscoring the importance ofearly identification and improved understanding of how these problems develop toadvance prevention and treatment (Hirshfeld-Becker & Biederman, 2002). Numerousfactors in children and their environment contribute to these problems, includinginhibited temperament, parental overcontrol and negativity, and environmental stressors(e.g., Rapee, 2001). How these factors interact to increase the risk of IP and the processesthrough which these problems develop are not well understood, however, due in part tothe lack of research in which multiple factors have been examined simultaneously inearly childhood. Such research requires studies in which children are assessed compre-hensively on factors both internal and external to the child. The present study addressedthis need using a novel cross-study method in which three independent samples wereintegrated at early childhood (e.g., Curran & Hussong, 2009).
Bioecological models suggest that children’s development occurs in nested multi-level contexts, from biological to cultural, with risk factors occurring across levels(e.g., Bronfenbrenner & Ceci, 1994). IP most likely originate from the interactiveeffects of inhibited temperament reflecting high physiological reactivity attributableto predisposing genetic vulnerabilities, experiences of parental overcontrol and nega-tivity, and exposure to stressful familial, economic, and social disadvantages(Zahn-Waxler et al., 2000). Children are increasingly likely to develop IP when moreof these multilevel risk factors are present and less likely to develop them if risk factorsare offset by the presence of protective factors (Ashford, Smit, van Lier, Cuijpers, &Koot, 2008). In the present study, we tested a model postulating pathways to IP fromchild temperamental inhibition; from parenting practices that are likely to increasechildren’s anxiety, especially if they are highly inhibited and hence more susceptible toanxiety; and from parental negative emotionality and family stressors that are likely tointerfere with effective parenting.
Twin and family studies have revealed shared genetic risk for temperamental inhi-bition, anxiety, and depression (e.g., Bolton et al., 2006; Boomsma, van Beijsterveldt,& Hudziak, 2005). Evidence linking physiological reactivity to anxiety problemssupports the idea that inhibition may contribute to the development of IP by affectingthe ability to cope with anxiety. Some children are physiologically prone to states ofanxiety due to over-activation of brain systems involved in stress responses. At thebehavioral level, these brain systems activate regulatory processes designed to lowerarousal, evidenced by distress in the face of novelty and avoidance of unfamiliarsituations (e.g., Kagan & Snidman, 1999). Distress decreases after withdrawal, rein-forcing avoidant coping. However, withdrawal also decreases opportunities to learn orto practice more effective coping, increasing the likelihood of future anxious reactions
230 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
through a positive feedback loop in which anxious avoidance maintains or increasesarousal and the sense of low control. Thus, more inhibited children may show more IPin early childhood.
Factors external to the child may contribute also to children’s IP. Parents are likelyto contribute by responding to the child in ways that reinforce the child’s distress. Bothinappropriate behavioral control (e.g., overdirective and punitive) and psychologicalcontrol (e.g., rejection and criticism) might contribute to the development of IP byundermining the child’s sense of self-agency and self-efficacy and increasing anxietyand distress (e.g., Barber, 1996; Rapee, 2001). Meta-analyses of research on relationsbetween parenting and childhood IP (McLeod, Weisz, & Wood, 2007; McLeod, Wood,& Weisz, 2007), indicated that child depression was associated with parental rejectionand to a lesser extent with parental overcontrol (overinvolvement) whereas anxiety wasassociated with parental overcontrol and to a lesser extent with parental rejection. Ahandful of prospective longitudinal studies suggest the same conclusion (e.g., Bayer,Sanson, & Hemphill, 2006; Letcher, Smart, Sanson, & Toumbourou, 2009), althoughyoung children’s inhibition also contributes reciprocally to these negative parentingdimensions (Rubin, Nelson, Hastings, & Asendorpf, 1999). Studies also point tosubstantial inverse relations between positive parenting (warm engagement) and bothIP and negative parenting (Bayer et al., 2006; Coplan, Arbeau, & Armer, 2008).However, it is unclear whether the presence of negative parenting and the absence ofpositive parenting are independently associated with children’s IP or whether they areat opposite ends of a single continuum. Less positive, more critical, and/or morepunitively controlling parents may have children with more IP. Given inhibited chil-dren’s greater susceptibility to parental influences (e.g., Gallagher, 2002), the linksbetween IP and parenting dimensions may be stronger among more inhibited children.
Parental behavior may be the final common pathway through which other riskfactors contribute to children’s IP. Parents’ personal characteristics may be an impor-tant risk factor for IP (Rubin & Mills, 1991), that operates through the mechanism ofparent behavior. Psychologically vulnerable parents are more likely to resort to inef-fective parenting practices during challenging childcare interactions and may directfeelings of frustration or distress toward their children. Emotionally negative parentsare more likely to engage in overcontrolling and critical or rejecting parenting (Coplanet al., 2008; Mills et al., 2007). Further, parenting behavior likely mediates relationsbetween parental emotionality and children’s IP (Elgar, Mills, Waschbusch, Brown-ridge, & McGrath, 2007). Thus, parental negative emotionality may be associated withchildren’s IP, and less positive, more critical, and/or more punitive parenting maymediate the association.
Parental behavior may also be the pathway through which family risk factorscontribute to children’s IP. Family socioeconomic disadvantages (e.g., low income andlow level of education) and challenging structural characteristics (e.g., single parent-hood and large family size) are well-established family risk factors in children’sdevelopment (Conger & Dogan, 2007; Weinraub, Horvath, & Gringlas, 2002). Theireffects on children have been attributed at least partly to the stresses they put uponparents leading to effects on the quality of parenting (Bronfenbrenner & Ceci, 1994).Economic hardship undermines parents’ ability to respond nurturantly to their children(Conger & Conger, 2002). Single-parent and blended families are at higher riskof parent–child conflict, leading to deterioration in the quality of parenting (e.g.,Weinraub et al., 2002). The more children there are in the family, the more parents’financial, psychological, and time resources are likely to be strained (Henderson,
Early-emerging Internalizing Problems 231
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Hetherington, Mekos, & Reiss, 1996), and larger family size has been linked to lessindividual attention to and more differential treatment of children, less positive parent-ing, and more autocratic discipline (e.g., Jenkins, Rasbash, & O’Connor, 2003). Themore young children there are in the family, the greater may be the strain on resources(e.g., Harris, Raley, & Rindfuss, 2002). Thus, contexts of greater family socioecon-omic stress (lower socioeconomic status) and structural stress (single-parent family,less than two biological parents, and more young children in the family) may beassociated with children’s IP, and low positive, more critical, and/or more punitiveparenting may mediate these relations.
In summary, the present study evaluated a comprehensive, ecological modelhypothesizing direct, mediated, and moderated pathways linking child temperamentalinhibition, maternal overcontrol and rejection, and contextual stressors to IP in earlychildhood. We predicted that: (1) more inhibited children would show more IP inearly childhood; (2) less positive, more critical, and/or more punitively controllingparents would have children with more IP; (3) the links between IP and parentingdimensions would be stronger among more inhibited children; (4) parental negativeemotionality would be associated with children’s IP, and less positive, more critical,and/or more punitive parenting would mediate the association; and (5) contexts ofgreater family socioeconomic and structural stress would be associated with children’sIP, and low positive, more critical, and/or more punitive parenting would mediate theserelations.
We evaluated the model using integrative data analysis (e.g., Curran & Hussong,2009), a new technique in which raw data from two or more existing independentsamples are integrated, and models are fit directly to the integrated dataset. Theadvantages of the technique include increased statistical power, greater heterogeneity insample demographic characteristics, the ability to test more complex models than wouldbe possible within any single sample, and the potential to replicate findings across thesamples. There are threats to parsimony associated with pooling datasets with differentsample characteristics, measurements, and response scales. However, the examinationof sample differences allows a richer understanding of the generalizability of the resultsas well as the accuracy of current theoretical notions. The present study is the first to usethis approach to evaluate a comprehensive model of the factors associated with IP inearly childhood. A large and regionally, socioeconomically, and linguistically diversesample of young Canadian children was obtained by integrating three samples: a sampleat familial and sociodemographic risk (sample 1; Campisi, Serbin, Stack, Ledingham, &Schwartzman, 2009), a sample of children varying in risk for anxiety problems (sample2; Daycare and PreschoolAdjustment Study, e.g., Hastings et al., 2008), and a sample offamilies recruited from a representative community group (sample 3; Shame in Child-hood Study, e.g., Mills, Imm, Walling, & Weiler, 2008). To maximize the size of theintegrated sample, assessments of parenting focused on mothers.
Method
Participants
Participants were 523 Canadian children ranging from 2.1 to 6.1 years of age (M = 3.95years; SD = .80) obtained by integrating three samples. Given that there is no evidencefor age differences in risk factors associated with IP in early childhood, the samplecomprised all children up to kindergarten age, including seven kindergarten children in
232 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
sample 1 who turned six before being assessed late in the school year. Twenty-threechildren with missing values on most of the key variables were excluded (7 in sample1 and 16 in sample 3), leaving a final integrated sample of 500. Demographic charac-teristics are shown in Table 1. On average, children in samples 1 and 3 were 7 monthsolder than those in sample 2, t(257) = 4.82, p < .001, and t(372) = 11.23, p < .001,respectively; children in samples 1 and 3 did not differ in age, t(365) = .09, NS.Mothers in sample 1 were significantly younger than those in sample 2, t(255) = 8.59,p < .001, and in sample 3, t(364) = 8.39, p < .001; mothers in samples 2 and 3 did notdiffer in age, t(369) = .03. Proportionately, more mothers had a low level of education(high school or less) in sample 1 (67 percent) than in sample 2 (12 percent) or sample3 (16 percent), c2(8, N = 500) = 213.87, p < .001, and proportionately more motherswere in a low income family (below $30 000/year) in sample 1 (40 percent) than insample 2 (11 percent) or in sample 3 (16 percent), c2(12, N = 487) = 73.01, p < .001.Details on the three samples follow.
Sample 1 (N = 126). The Concordia Longitudinal Risk Project is a study of long-termtrajectories of childhood social withdrawal and aggression. The original sampleincluded over 1770 elementary-school Francophone (French-speaking) children livingin predominantly lower income neighborhoods of Montreal in 1976–1978. As theoriginal participants (now in their thirties and forties) became parents, assessments oftheir children were conducted. Of over 700 representative participants in the mostrecent phase of the ongoing Concordia project, 126 had children of an appropriate agefor the current study and agreed to participate with their child (acceptance rate wasapproximately 78 percent of eligible families). Participating parents did not differsignificantly from either the ongoing longitudinal sample or the original Concordiasample of 1770 in their sociodemographic or behavioral characteristics. The 126participants involved in the present study consisted of those second-generation chil-dren for whom the target variables were assessed in early childhood and their mothers.Families were 95 percent White and 5 percent other ethnicities (Hispanic and African).Assessments were done in home when the children were 4–6 years old, at which time,mothers completed instruments used in the present study.
Sample 2 (N = 133). The Daycare and Preschool Adjustment Study comprised 133families in greater metropolitan Montreal. Targeted advertisements were used torecruit children ranging from ‘low risk’ to ‘at risk’ for early anxiety problems. Bymother report of behavior problems (Achenbach & Rescorla, 2001), 42 children werein the borderline-clinical to clinical range for IP for their age whereas 48 were less than1 SD above, and 43 were at or below the norm for their age. Children ranged from 2.1to 4.9 years old. The sample was 70 percent White and 30 percent other ethnicities(Asian, Caribbean, African, Indian, Hispanic, and North African). Half the familieswere Anglophone (first language is English), one or both parents were Francophone(first language is French) in 37 families, and one or both parents were Allophone (firstlanguage is neither English nor French) in 33 families. Mothers completed instrumentsused for the current study during four assessments over approximately 7 months,involving a telephone interview, a home visit, a mailed questionnaire, and a laboratoryvisit.
Sample 3 (N = 241). The Shame in Childhood Study is a community sample of 253English-speaking from three- to four-year-old children and their parents. Families with
Early-emerging Internalizing Problems 233
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Tab
le1.
Dem
ogra
phic
Cha
ract
eris
tics
ofth
eIn
divi
dual
and
Inte
grat
edSa
mpl
es
Cha
ract
eris
tics
Sam
ple
1(N
=12
6)S
ampl
e2
(N=
133)
Sam
ple
3(N
=24
1)In
tegr
ated
(N=
500)
Perc
ent
orM
Nor
SDPe
rcen
tor
MN
orSD
Perc
ent
orM
Nor
SDPe
rcen
tor
MN
orSD
Chi
ldag
e(y
ears
)4.
111.
233.
50.7
64.
10.2
63.
95.8
0C
hild
gend
erB
oys
48.4
%61
45.9
%61
54.4
%13
150
.6%
253
Gir
ls51
.6%
6554
.1%
7245
.6%
110
49.4
%24
7M
ater
nal
age
30.6
93.
3135
.32
5.11
34.9
85.
6234
.14
5.40
Mat
erna
led
ucat
ion
Bel
owhi
ghsc
hool
27.0
%34
.0%
0.4
%1
7.0%
35C
ompl
eted
high
scho
ol39
.7%
5012
.0%
1615
.4%
3720
.6%
103
Com
mun
ity
coll
ege
24.6
%31
26.3
%35
48.1
%11
636
.4%
182
Und
ergr
adua
tede
gree
7.9%
1042
.1%
5632
.8%
7929
.0%
145
Gra
duat
e/pr
ofes
sion
alde
gree
.8%
119
.5%
263.
3%8
7.0%
35Fa
mily
inco
me
($)
0–10
000
7.1%
9.0
%0
.9%
22.
3%11
1000
1–20
000
16.7
%21
6.2%
88.
6%20
10.1
%49
2000
1–30
000
15.9
%20
4.7%
67.
3%17
8.8%
4330
001–
4000
010
.3%
139.
3%12
10.3
%24
10.1
%49
4000
1–60
000
28.6
%36
17.8
%23
26.3
%61
24.6
%12
060
001–
7499
912
.7%
1614
.0%
1814
.2%
3313
.8%
6775
000
orm
ore
8.7%
1148
.1%
6232
.3%
7530
.4%
148
234 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
young children born between June 1, 1999 and May 31, 2000 were recruited througha letter of invitation sent to 3500 families drawn randomly from all 6358 familiesresiding in Winnipeg, a mid-size city (population 600 000) in the geographic center ofCanada. Details were provided about the study, and families were invited to return anenclosed stamped return postcard if they had a healthy child and wished to participate.Of 257 families volunteering to participate, four withdrew shortly thereafter. Parentspredominantly were White (94 percent), were of European ancestry (74 percent), wereat least second-generation Canadian (78 percent), and reported feeling very Canadian(94 percent rated themselves 8 or higher on a 10-point scale). Mothers completedinstruments used for the current study during an assessment that involved a mailedquestionnaire followed shortly by a laboratory visit.
Between-sample Heterogeneity. The samples varied considerably in method of recruit-ment, with the result that they differed in children’s risk for IP. Children in sample 1were at elevated risk of internalizing and/or externalizing problems by virtue of havingparents at high risk, children in sample 2 ranged from low to at risk of anxiety basedon screening scores, and children in sample 3 were recruited from the community andthus were at low risk of adjustment problems. By increasing the heterogeneity of thepooled sample, these differences increased the potential generalizability of the results,but it was also possible that the higher risk for IP in samples 1 and 2 would contributedisproportionately to the findings. For this reason, in the analyses described below, weexamined sample differences before pooling across samples and tested the measure-ment and structural models with a multiple-groups approach.
Measures
A central challenge in integrative data analysis, given that the measures are likely todiffer across samples, is to develop appropriate measures of the constructs and toestablish measurement invariance across samples. In the present study, in all samples,the same instrument was used to assess IP, and the same information was gatheredabout family socioeconomic and structural characteristics. In two of the three samples,the same instruments were used to assess child temperament and maternal emotion-ality. The three samples used different instruments to measure parenting, all of whichhave been used and validated in previous research. Where the instruments differedacross samples, parallel measures were created by selecting items that, although notidentical across samples, reflected the same construct and had common scales. Wherethe items were on different scales (4- or 5-point scale vs. 7-point scale), values of thesmaller scale were converted to those of the larger scale using graduated constants topreserve distributional properties (1 = 1, 2 = 2.5, 3 = 4, 4 = 5.5, 5 = 7; 0 = 1, 1 = 3,2 = 5, and 3 = 7). The use of rescaled original units was preferable to a transformation(e.g., z-transformation) because the latter can change distributions of data, and thus,relations among variables can be altered (Cudeck, 1989).
For child temperament, parenting, and maternal emotionality, we (five of the coau-thors with cumulatively many decades of experience in measurement and assessmentof child and parent characteristics) began by identifying candidate parallel items in theinstruments used with each sample on the basis of item content. Item and scalestatistics were examined in iterative processes to identify sets of relevant items, withthe highest possible reliability indices, computed for the integrated sample usingdata centered within each constituent sample. Centering served to control for
Early-emerging Internalizing Problems 235
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
between-group variance by ensuring that variances would range around zero. Thismethod of item selection produced subsets of items with the highest reliability acrossthe samples while accounting for between-sample differences. Selected items werethen rigorously tested with their original sample means included [multiple-groupconfirmatory factor analysis (CFA)] to ensure unidimensionality across the samples, asdescribed in the Results section. Thus, the parallel measures were sets of items thatreflected the same construct as indicated by acceptable reliability in the integratedsample. The instruments and procedures used to create parallel measures acrosssamples are described below. The items selected for each measure are shown in Table 2(see Table 3 for alpha coefficients in the three samples and the integrated sample).
Child Temperament. To assess child temperament, parents completed the children’sbehavior questionnaire (CBQ; Rothbart, Ahadi, Hershey, & Fisher, 2001), in samples2 and 3, and the emotionality, activity, and sociability (EAS) temperament survey forchildren (Buss & Plomin, 1984), in sample 1. For the CBQ, parents rate descriptionsof children’s reactions to situations on a 7-point scale from 1 (extremely untrue) to 7(extremely true). For the EAS, parents rate items using a 5-point scale from 1 (notcharacteristic or typical) to 5 (very characteristic or typical). Six items equivalent inmeaning across instruments (6 of 12 shyness items from the CBQ; all 5 shyness and 1of 5 sociability items from the EAS) were used to create a parallel measure of inhibitedtemperament [e.g., prefers to watch than join, at ease with almost anyone (reversed),acts shy around new people]. EAS items were rescaled converting 5-point to 7-pointscales.
Parenting. Mothers’ parenting styles and practices were assessed using the instru-ments and items shown in Table 2. Mothers rated the likelihood or frequency of usinga parenting strategy or indicated how descriptive the item was of their parentingpractices. To increase similarity, some items within an instrument for one sample wereaveraged to form a single composite item that cohered more than any of its constituentitems with the items on the instruments for the other samples. Individual and compos-ite items were used to create three-item parallel measures of positive, critical, andpunitive parenting. Items originally scored on 4- or 5-point scales were rescaled to7-point scales. Parenting scores were computed as the mean of item ratings. Therelatively low alphas (see Table 3) are likely due to the small number of items in eachmeasure necessitated by rigorous item selection. The composites reflect parcels, andthey are very beneficial empirically. Their use allows for the inclusion of informationfrom multiple items without requiring extra parameters within the model (Bagozzi &Edwards, 1998), a greater ratio of common-to-unique variance (Little, Cunningham,Shahar, & Widaman, 2002), and are most likely to capture the true relations amongconstructs of interest (Rushton, Brainerd, & Pressley, 1983).
Maternal Negative Emotionality. Two self-report scales were used to assess mothers’emotionality: the EAS temperament survey for adults (Buss & Plomin, 1984) insamples 1 and 3, and the positive and negative affect scales (PANAS; Watson, Clark,& Tellegen, 1988) in sample 2. The EAS for adults required mothers to rate itemsdescribing emotions and behaviors on scales ranging from 1 (not characteristic ortypical of you) to 5 (very characteristic or typical of you). The PANAS included a listof affect terms (e.g., bashful, nervous, and happy) that parents rated as describing theirown feelings in recent weeks on scales from 1 (not at all) to 5 (extremely). Eight items
236 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Tab
le2.
Inst
rum
ents
and
Item
sU
sed
toC
reat
eP
aral
lel
Mea
sure
sA
cros
sSa
mpl
es
Sam
ple
1S
ampl
e2
Sam
ple
3
Tem
pera
men
tIn
hibi
tion
Pre
fers
play
ing
wit
hot
hers
RP
refe
rsto
wat
chth
anjo
inP
refe
rsto
wat
chth
anjo
inV
ery
soci
able
RA
tea
sew
ith
alm
ost
anyo
neR
At
ease
wit
hal
mos
tan
yone
RM
akes
frie
nds
easi
lyR
Act
sfr
iend
lyan
dou
tgoi
ngR
Act
sfr
iend
lyan
dou
tgoi
ngR
Ver
yfr
iend
lyw
ith
stra
nger
sR
Join
sot
hers
quic
kly
RJo
ins
othe
rsqu
ickl
yR
Take
slo
ngti
me
tow
arm
upN
ervo
usta
lkin
gto
stra
nger
sN
ervo
usta
lkin
gto
new
adul
tsTe
nds
tobe
shy
Act
ssh
yar
ound
new
peop
leA
cts
shy
arou
ndne
wpe
ople
Pare
ntin
gPo
sitiv
eP
SI-
expe
cted
war
mer
feel
ings
RH
ave
war
min
tim
ate
tim
esH
ave
war
min
tim
ate
tim
esP
SI
(Ife
elun
like
dR
/una
ppre
ciat
edR
)I
give
com
fort
and
unde
rsta
ndin
gI
give
com
fort
and
unde
rsta
ndin
gP
DI-
like
lyto
disc
uss
situ
atio
n(R
easo
nw
ith/
RC
E-fi
ndou
tw
hy)
(Giv
ere
ason
s/em
phas
ize
reas
ons)
Cri
tica
lIn
sult
chil
dor
say
mea
nth
ings
(Tea
se/R
CE
-cal
lsi
lly/R
CE
-ang
ry)
Sco
ldan
dcr
itic
ize
Get
onch
ild’
sba
ck/c
urse
Sco
ldin
gm
akes
chil
dim
prov
eS
cold
whe
nfa
lls
shor
t(H
old
grud
ge/t
hing
sbu
ild
up)
(Ash
amed
/oft
enan
gry)
PC
S(s
ilen
t/un
frie
ndly
/dis
appo
inte
d)P
unit
ive
Chi
ldca
nse
eI’
mup
set
Whe
nan
gry,
Ish
owit
Exp
lode
inan
ger
Spe
akto
my
chil
dca
lmly
RR
CE
-yel
lat
my
chil
dY
ell
orsh
out
(Spa
nkgr
abhi
t/P
DI-
phys
ical
)(P
hysi
cal/
confl
ict/
RC
E-p
unis
h)(S
pank
/gra
b/sl
ap/s
hove
)Pa
rent
emot
iona
lity
Neg
ativ
eM
any
thin
gsan
noy
me
Irri
tabl
eM
any
thin
gsan
noy
me
Whe
nI
get
scar
ed,
Ipa
nic
Cal
mR
Whe
nI
get
scar
ed,
Ipa
nic
Few
erfe
ars
than
mos
tR
Afr
aid
Few
erfe
ars
than
mos
tR
Get
emot
iona
llyup
set
easi
lyU
pset
Get
emot
iona
llyup
set
easi
lyTa
kes
alo
tto
mak
em
em
adR
Ang
ryTa
kes
alo
tto
mak
em
em
adR
Eve
nts
mak
em
efr
etfu
lN
ervo
usE
vent
sm
ake
me
fret
ful
Freq
uent
lyge
tdi
stre
ssed
Dis
tres
sed
Freq
uent
lyge
tdi
stre
ssed
Eas
ilyfr
ight
ened
Frig
hten
edE
asily
frig
hten
ed
Not
es:
Pare
ntin
gw
asas
sess
edin
sam
ple
1m
ainl
yby
the
pare
ntin
gsc
ale
(PS
;A
rnol
d,O
’Lea
ry,W
olff
,&
Ack
er,
1993
),in
sam
ple
2,m
ainl
yby
the
chil
dre
arin
gpr
acti
ces
repo
rt(C
RP
R)
Q-s
ort
(Blo
ck,1
981)
,and
insa
mpl
e3,
mai
nly
byth
epa
rent
ing
styl
esan
ddi
men
sion
squ
esti
onna
ire
(PS
DQ
;R
obin
son,
Man
dlec
o,O
lsen
,&H
art,
2001
).W
here
asp
ecifi
cit
emw
asdr
awn
from
one
ofth
efo
llow
ing
mea
sure
s,th
eso
urce
isin
dica
ted
besi
deth
eit
em:
RC
E=
resp
onse
sto
chil
dren
’sem
otio
ns(H
asti
ngs
&D
e,20
08);
PC
S=
psyc
holo
gica
lco
ntro
lsc
ale
(Ols
enet
al.,
2002
);P
DI
=pa
rent
ing
dim
ensi
ons
inve
ntor
y(S
late
r&
Pow
er,1
987)
;PS
I=
pare
ntin
gst
ress
inde
x-sh
ortf
orm
(Abi
din,
1995
).R
espo
nse
scal
esra
nged
from
1to
7fo
rth
eP
San
dC
RP
R,f
rom
1to
5fo
rth
eP
SI,
RC
E,P
SD
Q,a
ndP
CS
,and
from
0to
3fo
rth
eP
DI.
Com
posi
teit
ems
that
wer
ecr
eate
dth
roug
hav
erag
ing
are
encl
osed
inpa
rent
hese
s,w
ith
com
pone
ntit
ems
inth
eco
mpo
site
sepa
rate
dby
slas
hes.
R=
reve
rse-
scor
ed.I
tem
sw
ith
para
llel
cont
ent
acro
ssth
ein
stru
men
tsw
ere
used
tocr
eate
mea
sure
sof
Posi
tive
Pare
ntin
g(i
nsa
mpl
e1,
3of
12pa
rent
–chi
lddy
sfun
ctio
nal
inte
ract
ion
item
sfr
omth
eP
SI
and
all
five
disc
uss
item
sfr
omth
eP
DI;
insa
mpl
e2,
2of
7ac
cept
item
sfr
omth
eC
RP
Ran
d2
of3
rew
ard
item
sfr
omth
eR
CE
;in
sam
ple
3,2
of5
war
mth
/sup
port
and
2of
5re
ason
ing/
indu
ctio
nit
ems
from
the
PS
DQ
),C
riti
calP
aren
ting
(in
sam
ple
1,5
of10
over
reac
tivit
yit
ems
from
the
PS
;in
sam
ple
2,2
of5
reje
ctan
d2
of16
auth
orit
aria
nit
ems
from
the
CR
PR
and
1of
3pu
nish
and
1of
3m
agni
fyit
ems
from
the
RC
E;i
nsa
mpl
e3,
2of
4ve
rbal
host
ilit
yit
ems
from
the
PS
DQ
and
3lo
vew
ithd
raw
al/g
uilt
indu
ctio
nit
ems
from
the
PC
S),
and
Pun
itiv
ePa
rent
ing
(in
sam
ple
1,3
of10
over
reac
tivit
yit
ems
from
the
PS
and
all
5ph
ysic
alit
ems
from
the
PD
I;in
sam
ple
2,3
of16
auth
orit
aria
nit
ems
from
the
CR
PR
and
1of
3pu
nish
item
san
d1
of3
mag
nify
item
sfr
omth
eR
CE
;in
sam
ple
3,2
of4
verb
alho
stil
ity
and
all
four
phys
ical
coer
cion
item
sfr
omth
eP
SD
Q).
Early-emerging Internalizing Problems 237
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Tab
le3.
Des
crip
tive
Stat
isti
csan
dR
elia
bilit
ies
ofth
eM
easu
res
bySa
mpl
e
Var
iabl
es
Sam
ple
1S
ampl
e2
Sam
ple
3In
tegr
ated
sam
ple
Mor
%SD or
Na
Mor
%SD or
Na
Mor
%SD or
Na
Mor
%SD or
Na
Chi
ldte
mpe
ram
ent
Inhi
biti
on2.
921.
03.7
53.
591.
37.8
53.
401.
37.8
53.
331.
31.8
3M
ater
nal
pare
ntin
gPo
sitiv
e6.
181.
01.6
66.
29.6
6.4
75.
93.6
9.4
56.
09.7
9.5
4C
riti
cal
2.03
.60
.43
1.55
.65
.58
2.39
.79
.52
2.08
.79
.51
Pun
itiv
e3.
261.
01.5
33.
27.8
7.5
42.
48.7
7.7
12.
89.9
5.6
1M
ater
nal
emot
iona
lity
Neg
ativ
eem
otio
nali
ty2.
19.6
9.7
92.
23.7
0.8
22.
33.7
0.7
82.
27.7
0.7
7Fa
mily
soci
oeco
nom
icst
ress
Mot
her’
sed
ucat
ion
2.16
.94
3.69
.92
3.23
.76
3.08
1.03
Fath
er’s
educ
atio
n2.
14.8
53.
58.9
93.
10.8
93.
001.
05O
ccup
atio
nal
pres
tige
40.9
412
.09
54.0
712
.15
51.9
311
.31
49.7
812
.79
Fam
ilyin
com
e4.
101.
755.
731.
545.
251.
645.
081.
75Fa
mily
soci
oeco
nom
icst
ress
(z)
-.05
.75
.67
-.02
.75
.68
-.01
.79
.78
-.02
.77
.74
Fam
ilyst
ruct
ural
stre
ssD
ual-
vs.
sing
le-p
aren
tfa
mily
(81.
0)(1
02)
(82.
7)(1
10)
(89.
6)(2
16)
(85.
6)(4
28)
Two
vs.
few
erbi
olog
ical
pare
nts
(73.
8)(9
3)(7
8.2)
(104
)(8
3.4)
(201
)(7
9.6)
(398
)N
umbe
rof
chil
dren
unde
rth
eag
eof
seve
n1.
65.6
51.
60.6
51.
97.7
31.
79.7
1
Chi
ldin
tern
aliz
ing
prob
lem
sIn
tern
aliz
ing
prob
lem
sa52
.40
8.49
.85/
.70
54.4
710
.21
.85
51.1
19.
15.8
152
.33
9.37
Not
es:
For
the
inte
grat
edsa
mpl
e,C
ronb
ach’
sal
pha
reli
abil
itie
sw
ere
com
pute
dus
ing
cent
ered
data
toco
ntro
lfo
rbe
twee
n-gr
oup
vari
ance
.a
Mea
nsan
dst
anda
rdde
viat
ions
are
for
the
Tsc
ores
;sa
mpl
e1
alph
asar
efo
rth
e2–
3an
d4–
8ve
rsio
ns,
resp
ecti
vely
,of
the
chil
dbe
havi
orch
eckl
ist.
238 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
with parallel content across the two instruments (3 of 4 distress, 3 of 4 fear, and 2 of4 anger items from the EAS; eight matching emotion words from the PANAS) wereused to create a measure of negative emotionality (e.g., many things annoy me and getemotionally upset easily).
Family Socioeconomic Stress. Family stress associated with socioeconomic status wasassessed from four components: mother’s education, father’s education, highest occu-pational prestige in the family, and family income. Occupational prestige was scoredusing a recently updated version of the standard international occupational prestigescale (SIOPS; Treiman, 1977). The SIOPS was originally constructed with reference tothe International Standard Classification of Occupation (ISCO68), an internationalmeasure of occupational codes and prestige criteria, to obtain a cross-national com-parative measure of occupational status. The original SIOPS has been updated to referto the ISCO88 (Hakim, 1998). The occupations listed in the SIOPS range from thehighest scores given to professionals (e.g., lawyers, physicians, and chief executiveofficers) to the lowest given to those who hold such occupations as domestic laborers,manufacturers, and farmhands.
In all three samples, mothers were asked to report the highest levels of educationcompleted by themselves and their children’s fathers (in two-parent households).Education levels were converted into 5-point scales ranging from 1 (did not completehigh school) to 5 (attained graduate or professional degree, e.g., MD and PhD).Mothers were also asked to report the annual total family income before taxes. Familyincome was converted into a 7-point scale ranging from 1 ($0–$10 000) to 7 ($75 000or more). A principal components analysis indicated that a single-factor solution wasappropriate for the four component measures (mother’s education, father’s education,highest SIOPS score in the family, and family income), eigenvalue = 2.23, varianceaccounted for = 56 percent. An overall measure of family socioeconomic stress for eachsample was created by averaging the four measures after transforming each of theminto a standard score (z) and reversing each scale so that higher values would reflectgreater stress.
Family Structural Stress. Family stress associated with structural characteristics wasassessed by three indices: dual- vs. single-parent family, two vs. fewer biologicalparents, and number of children under the age of seven. Preliminary analyses indicatedthat number of children under the age of seven did not cohere with either of the othertwo indices; therefore, rather than form a single index of structural stress, the threevariables were examined separately in the model.
Child IP. IP were assessed by two versions of the Achenbach child behavior check-list, the child behavior checklist (CBCL) and the Achenbach system of empiricallybased assessment (ASEBA CBCL) using the age 2–3 and 4–18 versions in samples2 and 3 (ASEBA CBCL; Achenbach & Rescorla, 2001). The CBCL is the mostwidely used tool for assessing IP. It has good reliability and validity. Across CBCLversions, the IP scale comprised more than 30 items rated from 0 (not true as far asyou know) to 2 (very true or often true). Total scores were obtained by summingmothers’ ratings of the items; to account for changes in measurement over time, Tscores derived from population norms were used. To check whether the use of Tscores would obscure gender differences, we performed analyses using mean itemscores. No significant gender differences within or across samples were found.
Early-emerging Internalizing Problems 239
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Results
Preliminary Analyses
An examination of missing values revealed that for the integrated sample about 2percent of data had missing values with the exception of one variable, maternalnegative emotionality, which had 9 percent missing values due to a delay in adding itto the sample 3 protocol. Given the moderate amount of missing data for maternalnegative emotionality, it was decided to perform the main analyses using an imputationapproach (Tabachnick & Fidell, 2007). Before imputation, 58 of the 500 cases had oneor two missing values, and these data were assumed to be missing at random. Toinvestigate this assumption, we used expectation maximization (EM) imputation withthe complete dataset, as recommended by Widaman (2006) for small amounts ofmissingness. The imputation method uses all available data to estimate a new datasetwith imputed values for those that originally had missing values. Next, the model ofinterest is estimated with the newly imputed dataset. Maximum likelihood (ML)estimation simply uses all available data from the original dataset to estimate the modelof interest (Allison, 2002). We compared the results using EM with those obtainedfrom ML, and we found that the parameter estimates under the two approaches werealmost identical. The similarity of the results suggested that the more rigorous treat-ment of the missingness via EM was not necessary. Therefore, we used ML to estimatethe final model and assumed that the missing values were missing at random.
For the most part, variables were normally distributed. Although some variablesshowed skewness or kurtosis, multivariate normality was not substantially violated anddid not affect the analysis. However, one case in sample 2 was excluded from thestructural analysis owing to an unusual impact on multivariate kurtosis. Therefore, N= 499 in the final, reported model.
Test of the Measurement Model
To assess construct validity of the measures, a CFA model was fitted to each sample(Anderson & Gerbing, 1988). This measurement model constructs the latent variablesfrom their corresponding manifest variable indicators and then correlates all of thelatent variables. Good fit of the measurement model within each sample indicates agood construction of latent variables for each sample (Anderson & Gerbing, 1988). Weevaluated model fit with the robust method (Byrne, 2006) using the comparative fitindex (CFI) and root mean square error of approximation (RMSEA), which is notoverly sensitive to sample size (Bentler, 2004). By convention, CFI values of .95 orabove and RMSEA values of .06 or below are indicative of a good empirical fit(Tabachnick & Fidell, 2007). The analyses yielded CFI values of .969, .984, and .924and RMSEA values of .023, .019, and .040 for samples 1, 2, and 3, respectively. Theseindices were within acceptable ranges, indicating reasonably good fit and providingevidence for factorial validity of the measures.
Next, we performed a multiple-group CFA model with the integrated sample toevaluate the extent to which the measures functioned similarly in relation to theirunderlying constructs across samples. Typically, multi-group comparisons require con-straints among the factor loadings as well as the correlations among the latent variables(Meredith, 1993). However, in the present context, in which the latent variables for thethree samples were derived from different sets of items, invariance of the factor
240 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
loadings would not be expected and was not required (Bentler, 2004; Byrne, Shavel-son, & Muthén, 1989). Because the latent variables were the constructs of interest, wewere concerned with invariance of the correlations among the latent variables but notwith invariance of the factor loadings. If the correlation pattern for the latent variablesis invariant across the samples, then the latent variables can be assumed to represent thesame constructs for each of the three samples (Nesselroade, Gerstorf, Hardy, & Ram,2007). It should be noted that two of the latent variable correlations were found to beproblematic. Firstly, a significant difference was found for the correlation betweenparent emotionality and punitive parenting for samples 2 and 3 (c2 = 5.03, df = 1,p = .025). Secondly, a marginally significant difference was found for the correla-tion between family socioeconomic stress and critical parenting for samples 2 and 3(c2 = 3.661, df = 1, p = .056). However, given the large sample size, there was a highamount of power for detecting differences among these correlations. Thus, the fit of themeasurement model was used to assess the invariance of the correlations among thelatent variables. The model with the constraints across all correlations yielded a CFI of.952 and a RMSEA of .031 across the three samples (c2 = 55.09, df = 51, p = .32),indicating a very good fit of the model. The fit of the model with all correlations freelyestimated across the samples was very poor (c2 = 529.904, df = 129, p < .0001). Thefit for the restricted model was much better than the fit of the free model (Dc2 =507.814, df = 78, p < .0001). This finding suggests that it is reasonable to assumethat the differences in the two problematic correlations are negligible and that theindicators within all samples are modeling the same latent variables (Nesselroadeet al., 2007). Furthermore, when correlations were constrained to be direct effects inthe hypothesized model, no significant differences were detected across the samples.
Preparation for the Structural Model
Given that each set of manifest variables loaded onto one latent factor, and these latentfactors had a similar correlation pattern across samples, we next focused our attentionon the reliability of the manifest variables across the samples. Because each set ofvariables loaded onto a single factor, we created a composite score for each set ofvariables. Simple averaging was used to create the composite score rather than aweighted score, which would be sample-dependent (e.g., Cohen, 1990; Wainer, 1976).The use of single composite indicators for the latent variables within the measurementmodel allowed for a simple method of accounting for unreliability in the set ofindicators (Kline, 2005). In this method, the uniqueness is defined as the total vari-ability of the manifest variable X (or sx
2) multiplied by the proportion of unreliabilityof the manifest variables (1 - rxx) for the particular construct of interest (where rxx
represents Cronbach’s alpha for variable X). Using these constraints, x becomes theremaining proportion of variability in X, or the part of the variability in X that iscommon across all the manifest variables.
In essence, this approach extracts the part variability in the set of items that isestimated to be reliable and models only this reliable portion of the variance at thestructural level. To incorporate this within our model, we defined the residual variancefor each of the composites as the proportion of variance in the composite that wasestimated to be unreliable (one minus Cronbach’s alpha). X can be thought of as oneparticular composite within our model, e represents the proportion of the variability ofthe manifest variables that is estimated to be unreliable, and x represents the proportionthat is reliable. Therefore, the unreliability of each set of manifest variables from each
Early-emerging Internalizing Problems 241
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
sample can be accounted for in the measurement model. Because only the reliable partin each sample will go to the structural level, we know that the low reliabilities of thesets of manifest variables will not adversely affect the robustness of the model.
Cronbach’s alpha was used to estimate the reliability for each set of manifestvariables (Bedeian, Day, & Kelloway, 1997). Cronbach’s alpha was chosen because itcan be used to estimate the reliability of a composite score for a set of variables(McDonald, 1999). Once Cronbach’s alpha had been calculated for each set of mani-fest variables, the composites that summarized each set were decomposed within themeasurement model into common and uncommon variance as described above. Thisdecomposition ensured that relations at the structural level were based on reliablevariance rather than random noise within the data.
In summary, our approach to estimating the measurement model allowed differentlevels of reliability across each of the samples, as well as a direct specification ofreliability based on Cronbach’s alpha. Given that the sets of manifest variables wereeach unidimensional (good fit of a one-factor CFA for each set of manifest variables),we knew that each set could be summarized via a composite score. To account forunreliability of the composite score, we extracted the proportion of variance that wasestimated to be reliable and used only that portion at the structural level. This methodensures that the relations at the structural level reflect only reliable measurements ofthe manifest variables rather than a mixture of reliable and unreliable portions ofvariability.
Sample Differences
Prior to investigating the structural model, sample differences, as well as correlationsamong the measured variables, computed as composite indicators, were investigated.Descriptive statistics and coefficient alpha reliabilities for the variables are shown inTable 3. To examine mean differences among the three samples, a series of one-wayanalyses of variance was performed; with Bonferroni’s correction applied to follow-upcomparisons, only sample differences at p < .001 are reported. The analyses revealednumerous differences among the samples. Children in sample 1 scored lower ontemperamental inhibition than those in samples 2 and 3, t(253) = -4.37, t(362) = -3.40,respectively; the latter two samples did not differ significantly. Mothers in sample 3rated themselves as less punitive than those in samples 1 and 2, t(365) = -8.27, t(372)= -9.06, respectively; the latter two samples did not differ significantly on punitiveparenting. Mothers in sample 3 rated themselves more critical than those in sample 1,t(365) = 4.37, who in turn rated themselves more critical than those in sample 2, t(257)= 6.19. Mothers in sample 2 rated themselves more positive than those in sample 3,t(372) = 4.90, but not those in sample 1, t(257) = 1.01, NS; there was no significantdifference between mothers in samples 1 and 3, t(365) = 3.02, p = .003 (below theBonferroni criterion). Family socioeconomic stress was higher in sample 1 than insample 3, as indicated by lower scores on all four components, smallest t(356) = 6.19,and higher in sample 3 than in sample 2 on mother’s and father’s education, smallert(344) = 4.60. Family structural stress did not differ significantly across samples exceptwith respect to the number of children under the age of seven, which was higher insample 3 than in samples 1 and 2, t(365) = 4.13, t(372) = 4.85, respectively, which didnot differ significantly. Finally, mothers in sample 2 reported more IP in their childrenthan did mothers in sample 3, t(371) = 3.26; sample 1 did not differ significantly fromsamples 2 or 3. In summary, the samples differed on several of the measured variables.
242 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Levels of IP were highest in sample 2 and lowest in sample 3, possibly due to sampledifferences in the recruitment of children at risk for IP. Sample 2 mothers showed morepositive and less critical parenting than mothers in the other samples, possibly relatedto lower family stress in sample 2.
Correlations among Variables
Zero-order correlations among the composite variables are summarized in Table 4 forthe integrated sample. Due to the age range of the sample and research indicating thatchild age is negatively associated with positive parenting (Brody, Stoneman, & Burke,1987; Stevenson, Leavitt, Thompson, & Roach, 1988), child age was included todetermine whether it should be considered as a covariate in the analysis. As shown inTable 4, children with more IP were significantly more inhibited, r = .27, p < .001. Theyhad mothers whose parenting was less positive, r = -.13, p < .01, more critical, r = .16,p < .001, and more punitive, r = .14, p < .01, and whose emotionality was morenegative, r = .25, p < .001; and they lived in families under greater socioeconomicstress, r = .09, p < .05, with fewer than two parents, rs = -.10, p < .05, or with fewerthan two biological parents, rs = -.11, p < .01. Child inhibition was not related tomaternal parenting or negative emotionality. Mothers who reported more negativeemotionality used parenting practices that were less positive, r = -.20, and morecritical, r = .23, or more punitive, r = .33, all p < .001. Mothers experiencing moresocioeconomic stress were less positive, r = -.09, or more punitive, r = .11, p < .05.Positive parenting was negatively related to both critical parenting, r = -.10, p < .05,and punitive parenting, r = -.28, p < .001, which were positively related to each other,r = .33, p < .001. Child age was not substantially related to other variables.
Test of the Hypothesized Model
The model postulated pathways to IP from inhibition; from parenting practices that arelikely to increase children’s difficulty regulating anxiety reactions, especially if theyare highly inhibited and hence more prone to anxiety reactions; from negative emo-tionality in the parent that interferes with effective parenting; and from family socio-economic disadvantage and structural stressors (single-parent family, fewer than twobiological parents, and having more young children) that are likely to interfere witheffective parenting. The model was tested with EQS 6.1 (Multivariate Software,Encino, CA, USA). To avoid potential bias caused by violations of normality inassessing model adequacy, we used maximum-likelihood estimation with and withoutthe robust option. The robust option corrects for violations of the assumption ofnormality (Chou, Bentler, & Satorra, 1991). The two methods yielded similar results.We evaluated model fit with the robust method using model chi-square, CFI, andRMSEA.
We tested the model using path analysis with a multiple-groups approach thatevaluated the equality of the model structure across samples. Although child age didnot have significant zero-order correlations with other variables, given its knownassociation with less positive parenting, it was examined as a covariate in the modeland retained due to its relations with parenting. Firstly, the model was tested in each ofthe three samples without any constraints on the parameters (sample 1: c2(48) = 63.9,p = .06, CFI = .969, RMSEA = .023; sample 2: c2(48) = 61.04, p = .10, CFI = .984,
Early-emerging Internalizing Problems 243
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Tab
le4.
Zer
o-or
der
Cor
rela
tion
sam
ong
Var
iabl
esin
the
Inte
grat
edSa
mpl
e(N
=50
0)
1.2.
3.4.
5.6.
7.8.
9.10
.
1.C
hild
age
2.In
hibi
ted
-.02
3.Po
sitiv
e-.
07-.
014.
Cri
tica
l.0
2.0
1-.
10*
5.P
unit
ive
.04
-.05
-.28
***
.34*
**6.
Neg
ativ
e-.
02-.
01-.
20**
*.2
3***
.33*
**7.
Fam
ilyso
cioe
cono
mic
stre
ss.0
0-.
11*
-.09
*-.
03.1
1*.0
58.
Dua
l-vs
.si
ngle
-par
ent
fam
ilya
.00
.06
.07
.07
-.04
-.03
-.33
***
9.Tw
ovs
.fe
wer
biol
ogic
alpa
rent
sa-.
03.0
5.0
8.0
1-.
03-.
01-.
36**
*.8
1***
10.
Num
ber
ofch
ildr
enun
der
the
age
ofse
ven
.04
.06
-.08
.08
.12*
*-.
02-.
13**
.14*
**.1
6***
11.
Inte
rnal
izin
gpr
oble
ms
.03
.27*
**-.
13**
.16*
**.1
4**
.25*
**.0
9*-.
10*
-.11
**-.
02
Not
es:
aA
ssoc
iati
ons
wit
hth
ese
vari
able
sw
ere
exam
ined
usin
gS
pear
man
rank
-ord
erco
rrel
atio
ns.
*p
<.0
5,**
p<
.01,
***
p<
.001
.
244 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
RMSEA = .019; and sample 3: c2(88) = 131.82, p = .002, CFI = .924, RMSEA = .04).These models included the specifications for the residual variance as described earlier(Kline, 2005). Because all CFI values were larger than .95 and all RMSEA values weresmaller than .05, these models were judged to be very good. Next, we constrained thepath coefficients to be equal across the three groups. This model showed equally goodfit, c2(51) = 55.09, p = .32, CFI = .991, RMSEA = .022. The hypothesis that associa-tions between IP and the three parenting indices would be stronger for more inhibitedchildren was tested by including parenting ¥ temperament interaction terms in themodel. All were nonsignificant, indicating no moderation effects by inhibition, b = .47,.11, and 1.23 for the interaction with positive, critical, and punitive parenting, respec-tively, all p > .40. Fit indices for the non-constrained model were c2(96) = 192.92, p <.01, CFI = .96, RMSEA = .08, and those for the constrained model (all interaction pathcoefficients constrained to zero) were c2(105) = 180.54, p < .01, CFI = .96, RMSEA =.10. No significant difference in model fit could be detected between the constrainedand non-constrained models (Dc2(9) = 12.37, p = .19). These results suggest thateffects of parenting on IP were not moderated by temperamental inhibition; therefore,these moderating effects were excluded from the final model. To examine reciprocalrelations between critical parenting and IP and between positive parenting and IP, weperformed two separate analyses including the inverse instead of the hypothesized pathin the model. The coefficient for the reverse path from IP to critical parenting wasnonsignificant [b = .01, p = .15; c2(38) = 46.14, p = .17, CFI = .98, RMSEA = .03], andthe coefficient for the path from IP to positive parenting was also nonsignificant [b =-.01, p = .10; c2(38) = 45.17, p = .20, CFI = .98, RMSEA = .03]. On the basis of thesefindings, reciprocal relations between parenting dimensions and IP were excluded fromthe final model. Finally, additional multiple-group analyses revealed no path coefficientdifferences as a function of dual- vs. single-parent family [Dc2(18) = 15.24, p = .65],two vs. fewer biological parents [Dc2(18) = 17.97, p = .46] or child gender [Dc2(18) =16.14, p = .58]. For each of these three analyses, constraints were placed on all of thepath coefficients across the two specified groups. The large p values (i.e., >.05) indicateno significant difference while constraining these paths across the groups; therefore,there was no detectable difference among the relationships of the latent variables as afunction of group membership. Fit statistics for the non-constrained models werec2(10) = 6.64, p = .78, CFI = 1.0 and RMSEA = .001 for dual- vs. single-parent family;c2(10) = 12.04, p = .36, CFI = .99 and RMSEA = .035 for two vs. fewer biologicalparents; and c2(10) = 11.78, p = .78, CFI = .99 and RMSEA = .03 for boys vs. girls.Therefore, these variables were excluded from the final model.
The results of the final model are shown in Figure 1. Independent variables in themodel were allowed to freely covary. Covariances for each pair of independent vari-ables were estimated to measure the level of dependence not accounted for by the directeffects in the model. The covariances for each sample are not provided in Figure 1because they differ across samples. Because multiple-groups analysis constrains theunstandardized coefficients to be equal across groups, it does not yield a common setof standardized coefficients (Kline, 2005). For this reason, the unstandardized solutionis shown in the figure, which provides coefficients for the significant paths. Coeffi-cients reflect, for each unit increase in a predictor, the expected unit increase ordecrease in the predicted variable, given that the other parameters remain unchanged.Note that the unstandardized coefficients for prediction of parenting variables aresmaller in magnitude than those for prediction of IP due to differences in the mea-surement scales for these variables (1–7 for parenting and 29–76 for IP).
Early-emerging Internalizing Problems 245
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
As Figure 1 shows, most hypothesized relations were significant. Unstandardizedcoefficients indicate strong direct predictive relations between IP and child inhibition(b = 2.64, p < .01), maternal low positive (b = -1.98, p = .02) and high critical (b = 3.20,p = .02) parenting, maternal negative emotionality (b = 3.22, p < .01), and familysocioeconomic stress (b = 2.25, p < .01). Maternal punitive parenting was stronglypredicted by maternal negative emotionality (b = .50, p < .01) and family socio-economic (b = .19, p < .01) and structural (children under the age of seven) (b = .17,p < .01) stress; however, there was no association between children’s IP and punitiveparenting over and above the associations with less positive and more critical parent-ing. There were pathways to IP from negative emotionality and family structural stress(having more children under the age of 7) via less positive parenting (b = -.29, p < .01,and b = -.12, p < .01, respectively) and more critical parenting (b = .31, p < .01, andb = .09, p = .02, respectively), and from socioeconomic stress via less positiveparenting (b = -.12, p < .01).
Mediation by Parenting. The hypothesis that maternal negative emotionality andfamily stress would be associated with IP via their impact on parenting was tested byexamining indirect effects estimated as the product of the path coefficients linking theindependent variable to the mediator and linking the mediator to the dependent vari-able. Indirect effects are shown in Figure 1 by the significant paths to IP from maternalnegative emotionality and from family structural stress (having more children underthe age of seven) through low positive and high critical parenting, and from familysocioeconomic stress through low positive parenting. To test the significance of
Child Age
Family StructuralStress
(Children Under the Age of Seven)
FamilySocioeconomic
Stress
MaternalNegative
Emotionality
Critical Parenting
PositiveParenting
PunitiveParenting
TemperamentalInhibition
InternalizingProblems
.08*
.09*
.31***
-.10* -.12**
-.12*
-.29***
.19** .17***
.50***
3.20*
-1.98*
2.25***
3.22***
2.64***
Figure 1. Unstandardized Path Coefficients Depicting Direct and Indirect pathwaysLinking Family, Parent, and Child Variables to Internalizing Problems in Early Child-hood. Notes: Significant paths are represented by solid lines and nonsignificant pathsby dashed lines.* p < .05, ** p < .01, *** p < .001.
246 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
mediation, we used bootstrapping procedures (MacKinnon, Lockwood, Hoffman,West, & Sheets, 2002; Shrout & Bolger, 2002), which involve taking multiple randomsamples of observations from the original sample, with replacement, and analyzingthem to provide a distribution of parameter estimates. Using EQS, we created bootstrapsamples and computed confidence intervals (CI) to assess the significance of theindirect effects. Distributions of parameter estimates tend to be skewed (Bollen &Stine, 1990; Lockwood & MacKinnon, 1998; MacKinnon et al., 2002); therefore,bias-corrected confidence intervals were computed (Efron & Tibshirani, 1993;MacKinnon, Lockwood, & Williams, 2004). Maternal negative emotionality was asso-ciated with IP through high critical parenting (95 percent CI: .121–3.308, p = .03) andthrough low positive parenting (90 percent CI: .029–1.375, p = .08). Family socioeco-nomic stress was associated with IP through low positive parenting (95 percent CI:.007–.909, p = .045).
Discussion
The purpose of this study was to test an ecological model of the development ofearly-emerging IP, postulating pathways to IP from temperamental inhibition, frommaternal parenting practices that are likely to increase children’s difficulty regulatinganxiety reactions, from maternal negative emotionality that interferes with effectiveparenting, and from family socioeconomic and structural sources of stress that arelikely to interfere with effective parenting. Using a form of integrative data analysis(e.g., Curran & Hussong, 2009), three independent samples were integrated to createa large and heterogeneous sample of children assessed on these multilevel factors.Structural analysis performed with a multiple-groups approach revealed that associa-tions were invariant across samples. Child inhibition, less positive and more criticalparenting, maternal negative emotionality, and family socioeconomic disadvantagewere found to have direct associations with IP. In addition, there was some evidence formediation by parenting. Maternal negative emotionality was associated with IPthrough more critical parenting, and both negative emotionality and socioeconomicstress were associated with IP through less positive parenting. Reciprocal relationsbetween IP and parenting dimensions were nonsignificant. These findings provide thefirst evidence in a large and heterogeneous sample that early-emerging IP may beunderstood from an ecological perspective as involving multilevel factors and that theimpact of more distal factors operates through effects on parenting.
In keeping with the model, most of the key factors hypothesized to contribute to IPwere found to be directly associated with IP. Each of these multilevel factors mayincrease the risk of IP in early childhood, and they may do so in an additive fashion.The finding of a direct link between inhibition and IP is consistent with theory andresearch identifying inhibited temperament as a vulnerability factor in the developmentof IP (e.g., Rapee, 2001). The finding that inhibition was relatively independent of theparenting and family contextual factors that were associated with IP could be inter-preted as evidence for the dispositional or genetic basis of this temperament trait(Biederman et al., 1990; Rosenbaum et al., 1993), and for the notion that inhibitedtemperament and parenting may not be strongly related in the absence of cumulativerisk (e.g., Paulussen-Hoogeboom, Stams, Hermanns, & Peetsma, 2007). Temperamen-tal inhibition is thought to reflect the behavioral manifestation of underlying biologicaldifficulties with emotion regulation (Rubin, Coplan, Fox, & Calkins, 1995). Thedevelopment of emotion regulation is fundamental to young children’s ability to
Early-emerging Internalizing Problems 247
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
manage their stress reactions. Inhibited children’s greater physiological reactivity tonovel and social stressors and resultant avoidant coping strategies leave them prone toheightened anxiety and IP. This risk may be independent of any effect inhibitedchildren’s reactivity has on parenting, at least in the context of low cumulative risk.
Also consistent with current theory and research (Coplan et al., 2008; Hastings,Nuselovici, Rubin, & Cheah, 2010), early IP were directly associated with less positiveand more critical maternal parenting. During early childhood, children lacking externalsupport to develop emotion regulation skills due to low parental warmth or engagementor high parental rejection or negativity will be more prone to states of anxiety and lessable to manage these states. Past research has linked punitive parenting to IP (McLeod,Wood, et al., 2007), and although such was the case in the present study when consid-ering their zero-order correlation, the relation was not significant once the contributionof critical parenting was taken into account. Given the association found betweenpunitive and critical parenting, possibly owing to the element of hostility common tothese parenting dimensions, critical parenting might reflect the aspect of emotionallynegative parenting specifically salient for children’s IP more strongly than punitiveparenting. These findings support models suggesting that psychological control(behaviors such as criticism that undermine self-efficacy) is an important risk factor forIP (e.g., Barber, 1996).
The association between IP and low positive and high critical parenting is consistentwith recent research linking IP not only to negative parenting but also to a lack ofpositive parenting (Bayer et al., 2006; Coplan et al., 2008). Either or both the absenceof positive responses and the presence of negative responses may play a role in themechanisms mediating the effect of parenting on children’s risk of developing IP.Parenting practices that foster understanding of rules and show respect for childautonomy should protect against IP by contributing to the development of self-regulation and a sense of competence and self-efficacy. Conversely, parenting practicesthat convey criticism and rejection of the child should increase the risk of IP bycontributing to poor self-regulation and a sense of incompetence. Research withmultiple time points is needed to determine whether parenting does in fact have alasting influence on the development of IP and whether it occurs through theseprocesses. Such research would also permit analysis of the reciprocal influences thatmay occur between parenting and child inhibition.
Family stress and parent characteristics showed both direct and indirect associationswith IP via parenting. IP were predicted by socioeconomic stress directly and indirectlythrough less positive parenting and by maternal negative emotionality directly andindirectly through more critical parenting, associations reflecting significant mediationby parenting. These associations provide evidence that family stressors may be linkedto young children’s IP through multiple processes, in accord with ecological models(Bronfenbrenner & Ceci, 1994), that posit a chain of reciprocating influences leadingfrom socioeconomic factors at the family level to the psychological resources ofparents, the quality of parenting, and children’s psychological adjustment (Conger &Dogan, 2007; Rubin & Mills, 1991). Thus, family and parent factors may contribute tothe development of children’s IP through multiple processes, including parentingpractices, genetic transmission of emotionality from parent to child, or other learningprocesses.
In summary, child temperamental inhibition, family socioeconomic status, maternalnegative emotionality, and maternal low positive and high critical parenting appearedto be key independent factors in early-emerging IP, and maternal negative emotionality
248 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
appeared to have both direct and indirect associations with IP through critical parent-ing. Interestingly, there were no moderating effects of inhibition, possibly reflecting therelative lack of consistency in studies of the interactive contributions of temperamentand parenting to adjustment (Bates & Pettit, 2007). This study’s failure to find signifi-cant moderating effects cannot be attributed to a lack of power, inasmuch as theintegrated sample afforded excellent power to detect even small effect sizes. Rather,these results suggest that, in early childhood, separate individual factors may besufficient by themselves to increase or decrease the risk of IP. To the extent that this isso, it would appear that developmental trajectories for IP may have multiple highlyspecific starting points, in accord with the principal of equifinality. Whether problemstend to persist, worsen, or improve thereafter, however, may depend on the cumulativeand interactive effects of multiple risk and protective factors. To test this idea, researchis needed to identify the factors associated with persisting, worsening, or improvingtrajectories of IP.
There are limitations of the study that must be considered. Most of the measureswere constructed by selecting items that were parallel in content due to the use ofdifferent measurement instruments in the three separate samples. Measurement errorwas likely higher than would have been the case with identical measures, and as such,the findings may represent an underestimate of the relations that exist. Nevertheless, asdescribed in the Preparation for the Structural Model subsection of the Results section,this limitation is mitigated by the use of structural equation modeling techniquesdesigned to account for measurement error by incorporating the individual samplereliabilities in the model. The results of the analyses supported most of the hypothesesand showed equality of the model across samples, providing replication of the findingsacross three independent samples. Another limitation is the use of reports frommothers only, which may have increased the degree of correspondence between them.Although the use of a single source of information may not entirely account for thepattern of relations that were found (e.g., consistent with theory, IP were predictedmore strongly by critical than by punitive parenting), future work should replicate thefindings using multiple raters and different methods of measurement.
Also limiting the study is its cross-sectional design, which necessitates caution ininterpreting the causal directions reflected by the present results. The findings areconsistent with theory and research suggesting that temperament, parenting, andfamily context contribute to the development of IP. Moreover, reciprocal relationsbetween IP and parenting dimensions were nonsignificant. However, the possibilityremains that IP are a cause as well as a consequence of less positive or more criticalparenting. Due to their deference and appearance of vulnerability, internalizing chil-dren may elicit parental over- or under-reactions, either of which could interfere withthe development of emotion regulation skills and, therefore, contribute to anxiety.Thus, an important research direction is to examine trajectories of developmentacross childhood. Such research is also needed to determine which aspects of chil-dren’s psychological functioning serve as the mechanisms by which internal dispo-sitional vulnerabilities and external relational and contextual influences becomemanifest as IP.
Despite these limitations, the present study has important implications. Firstly, itdemonstrates the power and efficacy of integrative data analysis and the advantagesthat can be afforded to researchers by combining complementary samples to investi-gate complex developmental questions. Secondly, the findings raise the possibility thatsingle risk factors could be starting points for the development of IP. Single risk factors
Early-emerging Internalizing Problems 249
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
may lead to the worsening of IP through ripple effects that elevate other risk factorscompounding the child’s vulnerability. Thirdly, the findings highlight the need formore investigation of the role parenting may play in the development of IP. Parentingpractices may be a key link in the chain of effects leading to or away from youngchildren’s IP, with family stressors affecting parenting practices both by decreasingpositive parenting and by increasing negative parenting. Whether the associationsfound in the present study are in fact attributable to these processes awaits longitudinalanalysis.
References
Abidin, R. R. (1995). Parenting stress index: Manual (3rd ed.). Odessa, FL: PsychologicalAssessment Resources.
Achenbach, T. M. (1991). Manual for the child behavior checklist/4–18 and 1991 profile.Burlington, VT: University of Vermont, Department of Psychiatry.
Achenbach, T. M. (1992). Manual for the child behavior checklist/2–3 and 1992 profile.Burlington, VT: University of Vermont, Department of Psychiatry.
Achenbach, T. M., & Rescorla, L. A. (2001). Manual for ASEBA school-age forms and profiles.Burlington, VT: University of Vermont, Research Center for Children, Youth, & Families.
Allison, P. D. (2002). Missing data. Thousand Oaks, CA: Sage.Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in practice: A review
and recommended two step approach. Psychological Bulletin, 103, 411–423.Arnold, D. S., O’Leary, S. G., Wolff, L. S., & Acker, M. M. (1993). The parenting scale:
A measure of dysfunctional parenting in discipline situations. Psychological Assessment,5, 137–144.
Ashford, J., Smit, F., van Lier, P. A. C., Cuijpers, P., & Koot, H. M. (2008). Early risk indicatorsof internalizing problems in late childhood: A 9-year longitudinal study. Journal of ChildPsychology and Psychiatry, 49, 774–780.
Bagozzi, R. P., & Edwards, J. R. (1998). A general approach to representing constructs inorganizational research. Organizational Research Methods, 1, 45–87.
Barber, B. K. (1996). Parental psychological control: Revisiting a neglected construct. ChildDevelopment, 67, 3296–3319.
Bates, J. E., & Pettit, G. S. (2007). Temperament, parenting, and socialization. In J. E. Grusec& P. D. Hastings (Eds.), Handbook of socialization: Theory and research (pp. 153–177). NewYork: Guilford.
Bayer, J. K., Sanson, A. V., & Hemphill, S. A. (2006). Parent influences on early childhoodinternalizing difficulties. Journal of Applied Developmental Psychology, 27, 542–559.
Bedeian, A. G., Day, D. V., & Kelloway, E. K. (1997). Correcting for measurement errorattenuation in structural equation models: Some important reminders. Measurement, 57,785–799.
Bentler, P. M. (2004). EQS 6 structural equations program manual. Encino, CA: MultivariateSoftware.
Biederman, J., Rosenbaum, J. F., Hirshfeld, D. R., Faraone, S. V., Bolduc, B. A., Gersten,M., et al. (1990). Psychiatric correlates of behavioral inhibition in young children ofparents with and without psychiatric disorders. Archives of General Psychiatry, 47, 21–26.
Block, J. H. (1981). The Child-rearing Practices Report (CRPR): A set of Q items for thedescription of parental socialization attitudes and values. Unpublished manuscript, Berkeley,CA: University of California, Institute of Human Development.
Bollen, K. A., & Stine, R. (1990). Direct and indirect effects: Classical and bootstrap estimatesof variability. Sociological Methodology, 20, 115–140.
Bolton, D., Eley, T. C., O’Connor, T. G., Perrin, S., Rabe-Hesketh, S., Rijsdijk, F., et al. (2006).Prevalence and genetic and environmental influences on anxiety disorders in 6-year-old twins.Psychological Medicine, 36, 335–344.
Boomsma, D. I., van Beijsterveldt, C. E. M., & Hudziak, J. J. (2005). Genetic and environmentalinfluences on anxious/depression during childhood: A study from the Netherlands TwinRegister. Genes, Brain, and Behavior, 4, 466–481.
250 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Brody, G., Stoneman, Z., & Burke, M. (1987). Child temperaments, maternal differentialtreatment and sibling relationships. Developmental Psychology, 23, 354–362.
Bronfenbrenner, U., & Ceci, S. J. (1994). Nature-nurture reconceptualized in developmentalperspective: A bioecological model. Psychological Review, 101, 568–586.
Buss, A. H., & Plomin, R. (1984). Temperament: Early developing personality traits. Hillsdale,NJ: Erlbaum.
Byrne, B. M. (2006). Structural equation modeling with EQS basic concepts, applications, andprogramming (Rev. ed.). Mahwah, NJ: Erlbaum.
Byrne, B. M., Shavelson, R. J., & Muthén, B. (1989). Testing for the equivalence of factor andmean structures: The issue of partial measurement invariance. Psychological Bulletin, 105,456–466.
Campisi, L., Serbin, L. A., Stack, D. M., Ledingham, J. E., & Schwartzman, A. E. (2009).Precursors of language ability and academic performance: An intergenerational study ofat-risk children. Infant and Child Development, 18, 377–403.
Chou, C.-P., Bentler, P. M., & Satorra, A. (1991). Scaled test statistics and robust standard errorsfor nonnormal data in covariance structure analysis: A Monte Carlo study. British Journal ofMathematical and Statistical Psychology, 44, 347–357.
Cohen, J. (1990). Things I have learned (so far). American Psychologist, 45, 1304–1312.Conger, R. D., & Conger, K. J. (2002). Resilience in Midwestern families: Selected findings
from the first decade of a prospective, longitudinal study. Journal of Marriage and the Family,64, 361–373.
Conger, R. D., & Dogan, S. J. (2007). Social class and socialization in families. In J. E. Grusec& P. D. Hastings (Eds.), Handbook of socialization: Theory and research (pp. 433–460). NewYork: Guilford.
Coplan, R. J., Arbeau, K. A., & Armer, M. (2008). Don’t fret, be supportive! Maternalcharacteristics linking child shyness to psychosocial and school adjustment in kindergarten.Journal of Abnormal Child Psychology, 36, 359–371.
Cudeck, R. (1989). Analysis of correlation matrices using covariance structure models. Psy-chological Bulletin, 105, 317–327.
Curran, P. J., & Hussong, A. M. (2009). Integrative data analysis: The simultaneous analysis ofmultiple data sets. Psychological Methods, 14, 81–100.
Efron, B., & Tibshirani, R. (1993). An introduction to the bootstrap. New York: Chapman &Hall/CRC.
Elgar, F. J., Mills, R. S. L., Waschbusch, D. A., Brownridge, D. A., & McGrath, P. J. (2007).Maternal and paternal depressive symptoms and child maladjustment: The mediating role ofparental behavior. Journal of Abnormal Child Psychology, 35, 943–955.
Gallagher, K. C. (2002). Does child temperament moderate the influence of parenting onadjustment? Developmental Review, 22, 623–643.
Hakim, C. (1998). Social change and innovation in the labour market: Evidence from theCensus SARs on occupational segregation and labour mobility, part-time work and studentjobs, homework and self-employment. Oxford: Oxford University Press.
Hammen, C., & Rudolph, K. D. (2003). Childhood mood disorders. In E. J. Mash & R. A.Barkley (Eds.), Child psychopathology (2nd ed., pp. 233–278). New York: Guilford.
Harris, K. M., Raley, R. K., & Rindfuss, R. R. (2002). Family configurations and child-carepatterns: Families with two or more preschool-age children. Social Science Quarterly, 83,455–471.
Hastings, P. D., & De, I. (2008). Parasympathetic regulation and parental socialization ofemotion: Biopsychosocial processes of adjustment in preschoolers. Social Development, 17,211–238.
Hastings, P. D., Nuselovici, J. N., Rubin, K. H., & Cheah, C. S. L. (2010). Shyness, parenting,and parent-child relationships. In K. H. Rubin & R. J. Coplan (Eds.), The development ofshyness and social withdrawal (pp. 107–130). New York: Guilford.
Hastings, P. D., Sullivan, C., McShane, K. E., Coplan, R. J., Utendale, W. T., & Vyncke, J. D.(2008). Parental socialization, vagal regulation and preschoolers’ anxious difficulties: Directmothers and moderated fathers. Child Development, 79, 45–64.
Henderson, S. H., Hetherington, E. M., Mekos, D., & Reiss, D. (1996). Stress, parenting,and adolescent psychopathology in nondivorced and stepfamilies: A within-family per-spective. In E. M. Hetherington & E. A. Blechman (Eds.), Advances in family research:
Early-emerging Internalizing Problems 251
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Stress, coping, and resiliency in children and families (pp. 39–66). Hillsdale, NJ:Erlbaum.
Hirshfeld-Becker, D. R., & Biederman, J. (2002). Rationale and principles for early interventionwith young children at risk for anxiety disorders. Clinical Child and Family PsychologyReview, 5, 161–172.
Jenkins, J. M., Rasbash, J., & O’Connor, T. G. (2003). The role of the shared family context indifferential parenting. Developmental Psychology, 39, 99–113.
Kagan, J., & Snidman, N. (1999). Early childhood predictors of adult anxiety disorders. Bio-logical Psychiatry, 46, 1536–1541.
Kline, R. B. (2005). Principles and practice of structural equation modeling (2nd ed.). NewYork: Guilford.
Letcher, P., Smart, D., Sanson, A., & Toumbourou, J. W. (2009). Psychosocial precursors andcorrelates of differing internalizing trajectories from 3 to 15 years. Social Development, 18,618–646.
Little, T. D., Cunningham, W.A., Shahar, G., & Widaman, K. F. (2002). To parcel or not to parcel:Exploring the question, weighing the merits. Structural Equation Modeling, 9, 151–173.
Lockwood, C. M., & MacKinnon, D. P. (1998). Bootstrapping the standard error of themediated effect. Paper presented at the 23rd Annual Meeting of SAS Users Group Interna-tional, Cary, NC.
MacKinnon, D. P., Lockwood, C. M., Hoffman, J. M., West, S. G., & Sheets, V. (2002). Acomparison of methods to test mediation and other intervening variable effects. PsychologicalMethods, 7, 83–104.
MacKinnon, D. P., Lockwood, C. M., & Williams, J. (2004). Confidence limits for the indirecteffect: Distribution of the product and resampling methods. Multivariate BehavioralResearch, 39, 99–128.
Majcher, D., & Pollack, M. H. (1996). Childhood anxiety disorders. In L. T. Hechtman (Ed.), Dothey grow out of it? Long-term outcomes of childhood disorders (pp. 139–169). Washington,DC: American Psychiatric Association.
McDonald, R. P. (1999). Test theory: A unified treatment. Mahwah, NJ: Erlbaum.McLeod, B. D., Weisz, J. R., & Wood, J. J. (2007). Examining the association between parenting
and childhood depression: A meta-analysis. Clinical Psychology Review, 27, 986–1003.McLeod, B. D., Wood, J. J., & Weisz, J. R. (2007). Examining the association between parenting
and childhood anxiety: A meta-analysis. Clinical Psychology Review, 27, 155–172.Meredith, W. (1993). Measurement invariance, factor analysis and factor invariance. Psy-
chometrika, 58, 525–544.Mills, R. S. L., Freeman, W. S., Clara, I. P., Elgar, F. J., Walling, B. R., & Mak, L. (2007). Parent
proneness to shame and the use of psychological control. Journal of Child and Family Studies,16, 359–374.
Mills, R. S. L., Imm, G. P., Walling, B. R., & Weiler, H. A. (2008). Cortisol reactivity andregulation associated with shame responding in early childhood. Developmental Psychology,44, 1369–1380.
Nesselroade, J. R., Gerstorf, D., Hardy, S. A., & Ram, N. (2007). Idiographic filters forpsychological constructs. Measurement, 5, 217–235.
Olsen, S. F., Yang, C., Hart, C. H., Robinson, C. C., Wu, P., Nelson, D. A., et al. (2002). Mothers’psychological control and preschool children’s behavioral outcomes in China, Russia, andthe United States. In B. K. Barber (Ed.), Intrusive parenting: How psychological controlaffects children and adolescents (pp. 235–262). Washington, DC: American PsychologicalAssociation.
Paulussen-Hoogeboom, M. C., Stams, G. J. J. M., Hermanns, J. M. A., & Peetsma, T. T. D.(2007). Child negative emotionality and parenting from infancy to preschool: A meta-analyticreview. Developmental Psychology, 43, 438–453.
Rapee, R. M. (2001). The development of generalized anxiety. In M. W. Vasey & M. R. Dadds(Eds.), The developmental psychopathology of anxiety (pp. 481–503). New York: OxfordUniversity Press.
Robinson, C. C., Mandleco, B. L., Olsen, S. F., & Hart, C. H. (2001). Parenting styles anddimensions questionnaire (PSDQ). In J. Touliatos, B. F. Perlmutter, & M. A. Straus (Eds.),Handbook of family measurement techniques (3rd ed., pp. 188–189). Thousand Oaks, CA:Sage.
252 Rosemary S. L. Mills, Paul D. Hastings, Jonathan Helm et al.
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012
Rosenbaum, J. F., Biederman, J., Bolduc-Murphy, B. A., Faraone, S. V., Chaloff, J., Hirshfeld,D. R., et al. (1993). Behavioral inhibition in childhood: A risk factor for anxiety disorders.Harvard Review of Psychiatry, 1, 2–16.
Rothbart, M. K., Ahadi, S. A., Hershey, K. L., & Fisher, P. (2001). Investigations of tempera-ment at three to seven years: The children’s behavior questionnaire. Child Development, 72,1394–1408.
Rubin, K. H., Coplan, R. J., Fox, N. A., & Calkins, S. A. (1995). Emotionality, emotionregulation and preschoolers’ social adaptation. Development and Psychopathology, 7, 49–62.
Rubin, K. H., & Mills, R. S. L. (1991). Conceptualizing developmental pathways to internal-izing disorders in childhood. Canadian Journal of Behavioural Science, 23, 300–317.
Rubin, K. H., Nelson, L. J., Hastings, P., & Asendorpf, J. (1999). The transaction betweenparents’ perceptions of their children’s shyness and their parenting styles. InternationalJournal of Behavioral Development, 23, 937–957.
Rushton, J. P., Brainerd, C. J., & Pressley, M. (1983). Behavioral development and constructvalidity: The principle of aggregation. Psychological Bulletin, 94, 18–38.
Shrout, P. E., & Bolger, N. (2002). Mediation in experimental and nonexperimental studies:New procedures and recommendations. Psychological Methods, 7, 422–445.
Slater, M. A., & Power, T. G. (1987). Multidimensional assessment of parenting in single-parentfamilies. In J. P. Vincent (Ed.), Advances in family intervention, assessment and theory(pp. 197–228). Greenwich, CN: JAI Press.
Stevenson, M. B., Leavitt, L. A., Thompson, R. H., & Roach, M. A. (1988). A social relationsmodel analysis of parent and child play. Developmental Psychology, 24, 101–108.
Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics (5th ed.). Boston, MA:Allyn and Bacon.
Treiman, D. J. (1977). Occupational prestige in comparative perspective. New York: AcademicPress.
Wainer, H. (1976). Estimating coefficents in linear models: It don’t make no nevermind.Psychological Bulletin, 83, 213–217.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measuresof positive and negative affect: The PANAS scales. Journal of Personality and Social Psy-chology, 54, 1063–1070.
Weinraub, M., Horvath, D. L., & Gringlas, M. B. (2002). Single parenthood. In M. H. Bornstein(Ed.), Handbook of parenting (2nd ed., Vol. 3, pp. 109–140). Mahwah, NJ: Erlbaum.
Widaman, K. F. (2006). Missing data: What to do with or without them. Monographs of theSociety for Research in Child Development, 71, 42–64.
Zahn-Waxler, C., Klimes-Dougan, B., & Slattery, M. J. (2000). Internalizing problems ofchildhood and adolescence: Prospects, pitfalls, and progress in understanding the develop-ment of anxiety and depression. Development and Psychopathology, 12, 443–466.
Acknowledgments
This research was supported by Canadian Institutes of Health Research grant MOP-74642awarded to the team, and by grants from Health Canada (Child and Youth Division), awarded toLisa Serbin and colleagues from the Concordia Longitudinal Risk Project, from Social Sciencesand Humanities Research Council of Canada and the Fonds de la Recherche en Santé du Québecgrants awarded to Paul Hastings for the Daycare and Preschool Adjustment Study, and fromCanadian Institutes of Health Research grant MOP-57670 to Rosemary Mills for the Shame inChildhood Study. We thank the children and parents who made this research possible andgratefully acknowledge the assistance of Farriola Ladha, Claude Senneville, Nadine Girouard,and Bobbi Walling. We also thank Keith Widaman for statistical consultation.
Early-emerging Internalizing Problems 253
© Blackwell Publishing Ltd. 2011 Social Development, 21, 2, 2012