Attachment and the Experience and Expression of Emotions in Romantic Relationships: A Developmental Perspective

Journal of Personality and Social Psychology
Copyright 2007 by the American Psychological Association
2007, Vol. 92, No. 2, 355–367
0022-3514/07/$12.00
DOI: 10.1037/0022-3514.92.2.355
Attachment and the Experience and Expression of Emotions in Romantic
Relationships: A Developmental Perspective
Jeffry A. Simpson, W. Andrew Collins, SiSi Tran, and Katherine C. Haydon
University of Minnesota
In this longitudinal study, the authors tested a developmental hypothesis derived from attachment theory
and recent empirical findings. Target participants were 78 individuals who have been studied intensively
from infancy into their mid-20s. When targets were 20 –23 years old, the authors tested the way in which
interpersonal experiences at 3 pivotal points in each target’s earlier social development—infancy/early
childhood, early elementary school, and adolescence—predicted the pattern of positive versus negative
emotions experienced with his or her romantic partner. A double-mediation model revealed that targets
classified as securely attached at 12 months old were rated as more socially competent during early
elementary school by their teachers. Targets’ social competence, in turn, forecasted their having more
secure relationships with close friends at age 16, which in turn predicted more positive daily emotional
experiences in their adult romantic relationships (both self- and partner-reported) and less negative affect
in conflict resolution and collaborative tasks with their romantic partners (rated by observers). These
results are discussed in terms of attachment theory and how antecedent life experiences may indirectly
shape events in current relationships.
Keywords: attachment, emotions, romantic relationships, peer relationships
Many of the most intense emotions arise during the formation, the
Close relationships are the setting in which some of life’s most
maintenance, the disruption, and the renewal of attachment relation-
tumultuous emotions are experienced. Echoing this viewpoint,
ships. The formation of a bond is described as falling in love, main-
Berscheid and Reis (1998) have argued that identifying both the
taining a bond as loving someone, and losing a partner as grieving
origins and the profile of emotions that are experienced in a
over someone. Similarly, threat of loss arouses anxiety, and actual loss
relationship is essential if one wants to understand the core defin-
gives rise to sorrow, while each of these situations is likely to arouse
ing features of a relationship. Against this backdrop, one might
anger. The unchallenged maintenance of a bond is experienced as a
expect that a great deal would be known about emotions in
source of security, and the renewal of a bond as a source of joy.
Because such emotions are usually a reflection of the state of a
relationships, especially how significant relationship experiences
person’s affectional bonds, the psychology and psychopathology of
at critical stages of social development forecast the type and
emotion is found to be in large part the psychology and psychopa-
intensity of emotions experienced in adult attachment relation-
thology of affectional bonds.
ships. Surprisingly little is known about these issues, however (see
—John Bowlby, Attachment and Loss: Vol. 3. Loss: Depression,
Berscheid & Regan, 2004; Shaver, Morgan, & Wu, 1996). Using
and Sadness
attachment theory (Bowlby, 1969, 1973, 1980) as an organizing
framework, we designed the current longitudinal study to fill these
crucial conceptual and empirical gaps in our knowledge.
Jeffry A. Simpson and SiSi Tran, Department of Psychology, University
of Minnesota; W. Andrew Collins and Katherine C. Haydon, Institute of
Attachment Theory and Emotions
Child Development, University of Minnesota.
Bowlby’s attachment theory provides a unique and comprehen-
The research described in this article was supported by National Insti-
sive account of the normative (i.e., species-typical) and individual
tutes of Health Grants R01-MH40864 to Byron Egeland, L. Alan Sroufe,
and W. Andrew Collins, by National Institute of Mental Health (NIMH)
difference (i.e., individual-specific) processes that generate emo-
Grant R01-MH49599 to Jeffry A. Simpson, and by NIMH Training Grant
tions in close relationships. According to Bowlby, the attachment
MH19893 to Katherine C. Haydon. The pioneering work of Byron Ege-
system serves two principal functions—to protect vulnerable indi-
land, L. Alan Sroufe, and Elizabeth A. Carlson in the early phases of this
viduals from potential threats and to regulate subsequent negative
longitudinal study is gratefully acknowledged. We thank Chris Federico for
affect. The normative component of the theory specifies the stimuli
his advice on the structural equation modeling analyses.
and contexts that typically elicit and extinguish specific kinds of
Correspondence concerning this article should be addressed to Jeffry A.
emotions as well as the sequence of emotions commonly experi-
Simpson, Department of Psychology, University of Minnesota, 75 E. River
enced in response to certain relational events (e.g., the sequence of
Road, Minneapolis, MN 55455-0344. E-mail: simps108@umn.edu
protest, despair, and detachment that normally unfolds during
Inquiries regarding the Minnesota Study of Risk and Adaptation from
prolonged separations from attachment figures). The individual
Birth to Adulthood should be addressed to W. Andrew Collins, Institute of
Child Development, University of Minnesota, 51 E. River Road, Minne-
difference component articulates how one’s personal history of
apolis, MN 55455-0345. E-mail: wcollins@umn.edu
receiving care and support from attachment figures across the life
355

---

356
SIMPSON, COLLINS, TRAN, AND HAYDON
span shapes the goals, working models, and coping strategies that
attached should yearn to achieve greater “felt security.” Those who
one uses when emotion-eliciting stimuli or events occur in rela-
are avoidantly attached should strive to obtain and maintain inter-
tionship contexts. Following Bowlby’s formulation, most research
personal autonomy and control. These countervailing goals, work-
on the significance of early attachment for later relationships relies
ing models, and coping strategies should shunt individuals who
on the distinction between secure and insecure attachment histories
have different attachment histories toward distinct patterns of
(e.g., G. I. Roisman, personal communication, April 13, 2006;
emotional experience in later relationships.
Roisman, Collins, Sroufe, & Egeland, 2005; also see Thompson,
1999; Waters & Cummings, 2000).
Experience of Emotion in Relationships
Kobak and Sceery (1988) have suggested that the way in which
individuals perceive and manage emotions in relationships should
Research has confirmed that the frequency and intensity of daily
depend on the nature of the working models formed in response to
emotions experienced in relationships act as a good barometer of
their specific attachment histories. Drawing on the attachment
how close individuals feel to their partners (Barrett, Robin, Pi-
classifications that are widely used in cross-sectional studies of
etromonaco, & Eyssell, 1998). Experiencing strong and frequent
adult attachment (see Ainsworth, 1989; Main & Goldwyn, 1998),
emotions in a relationship can communicate that one truly cares
Kobak and Sceery reasoned as follows:
about a partner and a relationship (Clark, Fitness, & Brissette,
2001). Indeed, individuals who view their relationships in
Secure attachment [should] be organized by rules that allow acknowl-
communal-oriented rather than exchange-oriented terms report
edgement of distress and turning to others for support, avoidant
expressing both more positive and more negative emotions when
attachment by rules that restrict acknowledgement of distress and the
interacting with their romantic partners (Clark & Brissette, 2000).
associated attachment attempts to seek comfort and support, and
Much of what is known about the experience of emotions in
ambivalent attachment by rules that direct attention toward distress
and attachment figures in a hypervigilant manner that inhibits the
relationships can be understood from an attachment perspective
development of autonomy and self-confidence. (p. 142)
(see Mikulincer & Shaver, 2005, for a review). In situations in
which relationship partners behave negatively, for example, indi-
Extending these ideas, Mikulincer and Shaver (2003) have pro-
viduals who are more securely attached should experience func-
posed a process model that outlines the conditions under which the
tional anger, which ought to facilitate their more constructive,
attachment system should be activated and terminated in individ-
relationship-enhancing goals. More insecurely attached individu-
uals who are securely attached. When potential threats are per-
als, on the other hand, should experience dysfunctional anger,
ceived, secure individuals should remain confident that current
which ought to promote either security-oriented goals (in the case
attachment figures will be attentive, responsive, and available to
of people who have anxious attachment histories) or control/
meet their needs and mitigate their distress. These beliefs should
autonomy-focused goals (in the case of those who have avoidant
increase their feelings of security, deactivating their attachment
histories).
systems and allowing secure individuals to use constructive,
problem-focused coping strategies.
Negative Partner Behaviors
Insecurely attached individuals, on the other hand, should be
more likely to experience attachment system activation, motivating
Studies investigating the emotional impact of negative partner
them to adopt interpersonal self-focused strategies to compensate
behaviors have found that individuals classified as secure on the
for uncertainty about their partners’ responses. Mikulincer and
Adult Attachment Interview (AAI; Main & Goldwyn, 1998) be-
Shaver (2003) proposed differing strategies for individuals who
have more constructively when confronted with negative partner
manifest the subcategories of insecurity described by Ainsworth,
behavior than do those classified as insecure (Zimmermann,
Blehar, Waters, and Wall (1978). For example, when individuals
Maier, Winter, & Grossmann, 2001). Furthermore, adults classi-
who are anxiously attached perceive threats, they are likely to be
fied as dismissive on the AAI are rated by their friends as being
uncertain that their attachment figures will be sufficiently atten-
more emotionally hostile (Kobak & Sceery, 1988), and dismissive
tive, available, and responsive to their needs. These worries should
teens display more dysfunctional and inappropriate anger (rated by
sustain their anxiety and keep their attachment systems online,
observers) when trying to resolve chronic problems with their
leading anxious persons to adopt emotion-focused coping strate-
mothers (Kobak, Cole, Ferenz-Gillies, Fleming, & Gamble, 1993).
gies (e.g., remaining hypervigilant to signs of possible loss, rumi-
Studies that have assessed the two dimensions underlying self-
nating about worst-case scenarios). When individuals who are
reported adult romantic attachment measures (anxiety and avoid-
avoidantly attached feel threatened, they are likely to experience—
ance; see Brennan, Clark, & Shaver, 1998) have also found that
but perhaps not consciously acknowledge— distress and anxiety at
women who report being more avoidantly attached to romantic
a physiological level. To keep their attachment systems deacti-
partners display greater dysfunctional anger toward their partners
vated, these individuals would be expected to strive to inhibit and
when they (avoidant women) are more distressed in a fear-
control their emotions by deploying avoidant coping strategies.
inducing situation and receive less support from their partners as
Each mode of coping should also be associated with unique
rated by observers (Rholes, Simpson, & Or?˜na, 1999). Highly
interpersonal goals (Mikulincer & Shaver, 2005). Securely at-
anxious women also display more dysfunctional anger but only
tached individuals, for example, should focus on building greater
after their distress has subsided (during a poststress recovery
intimacy with their attachment figures. Individuals who are inse-
period) and only if their partners behaved less supportively when
curely attached should cope differently, consistent with the partic-
the women were distressed (Rholes et al., 1999). Additional re-
ular form of insecurity they manifest. Those who are anxiously
search has revealed that anxiously attached persons typically re-

---

EMOTION IN RELATIONSHIPS
357
port a flood of negative feelings when their partners behave badly
vestigated the ways in which early attachment experiences are
toward them, experience intrusive tangential memories and per-
prospectively related to the quality and functioning of close rela-
ceptions, ruminate about and amplify their negative feelings, and
tionships in early adulthood.
subsequently feel worse about their partners and relationships
Bowlby’s fundamental hypothesis that internal working models
(Mikulincer, 1998). Highly anxious individuals also report and
(representations) of earlier relationship experiences should affect
display greater dysfunctional anger, hostility, and distress when
later relationship experiences has been very influential. Recent
they are trying to resolve major— but not minor—relationship-
findings, however, suggest that representations of early relation-
based problems with their romantic partners (Simpson, Rholes, &
ship experiences do not necessarily predict subsequent relationship
Phillips, 1996).
outcomes in a simple or straightforward manner. Instead, repre-
sentations tend to be modified continuously as individuals enter
Positive Partner Behaviors
and leave different types of attachment relationships across suc-
cessive periods of development (see Carlson, Sroufe, & Egeland,
When relationship partners behave positively, individuals who
2004). Recent studies, for example, indicate that relationship ex-
are securely attached should experience an assortment of positive
periences with early peers following infancy contribute signifi-
emotions, given that positive behaviors often signal availability,
cantly to the quality of close friendships in adolescence. Further-
responsiveness, support, or validation. Insecurely attached individ-
more, the quality of experiences with caregivers in infancy and
uals, by comparison, should experience less intense positive emo-
early childhood also predict the quality of adolescent friendships
tions— or perhaps even negative ones—in response to positive
over and above the contributions of more proximal (i.e., concur-
partner behaviors. Positive actions by partners might lead inse-
rent) experiences with same-age peers (see Sroufe, Egeland, &
curely attached individuals to feel as if they do not deserve, cannot
Carlson, 1999).
reciprocate, or might fail to meet their partner’s positive expecta-
With respect to adult relationships, individuals’ attachment his-
tions (in the case of anxiously attached persons) or to worry about
tories in infancy, assessed using the Strange Situation procedure
the loss of interpersonal control and autonomy (in the case of
(Ainsworth et al., 1978), also predict some aspects of their later
avoidantly attached persons).
behavior with romantic partners when individuals are in their early
Studies examining the emotional correlates of positive partner
20s (Collins & Van Dulmen, in press; Roisman et al., 2005).
behaviors have shown that persons classified as dismissive on the
Couple functioning has been assessed by both observer ratings of
AAI exhibit fewer signs of genuine positive emotions when ex-
videotaped couple interactions and self-reports from both romantic
posed to positive stimuli (Spangler & Zimmermann, 1999). In
partners (i.e., the longitudinal target person and his or her current
parallel fashion, individuals who report being more anxiously or
romantic partner). Studies have revealed that if the longitudinal
more avoidantly attached to romantic partners also display fewer
target person was classified as having had a disorganized pattern of
positive emotions during both important events (e.g., college en-
attachment during early infancy (Main & Solomon, 1990), his or
trance interviews; Horppu & Ikonen-Varila, 2001) and mundane
her interaction during conflict resolution with the current romantic
events (e.g., watching a pleasant film; Magai, Hunziker, Mesias, &
partner in early adulthood was rated by observers as containing
Culver, 2000). These individuals also report feeling fewer positive
fewer secure base behaviors, less balance between couple func-
emotions in group interactions (Rom & Mikulincer, 2003) and
tioning and each partner’s personal interests or needs, less caring,
report experiencing fewer positive emotions when interacting with
less trust, less emotional closeness, less sensitivity to one another’s
different types of partners in daily diary studies (e.g., Pietromo-
needs and wishes, and poorer overall outcomes than interactions
naco & Feldman Barrett, 1997; Tidwell, Reis, & Shaver, 1996).
involving people who were secure during infancy. In addition, if
Finally, individuals who report being more avoidantly attached
the target person was disorganized during infancy, the couple was
also indicate that they are less likely to feel gratitude when their
rated as displaying comparatively greater hostility during their
partners behave positively toward them, whereas those who report
conflict resolution interactions than a couple whose attachments
being more anxiously attached feel mixed or conflicting emotions
during infancy were secure.
when their partners act positively toward them (Mikulincer,
Attachment insecurity assessed during infancy and early child-
Shaver, & Slav, 2006).
hood also forecasts other relationship outcomes across develop-
ment, including the target’s social competence with peers between
Longitudinal–Developmental Perspectives on Emotion in
the ages of 6 and 8 as rated by classroom teachers (e.g., Sroufe et
Relationships
al., 1999) and the target’s interactions with his or her parents at age
13 as rated by trained observers (Sroufe et al., 2005). Further
Bowlby (1980) proposed that emotional reactions to relationship
evidence documents the relation between these chronologically
events are partially rooted in earlier relationship experiences, first
later measures of family interaction and subsequent romantic re-
with caregivers and then with other significant partners across
lationship behaviors and perceptions in early adulthood, but nei-
adolescence and adulthood (also see Ainsworth, 1989; Waters &
ther the role of parent– child relationships prior to age 13 nor the
Cummings, 2000). This core tenet of attachment theory has in-
implications of these relations across time for self-reported emo-
spired longitudinal studies in which the same individuals have
tional experience in romantic relationships have been considered
been studied repeatedly across time from infancy onward (see, for
(Roisman, Madsen, Hennighausen, Sroufe, & Collins, 2001).
example, Grossmann, Grossmann, & Waters, 2005; Sroufe, Ege-
Studies of associations between early attachment security and
land, Carlson, & Collins, 2005). Though differing in which spe-
behavior in later romantic relationships also suggest that chrono-
cific features of attachment were examined, all these studies in-
logically later measures of nonfamilial relationships (e.g., teacher

---

358
SIMPSON, COLLINS, TRAN, AND HAYDON
ratings of peer competence in early elementary school, degree of
early childhood should be rated as more socially competent by
security expressed about relationships with close friends in ado-
their grade school teachers than those who were classified as
lescence) often mediate connections between early infant–
insecure. Early social competence, in turn, should predict stronger
caregiver relationships and behavior in later romantic relation-
rated secure-base friendships during adolescence, and friendship
ships. Many mediation effects, however, are partial, with the
security then should predict the experience and expression of less
impact of early attachment measures remaining independent and
negative versus positive emotion in adult romantic relationships.
significant predictors of later developmental outcomes (Sroufe et
We also tested several competing alternative models.
al., 2005). Whether these patterns apply to the prediction of self-
This developmental model is based on the premise that relation-
reported emotions in adult relationships is not known, however,
ships at any stage of development are influenced by both familial
because previous longitudinal studies have not examined reports of
and extrafamilial relationships experienced at earlier stages (see
emotions in romantic relationships.
Sroufe et al., 2005). As a result, attachment relationships with
Although past research has emphasized the attributes of indi-
caregivers early in life should have an impact not only on later
viduals, concurrent relationship conditions, or interaction dynam-
relationships with caregivers but also on other important relation-
ics as determinants of the experience and expression of emotions
ships with peers, close friends, and romantic partners. Carlson et
in romantic relationships (see Mikulincer & Shaver, 2005), the
al. (2004), predicting observer ratings of individual global com-
longitudinal findings reviewed above suggest that the experience
petence at age 19, showed that this type of developmental process
and expression of emotions in romantic relationships in adulthood
involves dynamic interactions between experiences in one’s suc-
might reflect vestiges of important relationships experienced dur-
cessive relationships and the mental representations of those ex-
ing earlier periods of development. This fundamental developmen-
periences, which are constructed and revised across relationships
tal hypothesis of attachment theory was tested in the current study.
from each successive earlier period. The age-19 competence mea-
sure, however, is not a measure of emotional experience in roman-
The Current Study
tic relationships. In the few existing studies of emotions associated
with romantic relationships, predictors have been parent– child
The current study was based on longitudinal data from 78 target
variables either from assessments made at age 13 but not earlier
participants who have been studied continuously from infancy into
(Roisman et al., 2001) or from the quality of caregiving in infancy
their mid-20s as part of the Minnesota Study of Risk and Adap-
and early childhood but not measures of parent or peer relation-
tation from Birth to Adulthood (see Sroufe et al., 2005). Between
ships between infancy and early adulthood (Roisman et al., 2005).
the ages of 20 and 23, each target participant and his or her current
Determining whether both early relationships with parents and
romantic partner completed a battery of self-reported relationship
later relationships with nonromantic peers contribute to under-
measures. Each couple was also videotaped while trying to resolve
standing the experience and expression of emotions in adult ro-
a conflict in their relationship and while completing a collaborative
mantic relationships requires that these specific variables be tested
task. Our primary goal was to test whether and how attachment
and modeled.
experiences and relationships encountered during critical stages of
The measurement approach used in the current study is consis-
development (i.e., in infancy, early childhood, and adolescence)
tent with this conceptualization and with the principle of hetero-
are systematically related to the self-reported experience and the
typic continuity (see Caspi & Roberts, 2001; Rutter & Sroufe,
observer-rated expression of emotions with romantic partners in
2000). In particular, the infancy/early childhood measures ob-
early adulthood. In this initial attempt to examine mediation pro-
tained from target participants assessed their attachment and ex-
cesses between early attachment history and emotions in adult
ploratory behaviors with their caregivers in the Strange Situation at
romantic relationships, we adopted the prevailing practice of test-
the age of 12 months. The middle childhood measures assessed
ing the predicted differences between secure versus insecure at-
target participants’ competence between the ages of 6 and 8 at
tachment histories. Thus, when assessing the infant attachment
engaging peers in social interactions and their attunement to in-
security of target participants, subclassifications of insecurity in
terpersonal dynamics in organized peer groups in Grades 1–3. The
the Strange Situation were collapsed into a single insecure cate-
gory.
adolescence measure assessed the nature and quality of target
On the basis of theory and prior empirical findings, we hypoth-
participants’ behaviors at age 16 that were indicative of having
esized a double-mediation developmental model. According to this
secure attachment representations (e.g., greater disclosure, trust,
model, the emotional qualities of romantic relationships in early
and authenticity) with close friends. The early adulthood measures
adulthood should be predicted by a set of sequential links from the
indexed the experience and expression of emotions evident in
attachment security status in infancy and early childhood, to the
target participants’ romantic relationships when they were between
quality of peer relationships in childhood, to the quality of rela-
the ages 20 and 23. Although target participants’ behaviors, rela-
tionships with close friends in adolescence. We predicted that the
tionships, and relationship representations were assessed using
quality of childhood peer relationships and the quality of close
different measures in different relationships at different points
friendships in adolescence should mediate the link between early
during social development, the underlying meaning and function
attachment status (assessed using the Strange Situation procedure
of those behaviors and representations should have been consistent
when the targets were 12 months old) and the emotional tenor of
across time because the measures were designed to tap the general
adult romantic relationships (assessed at age 20 –23). More spe-
coherence of attachment representations and behaviors at each
cifically, individuals who were classified as secure in infancy and
developmental stage.

---

EMOTION IN RELATIONSHIPS
359
Method
pleted romantic relationship assessments in early adulthood, 61%
had been classified as secure and 39% had been classified as
Participants
insecure at 12 months old.
The present data were collected as part of the Minnesota Study
Peer competence.
Peer competence was assessed in Grades 1,
of Risk and Adaptation from Birth to Adulthood (see Sroufe et al.,
2, and 3. Each target participant’s classroom teacher was given a
2005). This prospective longitudinal study of at-risk children and
one-paragraph description of a hypothetical child who was well
their families began in 1976, when 267 women were recruited
liked and respected by peers, had mutual friendships, demonstrated
from Minneapolis public health clinics where they were receiving
understanding of other children’s perspectives and ideas, and con-
prenatal care. Although most (58%) of the child participants are
structively engaged peers in activities. The teacher then rank-
European American, 14% are African American, and 3% are
ordered all children in the classroom according to how closely
Native American or Latino. A portion of the sample (16%) is of
each student matched this criterion. Teachers were unaware of
mixed racial background, and 9% are unclassifiable because of
which child was the target child. Peer competence scores thus
missing data on their fathers’ race. The original sample was 55%
represent teachers’ perceptions of each target participant’s percen-
male and 45% female. Approximately 180 of the original partici-
tile rank in his or her class during Grades 1–3 divided by the total
pants are still participating in data collection 28 years after the
number of students in each respective class. Accordingly, each
study began.
target participant received a mean peer competence percentile
ranking relative to his or her classmates averaged across Grades 1,
2, and 3. Because each target participant’s score came from a
Procedure and Early Developmental Measures
different teacher, concordance was computed between scores in
The present analyses focused on a subset of the original sample:
each of the three consecutive years. The Pearson correlation was
those individuals who participated in the romantic relationship
.50, p
.01, between the participants’ scores in Grades 1 and 2
assessments in early adulthood (N
78). Target participants who
and was .45, p
.01, between the participants’ scores in Grades 2
were involved in a romantic relationship that had existed for at
and 3.
least 4 months participated with their partners in this assessment
Friendship security.
Each target participant’s level of friend-
phase when most target participants were between the ages of 20
ship security at age 16 was rated from a comprehensive interview.
and 23. The mean age of participants was 21.60 years (SD
3.75).
This measure was developed from the premise that attachment
The mean length of relationships was 25.06 months (SD
17.04).
security in later relationships should be facilitated by security in
All 78 couples were heterosexual.
earlier relationships (see Ainsworth, 1989; Bowlby, 1969; Thomp-
The subsample that participated in the couples assessments did
son, 1999; Waters & Cummings, 2000). Target participants were
not differ from the original sample on socioeconomic status mea-
asked to describe their close friendships, including whether and
sured prenatally, in middle childhood, or at age 16. The proportion
how they disclosed behaviors and feelings indicative of trust and
of subsample participants of African American and mixed racial
authenticity within their close friendships. The questions on which
backgrounds did not differ from that of the original sample;
ratings were based included the degree to which the adolescent felt
however, a slightly higher proportion of subsample participants
comfortable telling private things to close friends, how friends
was European American. Target participants and their partners were
responded to such disclosures, and whether the adolescent felt
first interviewed separately and then completed a battery of self-report
“close” to friends. Two trained coders then rated global friendship
measures that assessed the functioning of their relationship. Each
quality on a 7-point scale. The scale assessed the extent to which
couple then discussed and tried to resolve existing points of disagree-
target participants reported feeling that they could be themselves in
ment or contention in their relationship; they also completed a col-
their friendships, expected friends to be available and supportive,
laborative problem-solving task. These interactions were videotaped
and could mutually (jointly) share both positive and negative
and subsequently coded by trained observers (see below).
emotional and interpersonal experiences. The interrater reliability
In earlier phases of the Minnesota Study of Risk and Adaptation
(intraclass correlation) of this scale was .59; the Spearman–Brown
from Birth to Adulthood, several measures were collected at three
correction was .74.
different stages of participants’ social development: during very
early childhood, early elementary school, and adolescence. Assess-
Contemporary Self-Report Measures
ments were conducted at these specific periods of social develop-
ment because each one represents a unique stage at which new and
At ages 20 –23, target participants and their partners of at least
different kinds of relationships are being formed and developed.
4 months each completed a battery of measures about their rela-
The earlier developmental measures relevant to the hypotheses of
tionship. The 4-month minimum criterion was adopted to increase
the current study include the following:
the likelihood that participants were involved in meaningful and
Infant attachment security.
The quality of parent–infant at-
reasonably well-established romantic relationships. The contem-
tachment relationships was assessed with the Strange Situation
porary measures most relevant to the current hypotheses were the
(see Ainsworth et al., 1978) when the target participants were 12
following:
months old. Certified raters classified infants’ attachment patterns
Emotional tone of the relationship.
The Emotional Tone Index
as secure, avoidant, or anxious/resistant. The present analyses used
(ETI; Berscheid, Snyder, & Omoto, 1989) is a 27-item scale on
the conventional secure versus insecure distinction, in which
which individuals report the extent to which they typically expe-
avoidant and anxious/resistant classifications were collapsed into
rience different emotions in their relationships. Each item is an-
one group. Of the subsample of original participants who com-
swered on a Likert-type scale, with higher scores reflecting the

---

360
SIMPSON, COLLINS, TRAN, AND HAYDON
more frequent experience of certain emotions. The ETI includes 12
vulnerability (Balance Scale 1), (b) individual growth in the con-
positive emotions and 15 negative emotions that vary in intensity
text of the relationship (Balance Scale 2), and (c) effective com-
from high (e.g., elated, angry) to low (e.g., content, disappointed).
pletion of the problem-solving task (Balance Scale 3). Interrater
It contains three subscales: (a) the extent to which individuals
reliabilities (intraclass correlations) for these scales ranged from
experience positive emotions (the sum of the 12 positive emotion
.82 to .96. All scales were coded at the dyadic level. Thus, the
items), (b) the extent to which individuals experience negative
affect scales assessed the extent to which each couple engaged in
emotions (the sum of the 15 negative emotion items), and (c) the
reciprocal exchanges of positive affect, negative affect, anger, and
relative balance of positive versus negative emotions (the mean of
hostility. Our two global relationship observation measures were
the positive emotion index
the mean of the negative emotion
based on two composite variables originally developed by Rois-
index). The relative balance scores reported by both target participants
man et al. (2001). The first measure, adult romantic relationship
and their romantic partners were the primary focus of the present
process, was a unit-weighted composite of the positive affect,
analyses. Unless otherwise stated, ETI index in this study refers to the
secure base, conflict resolution, and overall quality measures and
relative balance subscale. The internal consistency (Cronbach’s alpha)
Balance Scales 1 and 2. The second measure, adult romantic
for the positive emotion subscale was .87, and for the negative
relationship negative affect, was a unit-weighted composite of the
emotion subscale, it was .86. Within participants, positive emotion
anger, hostility, and dyadic negative affect measures. The reliabili-
scores were negatively related to negative emotion scores (r
.46,
ties of the relationship process and negative affect composites
p
.001). Target participants’ relative balance of positive versus
were .95 and .91, respectively.
negative emotion scores was positively correlated with their partners’
We also calculated two additional measures: the ETI relative
balance scores (r
.31, p
.01).
balance score of each participant, in which his or her partner’s ETI
Contemporary relationship observation measures.
Couples
score was statistically controlled, and a composite (z-scored) index
also completed a videotaped observational procedure in our labo-
composed of observer-rated adult relationship process scores,
ratory that consisted of two interaction tasks: the Markman–Cox
observer-rated negative affect scores, and both partners’ self-
procedure and the Ideal Couple Q-sort. The Markman–Cox pro-
reported ETI relative balance scores.
cedure (Cox, 1991) is designed to elicit conflict between relation-
ship partners. In the first phase of the procedure, each partner
Results
completed a relationship problem inventory privately to identify
and rate the most salient problems in the relationship. Each couple
Preliminary Analyses
then reviewed their inventories together and chose the problem
that caused the most conflict in their relationship. In the second
As discussed above, the variables used in the present analyses
phase (which lasted 10 min), each couple was instructed to discuss
were constructed to broadly assess important relationship experi-
the problem and to attempt to reach a solution. During a “cool
ences at critical points of social development. Thus, we first
down” phase (which lasted 4 min), each couple then discussed the
examined correlations between the variables that formed each
areas on which they agreed the most in their relationship.
multimeasure developmental construct: peer competence and rat-
Following this, each couple completed an Ideal Couple
ings of the emotions expressed during the videotaped conflict and
Q-sort (Collins et al., 1999), which is designed to elicit collab-
collaborative discussions with romantic partners at ages 20 –23.
orative behaviors. In this task, each couple was given 45 cards,
The peer competence composite measure contained ratings of
and each card listed a potential quality of a romantic couple
each target participant’s level of social competence reported by
(e.g., make sacrifices for each other, have the same interests).
three different teachers when the participant was in Grades 1, 2,
The qualities included some items from the Dyadic Relation-
and 3. The correlations among teacher ratings for each grade
ship Q-Sort (Bengtson & Grotevant, 1999). Each couple was
ranged from .45 to .53.
instructed to read each card aloud and decide together into
With regard to the observer-rated scales, the adult romantic
which of three labeled baskets the card should be placed: a
relationship negative affect measure was composed of three vari-
basket labeled “Most like an ideal couple,” one labeled “Least
ables: anger, hostility, and negative affect. The correlations among
like an ideal couple,” or one labeled “Middle/unsure.” Couples
these variables ranged from .77 to .86. The adult romantic rela-
were told to base their decisions on their ideas of an ideal
tionship process measure involved six variables: Balance Scale
couple rather than on their own relationship. After sorting all of
I— openness versus concealment, Balance Scale II— development
the cards, each couple narrowed their sort by selecting 7 cards
of the relationship versus development of the individual, quality of
from the “least ideal” basket (i.e., those that least described an
conflict resolution, overall quality of relationship, secure base, and
ideal couple) and 7 cards from the “most ideal” basket (i.e.,
shared positive affect. The correlations among these variables
those that best described the ideal couple).
ranged from .60 to .94.
Seven trained observers then rated all of the interactions (both
the videotaped conflict discussion and the Ideal Couple Q-Sort
Descriptive Statistics and Correlations
discussion) on dyadic scales that assessed the amount of shared
positive affect, shared negative affect, anger, hostility, conflict
In Table 1, means and standard deviations are presented for all
resolution, secure base behavior, and overall quality (Collins et al.,
of the major variables included in the structural models testing the
1999; also see Sroufe et al., 2005). Ratings were also made on
main hypotheses. Table 2 reports zero-order correlations between
three “balance scales” that indexed the extent to which the partners
all of the major variables. It is important to recognize that these
in each relationship facilitated (a) acceptance of openness and
variables span almost 20 years and four critical points in the social and

---

EMOTION IN RELATIONSHIPS
361
Table 1
but those interactions were also rated as indicating poorer
Descriptive Statistics for the Major Variables
relationship process (i.e., partners were lower in mutual respon-
siveness and were more likely to express hostility).
Variable
M
SD
Infant attachment security
1.20
0.40
Primary Analyses
Mean peer competence—Grades 1–3
46.84
22.10
Friendship security at age 16
4.39
1.50
Romantic relationship process
3.97
1.25
In the primary analyses, we tested the hypothesized structural
Romantic relationship negative affect
2.01
1.21
relations between the antecedent measures and the nature and
ETI
3.33
1.37
quality of emotions experienced in adult romantic relationships
ETI (controlling for partner’s ETI)
0.00
0.99
in early adulthood (both self-reported and observer-rated). We
Composite index
0.00
1.00
used structural equation modeling (SEM) to test our hypothe-
Note.
For each variable except the ETI, N
78 observations. ETI
sized mediation models. SEM is a widely used data analytic
Emotional Tone Index.
approach that is well suited for testing the type of mediation
effects that we predicted (see Loehlin, 2004). We predicted that
interpersonal development of target participants, ranging from the
the association between measures of infant attachment security
quality of their attachment patterns at 12 months, to their social
and measures of emotion in adult romantic relationships would
competence with peers in early elementary school, to their degree of
be mediated by two intervening constructs—peer competence in
security with close friends in adolescence, to their emotional experi-
early elementary school and the degree of security with close
ences with adult romantic partners between the ages of 20 and 23
friends at age 16. The tested models are described below.
years.
Given our sample size (N
78 couples), it was impractical
As expected, infant attachment, peer competence in childhood,
to construct and conduct tests of measurement models for each
and friendship security in adolescence all correlated positively and
construct. Accordingly, the manifest composite variables were
sequentially, indicating that target participants who were secure at
tested. We tested our hypothesized structural model for each of
the beginning of their lives were rated as more socially competent
the three primary dependent variables: observer-rated adult
in grade school and, in turn, as having more secure friendship
romantic relationship process scores from the videotaped dis-
representations at age 16 than those participants who were insecure
cussions, observer-rated adult romantic relationship negative
at the beginning of their lives.
affect scores from the videotaped discussions, and both part-
Peer competence was significantly correlated with three de-
ners’ self-reported emotion balance scores on the ETI. Addi-
pendent measures. Target participants evaluated as less socially
tionally, a structural model was tested in which the three
competent by their teachers during elementary school had in-
dependent variables were aggregated into one composite depen-
teractions with their romantic partners in adulthood that con-
tained greater negative affect. In addition, they reported expe-
dent variable.
riencing relatively more negative than positive emotions in their
Structural Model 1: Predictors of adult romantic relationship
relationships on a daily basis, even when their partner’s ETI
process.
Model 1 tested whether the link between infant attach-
scores were partialed out. The same pattern emerged for secu-
ment security and the adult romantic relationship process index
rity of friendships at age 16, with one additional finding. Not
was mediated through the measures of peer competence in ele-
only did target participants who had less secure friendship
mentary school and friendship security at age 16 (see Figure 1a).
representations report relatively more negative than positive
As predicted, this model fit the data very well, comparative fit
emotions in their daily relationships and have interactions in the
index (CFI)
1.0, root-mean-square error of approximation (RM-
laboratory that were rated as containing greater negative affect,
SEA)
.00, 2(2, N
78)
0.24, ns. Hence, the CFI, RMSEA,
Table 2
Correlations Between the Major Variables
Variable
1
2
3
4
5
6
7
8
1. Infant attachment (12 months)
—
2. Peer competence (Grades 1–3)
.29**
—
3. Friendship security (age 16)
.12
.37**
—
4. Romantic relationship process (ages 20–23)
.09
.13
.43***
—
5. Negative affect (ages 20–23)
.12
.27*
.37**
.75***
—
6. ETI scores (ages 20–23)
.23*
.29**
.36**
.30**
.41***
—
7. ETI scores (ages 20–23; residualized score,
controlling for partner’s ETI score)
.19†
.25*
.32**
.19
.27*
.95***
—
8. Composite of relationship process, negative
affect, and ETI scores
.20†
.27*
.48***
.87***
.88***
.66***
.55***
—
Note.
ETI
Emotional Tone Index.
† p
.10.
* p
.05.
** p
.01.
*** p
.001.

---

362
SIMPSON, COLLINS, TRAN, AND HAYDON
78)
0.70, ns. Similar to the tests of Models 1 and 2, neither
analysis revealed a significant drop in either the results of the
chi-square test or the other fit indexes when the direct path from
infant attachment security to adult emotional tone was elimi-
nated from the models.
Structural Model 4: Predictors of the composite score for all
three dependent variables.
For the hypothesized double-
mediation model to be robust, evidence for it also should emerge
when the three dependent measures are aggregated. Thus, in
Model 4, we tested whether the association between the infant
Figure 1.
Structural Models 1– 4. Four tests of the structural model were
attachment index and the composite measure of all three dependent
conducted, one for each of the four dependent variables: a, adult romantic
variables—adult romantic relationship process, adult romantic re-
relationship process; b, adult romantic relationship negative affect; c, adult
lationship negative affect, and adult emotional tone—was medi-
romantic emotional tone; and d, the composite score for dependent vari-
***
ated through peer competence in elementary school and friendship
ables a, b, and c. †p
.10. *p
.05. **p
.01.
p
.001.
security at age 16 (see Figure 1d). As expected, this model also fit
the data well, CFI
1.0, RMSEA
.00, 2(2, N
78)
0.20,
ns. Again, neither the results of the chi-square test nor the other fit
and chi-square test all indicated good model fit.1 Neither the results of
indexes dropped significantly when the direct path from infant
the chi-square test nor the other fit indexes dropped significantly when
attachment security to the composite dependent variable was elim-
the direct path from infant attachment security to adult romantic
inated from the model.2
relationship process was eliminated from the model.
Structural Model 2: Predictors of adult romantic negative af-
fect.
Model 2 tested whether the association between infant at-
Tests of Alternative Models
tachment security and the adult romantic relationship negative
Two alternative models also were tested with each of the
affect measure was mediated through measures of peer compe-
dependent variables. First, we tested an alternative model in
tence in elementary school and friendship security at age 16 (see
which the direct path from infant attachment security to each
Figure 1b). As hypothesized, this model also fit the data well,
dependent measure was eliminated and replaced by a path from
CFI
1.0, RMSEA
.00,
2(2, N
78)
1.19, ns. Again,
peer competence to each dependent measure (see Figure 2). As
neither the results of the chi-square test nor the other fit indexes
expected, each of the paths from peer competence to the four
dropped significantly when the direct path from infant attachment
dependent variables—adult romantic relationship process, adult
security to adult romantic relationship negative affect was elimi-
romantic relationship negative affect, adult emotional tone, and
nated from the model.
the composite dependent variable—was nonsignificant. The fit
Structural Model 3: Predictors of adult romantic emotional
indexes remained good for all four tests of this alternative
tone.
Model 3 tested whether the link between infant attachment
model; CFIs
0.99 –1.0, RMSEAs
.00 –.04, and 2s(2, N
security and the ETI balance scores were mediated by measures of
78)
0.20 –2.31, ns, for all models. Akaike information crite-
peer competence in elementary school and friendship security at
rion (AIC) indexes for the tests of this first alternative model
age 16 (see Figure 1c). Unlike the dependent variables in the first
(which ranged from 16.20 to 18.27) were all slightly but not
two models, the ETI balance scale in Model 1c involved self-
appreciably larger than was true for the primary models re-
reports provided by both the target participant and his or her
ported above (which ranged from 16.20 to 17.18; see Loehlin,
romantic partner. If our hypotheses are correct, antecedent rela-
2004).
tionship experiences in an individual’s life should predict the
emotional tone (i.e., positive relative to negative emotions) of the
individual’s current romantic relationship, even when reports of
1 Previous research has shown that the adequacy of chi-square tests can
emotional tone provided by the partner are statistically controlled.
be questionable with small sample sizes (Herting, 1985). Given the sample
In other words, a stringent and precise test of our hypotheses
size in this study, the chi-square test may not be an adequate estimate of
should discount possible “partner effects.” To control for the
model fit. It should be noted, however, that the hypothesized model showed
partner’s influence on each target participant’s emotional tone
no significant declines in any of the fit indexes (including the chi-square
scores, we created a residualized variable in which the ETI balance
test) when the direct path from infant attachment security to each of the
scores reported by each partner were partialed out from each target
four dependent variables was eliminated from the model. Moreover, for all
participant’s ETI balance scores (see Cohen & Cohen, 1983, for
tests of the primary model, the robustness of fit across the different indexes
indicated that sample size was not a problem in this study.
the logic of creating and using residualized scores). This residu-
2
alized measure was then treated as the dependent measure in
We also performed five additional mediation tests examining portions
of our full model. Specifically, we examined whether the relation between
Structural Model 3.
infant attachment security and friendship security at age 16 was mediated
As predicted, the model fit the data well, CFI
.1.0, RM-
by peer competence in elementary school. We also tested whether the link
SEA
.00,
2(2, N
78)
0.59, ns. In addition, the same
between peer competence and each of the four dependent measures was
pattern emerged when we tested the target participant’s ETI
mediated by friendship security at age 16. In all five mediation analyses,
score as the dependent variable without partialing out his or her
the result of Sobel’s test was statistically significant (all zs
2.04, all ps
partner’s ETI score, CFI
1.0, RMSEA
.00,
2(2, N
.05), indicating at least partial mediation in each analysis.

---

EMOTION IN RELATIONSHIPS
363
premise of attachment theory. Assessing relationship experiences
Peer
Security at
competence
age 16
at four critical developmental stages, we found that both the
experience and expression of emotions in adult romantic relation-
ships were meaningfully linked to attachment-relevant experiences
Infant
Dependent variable:
earlier in social development. Specifically, the early attachment
attachment
a. Romantic process
security of target participants at 12 months of age predicted their
b. Negative affect
competence with peers (as rated by teachers) during early elemen-
c. Emotional tone
d. Composite index
tary school. Elementary school peer competence, in turn, predicted
the degree of security evident in target participants’ representa-
Figure 2.
Alternative structural models. Four tests of each alternative
tions of close friendships at age 16. This measure then predicted
model were conducted, one for each of the four dependent variables: a,
both daily reports of emotions experienced in romantic relation-
adult romantic relationship process; b, adult romantic relationship negative
ships (reported by both target participants and their partners) as
affect; c, adult romantic emotional tone; and d, the composite index for
well as the expression of emotions (as rated by observers) during
dependent variables a, b, and c. The first alternative model (depicted by the
videotaped interaction tasks. Support for this double-mediation
solid lines) tested the paths connecting infant attachment to peer compe-
model also remained when relationship partners’ self-reports of
tence in elementary school, peer competence in elementary school to
daily emotions in the relationship (i.e., partner effects) were sta-
friendship security at age 16, friendship security at age 16 to each depen-
tistically controlled. Thus, corroborating Bowlby’s conjectures,
dent variable, and peer competence in elementary school to each dependent
variable. The second alternative model tested the direct paths (depicted by
both the experience and expression of emotion in romantic rela-
the dashed lines) from each of the predictor variables to each dependent
tionships appear to be tied in significant and meaningful ways to
variable.
experiences rooted in earlier relationships and stages of social
development. However, the results also suggest that earlier devel-
opmental stages may have the strongest and most direct impact on
Next, we tested a second alternative model in which the medi-
the stages that immediately follow them.
ating paths were eliminated, and the direct paths from each pre-
We now discuss how the current findings extend attachment
dictor to a given dependent variable were included (see Figure 2).
theory, add to the developmental attachment literature, and expand
The direct path from infant attachment security to each of the
our understanding of the experience and expression of emotions in
primary dependent variables and the direct path from peer com-
adult romantic relationships. We then consider how the current
petence to each dependent measure were all nonsignificant. As
results might be understood within the Emotion in Relationships
expected, the four tests of this second alternative model revealed
Model (ERM; Berscheid & Ammazzalorso, 2001). We conclude
significant declines in goodness of fit, CFIs
.39 –.55, RMSEA
by highlighting limitations and important caveats of the current
.29 for all models, and
2(3, N
78)
20.62, for all models.
research.
Furthermore, AIC indexes of the primary models revealed consid-
erably better fit than was true of the second alternative model. AIC
Theoretical and Empirical Extensions
indexes for the second alternative model were all higher than
34.00, whereas those for the primary models were all significantly
Bowlby (1980) believed that life’s deepest and most intense
lower (16.20 –17.18).3
emotions arise in the context of attachment relationships. Indeed,
In sum, the predicted double-mediation model provided a stron-
one of the principal functions of the attachment system is to
ger or a more parsimonious fit than the two alternative models for
regulate negative affect, especially when individuals are ill, fa-
each of the three dependent measures. Three additional consider-
tigued, afraid, overly challenged, or in pain. Bowlby also believed
ations also highlight the robustness of the hypothesized double-
that experiences in and representations of attachment-based rela-
mediation model effects. First, the effects were very consistent
tionships from earlier periods of social development leave residual,
despite the fact that the two observer ratings of expressed emotion
lingering effects on attachment-based relationships later in life.
(adult romantic relationship process and negative affect measures)
Borrowing from Waddington’s (1957) epigenetic landscape
shared no method variance with the ETI self-reports. Second, all of
model, Bowlby (1973) likened social development as similar to a
the effects remained consistent even when variance associated
railway system in which individuals set out on a single develop-
with the partner’s self-reports of emotions experienced in the
mental track early in life and then encounter multiple branch points
relationship were partialed out from each target participant’s own
self-reports. Third, the double-mediation pattern also emerged
3
when the three primary dependent measures were aggregated into
For comparisons involving the two alternative models, we were unable
to conduct chi-square difference tests because of lack of nesting. Instead,
a single composite index. Viewed as a whole, these results pro-
we used the Akaike information criterion (AIC). There is no absolute AIC
vided fairly clear and compelling evidence for the double-
level that indicates a superior fitting model; rather, the AIC is used to
mediation model.
compare models. As a rule of thumb, an AIC difference greater than 2
suggests that the models being compared are not identical in fit, but a
Discussion
difference of 10 or more is viewed as more trustworthy. As expected, our
hypothesized double-mediation model had a slightly smaller AIC differ-
Bowlby (1979) proposed that attachment relationships contrib-
ence than did the first alternative model, indicating that these two models
ute to personality and social development from the cradle to the
had comparable fit. Our hypothesized model had a significantly smaller
grave. The results of this longitudinal study support this core
AIC than did the second alternative model, indicating a superior fit.

---

364
SIMPSON, COLLINS, TRAN, AND HAYDON
at critical stages of social development that can lead to different
The Findings in the Context of the ERM
outcomes in adulthood. According to Bowlby, the quality of the
caregiving environment figures prominently not only in determin-
Although the current study was not designed to test the ERM
ing which specific developmental track individuals take at critical
(see Berscheid & Ammazzalorso, 2001), one can envision how
junctures but also in sustaining movement down a particular de-
vestiges of an individual’s past relationships could shape the
velopmental track over time (see Fraley & Brumbaugh, 2004). The
experience and expression of emotions in his or her later romantic
current findings extend our understanding of critical attachment
relationships. According to the ERM, emotions are experienced in
processes not only by confirming that the quality of attachment-
relationships when expectations tied to important plans or goals
based re