Child Development, May/June 2000, Volume 71, Number 3, Pages 695–702
Attachment from Infancy to Early Adulthood in a High-Risk Sample:
Continuity, Discontinuity, and Their Correlates
Nancy S. Weinfield, L. Alan Sroufe, and Byron Egeland
This study explores the stability of attachment security and representations from infancy to early adulthood in a
sample chosen originally for poverty and high risk for poor developmental outcomes. Participants for this study
were 57 young adults who are part of an ongoing prospective study of development and adaptation in a highrisk
sample. Attachment was assessed during infancy by using the Ainsworth Strange Situation (Ainsworth &
Wittig) and at age 19 by using the Berkeley Adult Attachment Interview (George, Kaplan, & Main). Possible correlates
of continuity and discontinuity in attachment were drawn from assessments of the participants and their
mothers over the course of the study. Results provided no evidence for significant continuity between infant and
adult attachment in this sample, with many participants transitioning to insecurity. The evidence, however, indicated
that there might be lawful discontinuity. Analyses of correlates of continuity and discontinuity in attachment
classification from infancy to adulthood indicated that the continuous and discontinuous groups were differentiated
on the basis of child maltreatment, maternal depression, and family functioning in early adolescence.
These results provide evidence that although attachment has been found to be stable over time in other samples,
attachment representations are vulnerable to difficult and chaotic life experiences.
INTRODUCTION
Bowlby (1969/1982, 1980) wrote that internal working
models should become more resistant to change
over time given a stable, reinforcing environment, but
he did not go so far as to indicate that change is impossible
over the course of development. Bowlby
adopted and extended a concept from Waddington
(1957) that development can be conceived of as a
branching set of pathways, like a tree lying on its side.
If security and insecurity were described as two diverging
pathways early in development, the further
the branches one took moved away from the other
major pathway, the more difficult converging with
that pathway would become. Thus, if a person with an
insecure organization experiences more rejection and
implements stronger defenses with time, transitioning
to security will become less likely for that person.
These branching pathways, however, also leave open
the possibility of changes between security and insecurity.
For example, if an individual with a secure
organization had many negative relationship experiences,
he or she could eventually transition to an insecure
organization.
This study examines continuity and discontinuity
in attachment from infancy to age 19 longitudinally in
a sample at high risk for poor developmental adaptation
and also examines some possible correlates of
such continuity and discontinuity. Like the other
studies described here, this study examines the relation
between infant attachment and adult attachment.
It does differ from the other studies, however, in some
important ways. This sample is a low-income sample,
selected for high risk for poor developmental outcomes.
There is likely to be less stability in the environments
and interpersonal relationships of these participants
than in other samples, and likewise there may be
less stability in their attachment classifications.
Correlates of continuity and discontinuity are examined
somewhat differently in this study than in the
other studies presented here. The reason for this departure
is the high occurrence of negative life events
in this sample. Considering negative life events to consist
of being born to a single mother, parental divorce,
life-threatening illness of parent or child, serious parental
drug or alcohol problem, child experiencing
physical or sexual abuse, or a parent or other custodial
attachment figure dying, 91.2% of the participants in
this sample experienced at least one negative life
event. This high rate of negative life events makes it
statistically inadvisable to conduct analyses comparing
those participants who experienced negative life
events with those who did not. Consequently, correlates
of continuity and discontinuity are examined by
© 2000 by the Society for Research in Child Development, Inc.
All rights reserved. 0009-3920/2000/7103-0015
This is one of three long-term longitudinal studies assessing
infant attachment. See Waters, Hamilton, & Weinfield, “The Stability
of Attachment Security from Infancy to Adolescence and
Early Adulthood: General Introduction,” for an overall view of
study design, measures, and supporting references.
696 Child Development
conducting analyses comparing particular experiences
of stable versus nonstable attachment groups.
These experience variables were chosen because of
their theoretical and empirical relevance to attachment
relationships. The correlates explored here are
maternal life stress, maternal depression, child experience
of maltreatment, and family functioning in
early adolescence.
All four of these variables have empirical or theoretical
importance for the study of attachment. Maternal
life stress has been shown to predict changes in attachment
security during infancy (Egeland & Farber,
1984; Vaughn, Egeland, Sroufe, & Waters, 1979). Maternal
depression has been shown to have wide-ranging
negative effects on children (see Downey & Coyne,
1990, for a review), including a tendency toward insecure
attachment relationships (DeMulder & RadkeYarrow,
1991; Egeland & Farber, 1984; Gaensbauer,
Harmon, Cytryn, & McKnew, 1984; Radke-Yarrow,
Cummings, Kuczynski, & Chapman, 1985). Infants
who have been maltreated within their families are
more likely than nonmaltreated infants to show a
change in their attachment relationships away from
security (Egeland & Sroufe, 1981; Schneider-Rosen,
Braunwald, Carlson, & Cicchetti, 1985). Finally, if the
quality of attachment representations are responsive
to changes in attachment relationships, quality of
family functioning should be related to stability and
change in attachment classification.
METHOD
Participants
Participants were 59 young adults, ages 18–19, a
subset of the participants from the Minnesota
Mother–Child Project, which is an ongoing prospective
longitudinal study of development and adaptation
in children at risk for poor developmental outcomes.
The original full sample consisted of 267
primiparous mothers and their infants. Mothers were
recruited from public health clinics during the third
trimester of pregnancy. Mothers were young (M ϭ 20,
range ϭ 12–34), mostly single (62%), and many (40%)
had not completed high school. The majority of the
pregnancies were unplanned (82%), and most families
were earning incomes at or below the poverty
level. Overall, the lives of the sample members were
characterized by high levels of stress and instability
and low levels of social support (Egeland & Brunnquell,
1979). Eighty percent of the women were European
American, 13% were African-American, and 7%
were Hispanic or Native American.
The current study focuses on development in a
subgroup of the children from the original sample.
The subgroup examined in this study consisted of
participants from two groups (not mutually exclusive).
One group consisted of participants who had
participated in a nursery school through the Mother–
Child Project and attended a summer camp together
through the Project at age 10. This group was originally
chosen for more intensive study, in part because
of stable infant attachment classifications from the 12and
18-month assessments. The second group consisted
of those participants who had become parents
by the age of 19. The stable infant attachment group
was chosen to increase confidence in the accuracy of
the infant attachment classifications. The group that
had become parents was added because the Adult Attachment
Interview had been used previously with
teen mothers. Base rates of infant attachment security
were similar in both subgroups; security was slightly
more strongly represented in the teen parent group.
One participant was excluded from analyses because
low IQ and medication brought the validity of her
Adult Attachment Interview into question, and another
participant was excluded because infant attachment
information was not available; therefore, the
participant total was 57 (25 males and 32 females). Racial
distribution of this subgroup was as follows: 61%
European American, 16% African American, and 23%
mixed racial background. This subgroup did not differ
significantly from the original sample of 267 on
base rates of infant attachment security, maternal education,
maternal age, family socioeconomic status
(SES), proportion of single mothers, or mother’s racial
or ethnic group membership. Sample size may
differ slightly across analyses because of missing data
for some participants at some time points.
Procedures and Measures
General Procedures
Data for the larger longitudinal study have been
collected at regular intervals, beginning during the
mothers’ third trimester of pregnancy and extending
through their children’s 19th birthdays. Information
used in this study has been collected through interviews
with the mothers or their children and videotaped observations
of the mothers and children together.
Attachment Measures
The Strange Situation. Infant attachment was assessed
in the Strange Situation (Ainsworth, Blehar,
Waters, & Wall, 1978; Ainsworth & Wittig, 1969) when
the children were 12 months old and again when they
Weinfield, Sroufe, and Egeland 697
were 18 months old. Each Strange Situation was coded
by two coders. Twelve- and 18-month Strange Situations
were coded by separate coders, and 18-month coders
were blind to 12-month attachment classification. Interrater
agreement for the 12-month classifications
was 89%; agreement for the 18-month classifications
was 93%. Disagreements at both times were resolved
through discussion. Participants were classified
as insecure-avoidant, secure, or insecure-resistant. The
disorganized-disoriented classification (Main & Solomon,
1990) was not considered in these analyses.
For the purposes of analyses in this study, only
one designation of infant attachment classification is
used for each participant. The single infant attachment
classification was determined by using all
available infant attachment data to further increase
confidence in the accuracy of the infant attachment
classifications. For the portion of this subsample that
was chosen specifically for stable attachment classification
from 12 to 18 months, the classification that
was made at both 12 and 18 months is used to represent
infant attachment status (n ϭ 36). For participants
for whom only one Strange Situation assessment
is available, that single classification is used to
represent infant attachment status (n ϭ 6). For participants
whose infant attachment classification differed
at 12 and 18 months, a single classification was
determined through interrater conferencing, considering
the 12- and 18-month classifications as well as
an evaluation at 24 months (n ϭ 9). If no 24-month
evaluation is available and the 12- and 18-month
classifications differ, the 12-month classification is
used to represent infant attachment classification because
the Strange Situation procedure was originally
designed for use at that age (n ϭ 6).
The Berkeley Adult Attachment Interview. The Berkeley
Adult Attachment Interview (George, Kaplan, &
Main, 1985) was administered as a part of a battery of
measures during an interview that took place around
the participants’ 19th birthdays.
During the transcription process, interviews were
blinded for all identifying information, including
original participant number and names of people
and places so that coders familiar with some participants
would be blind to the identities of participants
while coding. All interviews were coded by the first
author, with the exception of five interviews where
the identity of the participant could not be sufficiently
blinded for that coder. All coders were trained
in coding by Mary Main and Erik Hesse and have established
reliability with them. Two particularly
challenging interviews were sent out for expert coding
by Mary Main and Erik Hesse. Interrater reliability
was established through double coding of a randomly
selected set of cases from this sample.
Percent agreement between coders was 82% across
the three major classifications (␬ ϭ .67, p Ͻ .001).
Participants were classified as insecure-dismissing,
secure-autonomous, or insecure-preoccupied. Although
unresolved status was assessed in this sample,
it will not be considered in these analyses because
no participants in this subsample were classified as
unresolved but autonomous.
Measures of Correlates of Continuity
and Discontinuity of Attachment
Maternal life stress. Maternal life stress was assessed
at 11 time periods between child age 18
months and 19 years by using the Life Events Scale
(Egeland & Deinard, 1975), adapted from Cochrane
and Robertson’s (1973) Life Events Inventory for use
with a low SES sample. The Life Events Scale consists
of a 40-item interview that inquires about a variety of
stressful life experiences that may have occurred in
the period of time since the last interview, including
such topics as family conflict, difficulties with neighbors,
instability of living situation, work-related
problems, health problems, and trouble with the law.
Events were rated for both for occurrence and severity.
The Life Events Scale has been shown to predict
both concurrent and subsequent child functioning
(Herzog, 1984; Pianta, 1985), thus demonstrating that
self-reported maternal life stress does relate to child
outcomes. Scores from these measures were averaged
to obtain one score for each participant that represented
average total maternal life stress over the period
between the Strange Situation assessment and
the Adult Attachment Interview. Reliability was evaluated
through interrater agreement on the total life
stress scores. Reliability was assessed with Pearson r
correlations and ranged from r ϭ .86 to r ϭ .96 for individual
time points.
Maltreatment. The Mother–Child Project began as
a prospective study of child maltreatment. As such,
the staff of the Mother–Child Project has assessed
regularly whether there is any evidence of maltreatment
of the children in this study by their caregivers
(see Egeland & Sroufe, 1981). Using information
from mother interviews and home observations,
groups of participants were identified as having been
physically abused, verbally abused, neglected, or
subject to maternal psychological unavailability during
childhood. Maltreatment is collapsed across category
of abuse for analyses. Maltreatment scores are
coded 0 (not identified as maltreated) or 1 (identified
as maltreated).
Maternal depression. Maternal depression was as-
698 Child Development
sessed with two widely used self-report inventories
of depressive symptomatology. The Center for Epidemiological
Studies Depression Scale, or CES-D
(Radloff, 1977), was administered at 48 months, 16
years, and 17.5 years. The Beck Depression Inventory
(Beck, Ward, Mendelson, Mock, & Erbaugh,
1961) was administered at the second-grade and
third-grade interviews.
For the purposes of analyses, and to maintain the integrity
of the individual measures, depression was
evaluated in terms of the number of times a mother
scored either (1) at high risk for depression on the
CES-D or (2) above borderline risk for depression on
the BDI. To allow for missing data, scores were then
transformed to indicate the proportion of assessments
during which a mother met the criteria for depression
(taking into account the number of assessments that
were available for each case). Consequently, the maternal
depression score ranges from 0 to 1.
Family functioning. Qualities of parent–child interaction
were assessed during an observational assessment
of family functioning during interaction tasks
at age 13. One group of scales, the “balance” scales, is
used in these analyses (for the full scales, see Fleeson,
1988). The three balance scales were developed by
using concepts drawn from family systems theory
and are so called because they assess how well the
dyad balances the multiple functions of their relationship
roles during the session. The scales assess
how secure the family members are in their roles and
whether they seem to feel free to act naturally and
spontaneously within those roles; whether the relationship
supports the development and needs of the
individuals; whether the individuals take pride in
and enjoy the relationship; and whether the relationship
allows a commitment to goal-oriented teamwork
to meet the requirements of the tasks. The balance
scales have been shown to relate to the child’s egoresiliency
(Sroufe, 1991), as well as to subsequent
adolescent reports of family functioning (Ogawa &
Weinfield, 1997). Interrater reliability on the three
balance scales had an average intraclass correlation
of ␳I ϭ .61.
For the purposes of these analyses the balance
scales are composited by summing across the three
scale scores. The reason for this compositing is to provide
a broader index of the quality of relationship
functioning at age 13. The family balance composite
has a possible range from 3 to 21.
Relations between these four potential correlates
of continuity and discontinuity of attachment are presented
in Table 1. Correlations range in magnitude
from .17 to .38.
RESULTS
Three-Level Contingency Analysis
A␹2 analysis was conducted to assess continuity in
three-category attachment classification (avoidant/
secure/resistant to dismissing/secure/preoccupied)
from infancy to adulthood (see Table 2). The rate of
continuity was 38.6%. The nonsignificant ␹2, ␹2(4, N ϭ
57) ϭ 1.19, ns, indicates that there was no significant
continuity between infant and adult attachment. For
infant attachment, a secure classification was the predominant
classification (59.6%). In contrast, for adult
attachment, an insecure-dismissing classification was
the predominant classification (59.6%).
Two-Level Contingency Analyses
Another ␹2 analysis was conducted to assess continuity
in two-category attachment classification from
infancy to adulthood, collapsing across the two insecure
classifications at each time point to obtain a secure/insecure
differentiation. The rate of continuity
was 50.9%. The nonsignificant ␹2, ␹2(1, N ϭ 57) ϭ .54,
ns, indicates there was no significant continuity between
infant and adult attachment.
To clarify this finding and explore whether certain
characteristics of the data or the sample can account
for it, the two-level contingency analysis was recalculated
by using several different criteria: (1) female parTable
1 Intercorrelations between Attachment-Related Life
Events
Variable 1 2 3 4
1. Maternal life stress
2. Child maltreatment .35**
3. Maternal depression .38** .17
4. Family balance Ϫ.35** Ϫ.34* Ϫ.26
* p Ͻ .05; ** p Ͻ .01.
Table 2 Stability of Attachment Classifications from Infancy to
Adulthood
Adult Attachment
Classification (Adult
Attachment Interview)
Infant Attachment Classification
(Strange Situation)
Avoidant
(A)
Secure
(B)
Resistant
(C) Total
Dismissing (Ds) 9 20 5 34
Secure (F) 4 12 2 18
Preoccupied (E) 2 2 1 5
Total 15 34 8 57
Weinfield, Sroufe, and Egeland 699
ticipants only, (2) male participants only, (3) participants
with stable infant attachment classifications
across 12 and 18 months only, (4) using 12-month infant
attachment classifications only, (5) using 18-month infant
attachment classifications only, and (6) teen parents
only. Continuity was not strikingly different, and certainly
not significantly different, for any of the vari-
ants.
The categorical nature of attachment classifications
could also possibly mask continuity that would be revealed
if continuous measures were used. To test this
hypothesis, continuous measures were derived for infant
and adult attachment. For infant attachment, the
number of times (across the 12- and 18-month assessments)
that a participant was classified as secure was
used as a continuous measure of attachment, resulting
in a score ranging from 0–2. For adult attachment
the Coherence of Transcript rating scale was used.
This scale is strongly related to classification: high coherence
is a quality of secure transcripts, and low
coherence is a quality of insecure transcripts. The correlation
between these two continuous measures of
attachment was r(51) ϭ .14, ns, and covarying out
attachment-related negative life events did not bring
the correlation to significance. It seems, then, that this
finding of a lack of significant continuity between infant
and adult attachment is a consistent finding within
this sample.
Analyses of Correlates of Continuity
and Discontinuity
Despite the lack of significant continuity between
infant and adult attachment classifications, an a priori
examination of correlates of stability and change was
carried out. The rationale for doing so was that continuity
has already been established between infant
and adult attachment classifications in other samples
(Hamilton, 2000; Waters, Merrick, Treboux, Crowell,
& Albersheim, 2000), thus the two classification variables
can be assumed to be related, and an exploration
of correlates is justified.
For the purpose of these analyses, participants
were divided into four groups representing the possibilities
of continuity and discontinuity: (1) infant insecureadult
insecure (n ϭ 17), (2) infant secure-adult secure
(n ϭ 12), (3) infant insecure-adult secure (n ϭ 6), and
(4) infant secure-adult insecure (n ϭ 22). For each potential
correlate, planned contrasts were conducted
comparing the groups that have the same infant classification
but differing adult classification (e.g., infant
secure-adult secure with infant secure-adult insecure).
Planned contrasts were chosen over ANOVA
with regular post hoc tests to take a theoretically
driven approach, thereby eliminating analyses that
were not of direct interest (e.g., stable secure versus
stable insecure groups, although likely differing on
correlates, are not directly relevant to the stability versus
change question). Planned contrasts protect
against Type II error by reducing the likelihood that a
true effect will be obscured by variation that is not of
theoretical interest. Planned contrasts do not require
the precondition of a significant omnibus test; thus no
omnibus tests are reported (for a full discussion of
planned contrasts, see Hays, 1988). Means and standard
deviations on the correlates for the four groups
appear in Table 3. Although child maltreatment is a
categorical variable, it is statistically permissible to
use the variable as continuous given that the magnitude
of the dimension in question (maltreatment) differs
across values of the variable (Cohen & Cohen,
1975). Because child maltreatment is a dichotomous
variable, however, standard deviations for the variable
are not reported.
Stressful life events. Although the secure-secure
group did have a lower level of average total maternal
life stress than the secure-insecure group, this finding
was nonsignificant, t(53) ϭ Ϫ1.24, ns. The means for
Table 3 Means and Standard Deviations on Correlates of Continuity
and Discontinuity in Attachment
Adult Attachment
Classification
Correlate
Infant Attachment
Secure
M (SD)
Insecure
M (SD)
Average maternal life stress
sample range ϭ 3.6–17.7,
N ϭ 57
Secure 8.3 (3.6) 10.0 (3.9)
Insecure 11.2 (3.5) 11.4 (4.1)
Child maltreatmenta
sample range ϭ 0–1,
N ϭ 57
Secure .25 .14
Insecure .00 .41
Maternal depression
sample range ϭ .0–.6,
N ϭ 57
Secure .05 (.09) .15 (.19)
Insecure .23 (.15) .13 (.22)
Family balance scores
sample range ϭ 8.0–19.5,
N ϭ 54
Secure 15.0 (1.8) 14.1 (2.7)
Insecure 14.5 (2.1) 12.0 (2.3)
a No standard deviations are indicated for the child maltreatment
variable because the variable is dichotomous. The mean indicates
the proportion of participants in the group who were maltreated.
700 Child Development
the insecure infant groups were very similar, and
again the difference was nonsignificant, t(53) ϭ .09,
ns. Overall, maternal stressful life events were not
found to be a correlate of stability and change in attachment
classification within this sample.
Maltreatment. Child maltreatment did not differentiate
the secure-secure group from the secure-insecure
group, t(18.4) ϭ .76, ns. It did, however, differentiate
the insecure-insecure group from the insecure-secure
group, t(16) ϭ 3.35, p ϭ .004. The descriptive data reveal
that whereas the mean for the insecure-insecure
group was .41 (41% of this group had been identified
as maltreated), the mean for the insecure-secure
group was actually .00 (no participants in this group
had been identified as maltreated). Thus it appears
that those participants who had an insecure infant attachment
relationship and an insecure adult attachment
classification were significantly more likely to
have experienced maltreatment than those participants
with insecure infant attachment who transitioned
to adult security.
To explore whether this finding could be attributed
to a general difference between secure and insecure
adult attachment classification rather than a difference
specific to the infant insecure group, a t-test was
conducted to determine whether the mean for maltreatment
differed across adult attachment classification
(secure versus insecure adult groups). The difference
between the mean for the adult secure group
(.17) and the mean for the adult insecure group (.28)
was nonsignificant, t(55) ϭ .74, ns. Consequently, the
finding may be interpreted to be specific to the infant
insecure group (see Table 3).
Maternal depression. Maternal depression did not
differentiate the insecure-insecure group from the insecure-secure
group, t(13.3) ϭ Ϫ1.27, ns. Maternal depression
did differentiate the secure-secure group
from the secure-insecure group, t(31.5) ϭ Ϫ2.13, p ϭ
.04. The secure-insecure group had mothers who
were significantly more often depressed than the
mothers of the secure-secure group.
To examine the possibility that this finding could
be attributed to adult attachment outcomes rather
than change groups specifically, a t-test was conducted
to examine differences in maternal depression
across adult attachment classification (secure versus
insecure adult groups). Results indicated that mean
maternal depression did not differ specifically by
adult attachment classification, t(55) ϭ .61, ns. The
finding, then, may be attributed specifically to the infant
secure stability and change groups.
Family functioning. Although the secure-secure
group had a higher mean family balance composite
score than the secure-insecure group, this finding was
nonsignificant, t(50) ϭ 1.00, ns. The insecure-secure
group had a higher mean family balance composite
score than the insecure-insecure group, and this difference
was significant, t(50) ϭ Ϫ2.24, p ϭ .03. The
insecure-secure group seems to have had better family
functioning at age 13 than the stable insecure group.
To examine the possibility that this finding could be
attributed to adult attachment outcomes rather than
change groups specifically, a t-test was conducted to examine
differences in family balance composite scores
across adult attachment classification (secure versus insecure
adult groups). Results indicated that the mean
family balance composite score was higher for the adult
secure group than for the adult insecure group t(52) ϭ
Ϫ2.32, p ϭ .03. Although this does not indicate that the
within infant insecure group finding is spurious, it indicates
that caution is warranted in interpreting this finding
because it may be associated with outcome rather
than change specifically.
DISCUSSION
In looking at the distribution of adult attachment classifications,
one aspect of this sample is particularly
striking: the predominance of insecure classifications,
specifically insecure-dismissing classifications. This
is unusual because in most middle-class samples the
secure-autonomous classification is the predominant
classification (e.g., Benoit & Parker, 1994; Das Eiden,
Teti, & Corns, 1995; Fonagy, Steele, & Steele, 1991). In
a study looking at attachment in high-risk adolescent
mothers and their infants, however, the adult attachment
classification distribution also showed a preponderance
of insecure dismissing classifications (Ward &
Carlson, 1995). Possibly there is something about the
combination of high rates of attachment-related negative
life events and the period of late adolescence that
makes a dismissing state of mind with respect to attachment
more likely at this point of development. Waters,
Weinfield, and Hamilton (2000) address this issue
further in the General Discussion for this set of papers.
There was no significant continuity in attachment
classification from the Strange Situation to the Adult
Attachment Interview. One possible explanation is
that within a high-risk sample that has a less stable environment
and less stable relationships than a middleclass
sample, there would be less continuity. This has
been suggested theoretically by Bowlby (1980) and
borne out in studies of continuity of infant attachment
classifications in high-risk samples from 12 to 18 months
(Schneider-Rosen et al., 1985; Vaughn et al., 1979). Waters
et al. (2000) and Hamilton (2000) found that change
was more likely in participants who had experienced
negative life events. Within this sample, negative life
Weinfield, Sroufe, and Egeland 701
events were all too common. Therefore, the low stability
rate could likely be accounted for by the chaotic
lives and negative experiences within this sample.
Another possible source of discontinuity is measurement
error. The concern over the role of measurement
error is minimized, however, by the fact that all
AAI coders had established reliability with Mary
Main, and interrater reliability, although not high, was
well within acceptable ranges. The finding that family
functioning in adolescence differentiates participants
who will later be classified as secure or insecure on the
Adult Attachment Interview also reinforces the validity
of the classifications at age 19.
Several variables representing life circumstances
were found to relate to continuity and change in attachment
classification from infancy to adulthood.
One variable that did not relate to continuity or discontinuity
was maternal life stress. One possibility is
that low power and large standard deviations were
responsible for the nonsignificant finding. Another
possibility is that as these children grew into adolescence
and early adulthood the influence of their
mothers’ life stress diminished and their own, perhaps
different, life stress became more crucial.
Stable insecure participants were significantly more
likely to have been maltreated during childhood than
infant insecure participants who transitioned to security.
Interestingly, rather than being instrumental in a
change from security to insecurity as SchneiderRosen
and her colleagues (1985) found in infancy,
maltreatment seems to be relevant to the maintenance
of insecurity from infancy to adulthood within this
high-risk sample. It may be that within this sample
maltreatment was what Bowlby (1980) described as a
reinforcing experience for the insecure pathway.
The measure of family functioning must be interpreted
with some caution because it also differentiates
secure and insecure participants in general on theAdult
Attachment Interview. Insecure participants who transitioned
to security had better family functioning at age
13 than the insecure participants who remained insecure.
This can best be interpreted as a marker of positive
change within the family. The families of the insecure
participants who changed to security seemed to
balance the needs of the family, the individuals, and the
task better; they allowed for comfortable expressions of
opinions and unfettered cooperation toward task goals.
Mothers of individuals who transitioned from security
to insecurity were more likely to report clinically
significant depressive symptoms than mothers
of stable secure individuals. One possibility is that the
mother’s depression itself is affecting her parenting
and consequently the relationship and promoting a
change to insecurity. Another equally plausible possibility
is that a third as yet unspecified variable is influencing
both the maternal depression and the
child’s shift to insecurity.
Overall, this study emphasizes the nature of risk in
the lives of children who grow up in poverty and chaotic
environments. Although research has demonstrated
that infant security might provide a buffer
that allows for continued competence in the face of
adversity, it does not seem to make certain continued
security. Difficult life experiences take their toll on
children despite possible avenues of resilience. Research
has, however, indicated that individuals with
secure early histories may show a rebound to better
functioning after setbacks (Sroufe, Egeland, & Kreutzer,
1990), and possibly this will be true in this sample.
Some of the dismissing participants (particularly
those who were secure in infancy) might transition to
“earned” security later if, as they mature, they are able
to explore the meaning of their negative experiences.
Two particularly important directions for future research
are suggested by this study. One important step
would be to replicate these results in another high-risk
poverty sample. This type of research is difficult to do,
however, because it requires extensive longitudinal
work extending from infancy to adulthood. Such work
with high-risk samples is unusual. The next step that is
possible to do is confirming these findings with the full
Mother–Child project sample. The issue that would be
addressed by doing a full sample analysis would be
similar to that addressed in replication: ensuring that
characteristics particular to this subsample are not
driving results that would not hold true in a different
sample. It would also be informative to reinterview a
selected group of these participants at a later age. Of
specific interest would be looking at those participants
who were secure in infancy but insecure as adults to
see if time, maturity, and autonomy from their families
of origin allow them to work through their difficult
pasts and reclaim their security.
ACKNOWLEDGMENTS
This research was supported in part by a grant from the
National Institutes of Mental Health (MH40864-08), as
well as by a Dissertation Fellowship Award granted by
the University of Minnesota to the first author.
ADDRESSES AND AFFILIATIONS
Corresponding author: Nancy S. Weinfield, Department
of Psychology, Gilmer Hall, P.O. Box 400400,
University of Virginia, Charlottesville, VA 22904;
e-mail: nweinfield@virginia.edu. L. Alan Sroufe
and Byron Egeland are at the University of Minnesota,
Minneapolis.
702 Child Development
REFERENCES
Ainsworth, M. D. S., Blehar, M., Waters, E., & Wall, S. (1978).
Patterns of attachment: A psychological study of the strange
situation. Hillsdale, NJ: Erlbaum.
Ainsworth, M. D. S., & Wittig, B. A. (1969). Attachment and
exploratory behavior in one-year-olds in a strange situation.
In B. M. Foss (Ed.), Determinants of infant behavior
(Vol. 4.) London: Methuen.
Beck, A., Ward, C., Mendelson, M., Mock, J., & Erbaugh, J.
(1961). An inventory for measuring depression. Archives
of General Psychiatry, 4, 53–63.
Benoit, D., & Parker, K. (1994). Stability and transmission of
attachment across three generations. Child Development,
65, 1444–1456.
Bowlby, J. (1969/1982). Attachment and loss: Vol. 1. Attachment.
New York: Basic Books.
Bowlby, J. (1980). Attachment and loss: Vol. 2. Loss. New York:
Basic Books.
Cochrane, R., & Robertson, A. (1973). The life events inventory:
A measure of the relative severity of psychosocial
stressors. Journal of Psychosomatic Research, 17, 135–139.
Cohen, J., & Cohen, P. (1975). Applied multiple regression/correlation
analysis for the behavioral sciences. Hillsdale, NJ: Erlbaum.
Das Eiden, R., Teti, D., & Corns, K. (1995). Maternal working
models of attachment, marital adjustment, and the parent–
child relationship. Child Development, 66, 1504–1518.
DeMulder, E., & Radke-Yarrow, M. (1991). Attachment with
affectively ill and well mothers: Concurrent behavioral
correlates. Development and Psychopathology, 3, 227–242.
Downey, G., & Coyne, J. (1990). Children of depressed parents:
An integrative review. Psychological Bulletin, 108, 50–76.
Egeland, B., & Brunnquell, D. (1979). An at-risk approach to
the study of child abuse: Some preliminary findings.
Journal of the American Academy of Child Psychiatry, 18,
219–235.
Egeland, B., & Deinard, A. (1975). Life stress scale and manual.
Unpublished manuscript, University of Minnesota.
Egeland, B., & Farber, E. (1984). Infant–mother attachment:
Factors related to its development and changes over
time. Child Development, 55, 753–771.
Egeland, B., & Sroufe, L. A. (1981). Developmental sequelae
of maltreatment in infancy. In R. Rizley & D. Cicchetti
(Eds.), Developmental perspectives in child maltreatment
(pp. 77–92). San Francisco: Jossey-Bass.
Fleeson, J. A. (1988). Assessment of parent–adolescent relationships:
Implications for adolescent development. Unpublished
doctoral dissertation, University of Minnesota.
Fonagy, P., Steele, H., & Steele, M. (1991). Maternal representations
of attachment during pregnancy predict the
organization of infant-mother attachment at one year of
age. Child Development, 62, 891–905.
Gaensbauer, T. J., Harmon, R. J., Cytryn, E. M., & McKnew,
D. H. (1984). Social and affective development in infants
with a manic-depressive parent. American Journal of Psychiatry,
141, 223–238.
George, C., Kaplan, N., & Main, M. (1985). An adult attachment
interview: Interview protocol. Unpublished manuscript,
University of California at Berkeley.
Hamilton, C. E. (2000). Continuity and discontinuity of attachment
from infancy through adolescence. Child Development,
71, 690–694.
Hays, W. L. (1988). Statistics, 4th Edition. Fort Worth, TX:
Holt, Rinehart & Winston, Inc.
Herzog, J. (1984). Life events as indices of family stress: Relationships
with children’s current levels of competence. Unpublished
doctoral dissertation, University of Minnesota.
Main, M., & Solomon, J. (1990). Procedures for identifying
infants as disorganized/disoriented during the Strange
Situation. In M. T. Greenberg, D. Cicchetti, & E. M. Cummings
(Eds.), Attachment in the preschool years. Chicago:
University of Chicago Press.
Ogawa, J. R., & Weinfield, N. S. (1997,April). Amulti-method investigation
of precursors of adult attachment classification.
Poster session presented at the biennial meeting of the Society
for Research in Child Development, Washington, DC.
Pianta, R. (1985). The longitudinal effects of maternal life stress
on the developmental outcomes of first grade children in a
high-risk sample. Unpublished doctoral dissertation, University
of Minnesota.
Radke-Yarrow, M., Cummings, E. M., Kuczynski, L., &
Chapman, M. (1985). Patterns of attachment in two- and
three-year olds in normal families and families with parental
depression. Child Development, 36, 884–893.
Radloff, L. (1977). The CES-D scale: Aself-report depression
scale for research in the general population. Applied Psychological
Measurement, 1, 385–401.
Schneider-Rosen, K., Braunwald, K., Carlson, V., & Cicchetti,
D. (1985). Current perspectives in attachment theory:
Illustration from the study of maltreated infants. In
I. Bretherton & E. Waters (Eds.), Monographs of the Society
for Research in Child Development, 50(1–2, Serial No. 209),
194–210.
Sroufe, J. (1991). Assessment of parent–adolescent relationships:
Implications for adolescent development. Journal
of Family Psychology, 5, 21–45.
Sroufe, L.A., Egeland, B., & Kreutzer, T. (1990). The fate of
early experience following developmental change: Longitudinal
approaches to individual adaptation in childhood.
Child Development, 61, 1363–1373.
Vaughn, B., Egeland, B., Sroufe, L. A., & Waters, E. (1979).
Individual differences in infant–mother attachment at
twelve and eighteen months: Stability and change in
families under stress. Child Development, 50, 971–975.
Waddington, C. H. (1957). The strategy of genes. London:
Allen & Unwin.
Ward, M. J., & Carlson, E. (1995). Associations among adult
attachment representations, maternal sensitivity, and
infant–mother attachment in a sample of adolescent
mothers. Child Development, 66, 69–79.
Waters, E., Merrick, S., Treboux, D., Crowell, J., & Albersheim,
L. (2000). Attachment security in infancy and early
adulthood: A 20-year longitudinal study. Child Development,
71, 684–689.
Waters, E., Weinfield, N. S., & Hamilton, C. E. (2000). The stability
of attachment security from infancy to adolescence
and eary adulthood: General discussion. Child Development,
71, 703–706.