You are on page 1of 14

PERSONALITY AND SOCIAL PSYCHOLOG Y BULLETIN

Cozzarelli et al. / MENTAL MODELS OF ATTACHMENT

General Versus Specific Mental Models of Attachment:


Are They Associated With Different Outcomes?

Catherine Cozzarelli
Kansas State University
Steven J. Hoekstra
Kansas Wesleyan University
Wayne H. Bylsma
American College of Physicians–
American Society of Internal Medicine

The present study examined the relationships between mental current romantic partner differ in the way they relate to
models of attachment and both overall psychological adjustment overall psychological adjustment and relationship-
and relationship-related outcomes. Mental models were assessed related outcomes.
in the context of a specific relationship as well as for people in Construct overlap is another issue of concern when
general. Overall, partner-specific mental models had stronger measuring or theorizing about attachment styles or men-
and more numerous associations with the outcome variables tal models. Theoretical conceptualizations of attach-
than did general mental models. General models (especially ment and recent research have both suggested that men-
model of self) were most strongly associated with measures of tal models of attachment may overlap significantly with
overall psychological adjustment. However, specific mental mod- global constructs such as self-esteem and interpersonal
els were much more strongly associated with relationship-specific trust. Therefore, a secondary purpose of this article is to
outcomes such as feelings of romantic love or relationship satis- examine the extent to which mental models of attach-
faction than were general mental models. Self-esteem was more ment, assessed at both the general level and with refer-
strongly related to global psychological adjustment than was ence to a specific relationship partner, are related to a
model of self, but specific model of self was more strongly related variety of outcome variables independently of their over-
to outcomes in relationships. Trust was not related to any of the lap with self-esteem and trust.
dependent variables in the study.
Brief Overview of Attachment Theory
The attachment theory approach to adult romantic
Imajor
n the past decade, attachment theory has become a relationships is rooted in the work of John Bowlby (1969,
1973, 1980), which detailed the origins, functioning,
perspective on adult romantic relationships. The
measurement of attachment styles, however, has con- and implications of an innate attachment system
tinually posed both empirical and conceptual chal- designed primarily to maintain the proximity of an
lenges. One issue that may have important implications infant to her or his caregiver. As interactions with a care-
for measurement and theory is the level at which mental giver proceed, infants develop stable experience-based
models of attachment are assessed. Attachment theorists beliefs about themselves as lovable or worthwhile (men-
are only recently beginning to explicitly acknowledge
that individuals may have multiple mental models of Authors’ Note: This study was supported in part by a grant from the Na-
attachment and that these mental models may differ in tional Institute of Mental Health (No. 5RO3MH53599-02) to the first
and third authors. Correspondence concerning this article may be
their level of specificity (e.g., they may refer to people in addressed to Catherine Cozzarelli, Department of Psychology, Kan-
general or to a specific romantic partner). The primary sas State University, 468 Bluemont Hall, Manhattan, KS 66506; e-mail:
purpose of our study is to examine whether attachment psycozz@ksu.edu.
representations assessed with regard to people in gen- PSPB, Vol. 26 No. 5, May 2000 605-618
eral and those assessed with reference to an individual’s © 2000 by the Society for Personality and Social Psychology, Inc.

605
606 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

tal model of self) and about the responsiveness and viduals may simultaneously hold multiple mental mod-
accessibility of others (mental model of others). A funda- els of attachment (e.g., Baldwin & Fehr, 1995; Collins &
mental tenet of attachment theory is that these mental Read, 1994; Shaver et al., 1996). For example, psycholo-
models influence thoughts, emotions, and behaviors gists have shown that an individuals’ attachment style
automatically and are self-perpetuating; they are the with one parent does not necessarily predict his or her
mechanisms by which an attachment style (or typical style with regard to the other parent (Baldwin, Keelan,
mode of responding in attachment-relevant situations) Fehr, Enns, & Koh-Rangarajoo, 1996; Bridges, Connell,
is thought to be maintained over time (Rothbard & & Belsky, 1988; Lamb, 1978; see Fox, Kimmerly, &
Shaver, 1994). Schafer, 1991, for a review) or his or her attachment style
Based on this assumption of continuity, Hazan and in relation to a romantic partner (Baldwin et al., 1996).
Shaver (1987) proposed that attachment processes simi- And, it is likely that the situation is even more complex
lar to those characterizing attachment to primary care- because adults may have different working models for
givers during childhood should govern an individual’s each of the different roles they occupy (e.g., son, lover,
thoughts, feelings, and behaviors in romantic relation- father, friend) (Bartholomew & Horowitz, 1991;
ships. Numerous studies have supported Hazan and Bretherton, Biringen, Ridgeway, Maslin, & Sherman,
Shaver’s initial proposition, showing that individuals’ 1989). Collins and Read (1994) have argued that an indi-
recollections of parental relationships are consistent vidual’s various mental models are structured as a hierar-
with their attachment style in adulthood and that attach- chical network of attachment representations, with gen-
ment styles are related to satisfaction and adjustment in eral models of self and others at the top of the hierarchy
romantic relationships (e.g., Collins & Read, 1990; Kirk- and more specific mental models at lower levels, whereas
patrick & Davis, 1994; Simpson, 1990), the overall posi- Baldwin et al. (1996) suggest that the structure of rela-
tivity of individuals’ self-views (e.g., Bartholomew & Hor- tionship knowledge may look more like a tangled web
owitz, 1991), disclosure in relationships (e.g., Feeney & than a hierarchy.
Noller, 1990; Mikulincer & Nachshon, 1991), caregiving Despite the plausibility of the existence of multiple
under stress (e.g., Kobak & Hazan, 1991; Simpson, mental models, measures of attachment styles are typi-
Rholes, & Nelligan, 1992), perceptions of social support cally presented and/or used without any explicit consid-
and social conflict (e.g., Cozzarelli, Sumer, & Major, eration of the specificity of the attachment object that is
1998), as well as many other variables (see Shaver, Col- referred to in the measure. A review of the literature
lins, & Clark, 1996). reveals that in practice, most research has assessed
In 1991, Bartholomew and Horowitz constructed a attachment styles in terms of general others or relation-
measure of attachment styles organized around the posi- ships and not in terms of specific relationship partners.
tivity/negativity of beliefs about the self (as worthy vs. In some cases, existing measures simply intermix refer-
unworthy of love) and others (as trustworthy and avail- ences to people in general and to more specific others,
able vs. unreliable and rejecting). When crossed, these usually romantic partners. For example, Hazan and
two dimensions produce the four attachment classifica- Shaver’s (1987) three-category measure specifically
tions assessed in Bartholomew and Horowitz’s measure: instructs respondents to describe the “way you generally
(a) the secure attachment style (positive models of both are in love relationships,” and the attachment style
self and others), which is characterized by a general com- descriptions themselves refer variously to “others,” “any-
fort with closeness and trust in others; (b) the preoccu- one,” “love partners,” and “my partner.” Bartholomew
pied attachment style (negative model of self combined and Horowitz’s (1991) popular four-category measure
with a positive model of others), which reflects a sense of contains references only to close “others” in general, and
unlovability but a belief that others are generally trust- Simpson et al. (1992) instruct respondents to complete
worthy and available; (c) the dismissing style (positive their measure with reference to “romantic partners in
model of self and negative model of others), character- general” (see also Pietromonaco & Feldman Barrett,
ized by a belief in one’s lovability but a general distrust of 1997). In addition, regardless of how attachment styles
others or the expectation that they will be unreliable or are measured, attachment researchers often use these
rejecting; and (d) the fearful style (negative self and styles or mental models to predict a wide variety of out-
other models), which is typified by the belief that one is comes, paying scant attention to the level of specificity
unlovable and that others are unreliable or rejecting. with which either the attachment or outcome variables
are assessed. Thus, attachment styles have been related
Multiple Attachment Representations
to relationship-specific behaviors and outcomes (e.g.,
Although researchers often discuss attachment as if seeking or providing social support, conflicts with a
individuals have only one attachment style, theoretical romantic partner, relationship satisfaction, commit-
accounts of attachment processes point out that indi- ment, relationship longevity) as well as to indices of gen-
Cozzarelli et al. / MENTAL MODELS OF ATTACHMENT 607

eral adjustment or adaptive functioning such as anxiety, ing the self-concept (for which self-esteem was a marker
depression, loneliness, eating disorders, alcohol prob- variable). In the realm of stress and coping, Cozzarelli et al.
lems, and physical symptoms (see Shaver & Hazan, 1993, (1998) demonstrated that controlling for self-esteem
for a review). substantially reduced the relationship between mental
We believe that it is important to distinguish the cor- model of self and psychological adjustment subsequent
relates and effects of general mental models about “peo- to an abortion. Similarly, conceptual definitions of men-
ple” from those of specific mental models about a tal model of others and of interpersonal trust (e.g., Rem-
romantic partner. Logically, mental models assessed at pel, Holmes, & Zanna, 1985) suggest that these variables
different levels of specificity are likely to be related to dif- should also overlap (see also Mikulincer, 1998). Empiri-
ferent outcomes (see also Shaver et al., 1996). A similar cal support for this proposition was found by Sumer, Coz-
conceptual point has been made repeatedly in the psy- zarelli, and Burns (1996), who showed that a latent vari-
chological literature on attitudes, where it has been dem- able representing model of other and a latent variable
onstrated that dependent variables will generally be best representing interpersonal trust were strongly related
predicted by independent variables measured at the (see also Lipschultz, Wall, Scotilla, Bylsma, & Cozzarelli,
same level of specificity (e.g., Brannon, 1976; Fishbein & 1995, for corroborating evidence).
Ajzen, 1975). Generally speaking, global predictors can Given this evidence, it is increasingly important to
be expected to relate to a wide array of general out- consider the degree to which mental models differ from
comes, whereas specific outcomes will be better pre- and have relationships with outcome variables that can-
dicted by variables assessed at the same level of specific- not be attributed to these long-standing, global con-
ity. Consistent with this argument, Davis, Morris, and structs. We believe that such overlap may be especially
Kraus (1998) found that global and relationship-specific likely when attachment styles or mental models are
social support best predicted outcomes that were assessed at the general level, as has frequently been the
assessed at the same level of specificity as the support case in the adult attachment literature. General beliefs
measure. about the availability or trustworthiness of other people
The same general logic should hold true for the rela- would logically bear a stronger resemblance to a disposi-
tionships between general and partner-specific mental tional measure of trust than would beliefs about the
models and global and relationship-specific outcomes. trustworthiness of another person in a specific context
To address this issue in the current study, we assessed the or relationship. If this reasoning is correct, attachment
relationships between both general and partner-specific styles or mental models would only have sizable, unique
mental models and a set of outcome variables, including relationships with outcome variables when they are
relationship-specific outcomes and global indices of psy- assessed at the more specific level; when assessed gener-
chological well-being. We hypothesized that working ally, these variables would have little to offer above and
models assessed with respect to a specific romantic part- beyond existing dispositional measures.
ner would be more strongly related to relationship- To address the issue of construct overlap, we exam-
relevant outcomes (e.g., relationship satisfaction, feel- ined the relationships between general and partner-
ings of romantic love) than would working models specific mental models, a set of relationship-specific and
assessed at a more general level. In contrast, we pre- global outcome variables, and trust and self-esteem. We
dicted that mental models about people in general expected to find that trust and self-esteem would be
would be more strongly related to overall psychological more highly correlated with general mental models than
functioning (e.g., psychological well-being, life satisfac- with partner-specific mental models. We also explored
tion, depression) than would mental models of one’s the relationships between general and partner-specific
romantic partner. mental models and the set of outcome variables, holding
trust and self-esteem constant. We expected to find sub-
Construct Overlap With Self-Esteem and Trust
stantial overlap between trust/self-esteem and general
Another issue that needs to be addressed more fully in mental models in relation to all of the outcome variables,
the attachment literature is the extent to which the but especially global outcomes such as life satisfaction or
apparent effects of attachment styles or mental models depression. In contrast, we did not expect to find that
are due to their overlap with related constructs. Concep- the relationships between partner-specific mental mod-
tual accounts of attachment styles suggest that model of els and any of the outcome variables would be much
self overlaps with global self-esteem, and empirical stud- attenuated when partialing out the effects of self-esteem
ies indicate that these two constructs are indeed related. and trust.
For example, Griffin and Bartholomew (1994) showed In summary, we made the following five predictions
that a latent variable representing model of self over- with respect to general and partner-specific mental mod-
lapped almost perfectly with a latent variable represent- els of attachment: (a) general mental models were ex-
608 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

pected to be more strongly related to global outcomes second session, global adjustment and relationship-
(e.g., overall psychological well-being, life satisfaction) specific outcomes were assessed, also in counterbal-
than were partner-specific mental models, (b) partner- anced order. These scales are described more fully
specific mental models were expected to be more below. Participants were tested in groups of 5 to 10 and
strongly related to relationship-specific outcomes (e.g., were debriefed after the second session.
relationship satisfaction, feelings of love) than were gen- Attachment measures. Two different forms of Bartholo-
eral mental models, (c) self-esteem and trust were mew and Horowitz’s (1991) Relationship Questionnaire
expected to be more highly correlated with general than (RQ) were used to access participants’ attachment styles.
partner-specific mental models, (d) holding self-esteem One of these referred to how participants feel about peo-
and trust constant was expected to substantially attenu- ple in general, whereas the second referred to how par-
ate the relationships between general mental models ticipants feel about their current romantic partner. The
and all of the outcome variables, and (e) holding self- RQ consists of short prototypical descriptions of each of
esteem and trust constant was not expected to substan- the four attachment styles—secure, dismissing, preoccu-
tially attenuate the relationships between partner- pied, and fearful. Participants are asked to indicate the
specific mental models and the outcome variables. In degree to which each prototype is descriptive of them on
testing these hypotheses, we used Bartholomew and 7-point rating scales ranging from 1 (this is not like me) to
Horowitz’s (1991) measure of attachment styles. Thus, 7 (this is very much like me). They then indicate which of
we implicitly adopted their working definition of model the four prototypes describes themselves best.
of self as the extent to which one views the self as worthy The first form of the scale was worded in the tradi-
of love and support and their definition of model of oth- tional way (as per Bartholomew & Horowitz, 1991; Grif-
ers as the extent to which people are seen as trustworthy fin & Bartholomew, 1994), referring to relationships
and available versus unreliable and rejecting. with people in general. This version will be hereafter
referred to as the general measure of attachment styles.
METHOD On this measure, 49% of the participants reported being
secure, 25% being dismissing, 8% being preoccupied,
Participants
and 18% being fearful. Other research (Bartholomew &
The participants were 112 undergraduates from Kan- Horowitz, 1991; Brennan & Bosson, 1998) has found
sas State University (24 men, 56 women) and Pennsylva- roughly equivalent numbers of secures but slightly fewer
nia State University (9 men, 23 women) who volunteered dismissings and more preoccupieds and fearfuls.
to complete a set of questionnaires to fulfill a course The second version of the RQ was worded so as to
requirement.1 The mean age of participants was 19.7 refer to the participant’s relationship with his or her cur-
years (SD = 3.4), with 87% of participants between 18 and rent romantic partner. For example, the descriptor for
21 years. All participants were required to be currently in secure was changed from “It is easy for me to become
a romantic relationship and to have been in that rela- emotionally close to others. I am comfortable depend-
tionship for at least 1 month. Participants reported ing on others and having others depend on me. I don’t
knowing their partner an average of 33 months (range = worry about being alone or having others accept me” to
1 month to 15 years) and to have been romantically “It is easy for me to become emotionally close to my part-
involved with their partner for an average of 20 months ner. I am comfortable depending on him or her and hav-
(range = 1 month to 10 years). Of the participants, 50% ing him or her depend on me. I don’t worry about being
were in the 1st year of their relationship and 95% of par- alone or having my partner not accept me.” This version
ticipants were in the first 5 years of their relationship. will be hereafter referred to as the relationship-specific
measure of attachment styles. When measured in this
Materials and Procedure
relationship-specific way, 55% of the participants
Participants were asked to complete a variety of scales, reported being secure, 20% being dismissing, 14%
including measures of attachment style, self-esteem, being preoccupied, and 11% being fearful. A chi-square
trust, global adjustment, and relationship-specific out- test of independence revealed no significant differences
come measures. Because of the questionnaire’s length, between the distributions that were obtained using the
testing occurred in two sessions, the second occurring global and relationship-specific attachment question-
approximately 2 weeks after the first. In the first session, naires, χ2(4) = 4.55, p > .05.
global and relationship-specific attachment styles were Mental models of attachment were calculated for
assessed, as was self-esteem and interpersonal trust. both the global and relationship-specific attachment
These measures were embedded within a counterbal- measures, following the procedure outlined by Griffin
anced set of attachment-related scales that were adminis- and Bartholomew (1994). Model of self was calculated
tered for purposes unrelated to the present study. In the by summing the ratings of the two attachment styles with
Cozzarelli et al. / MENTAL MODELS OF ATTACHMENT 609

a positive model of self (secure and dismissing) and sub- to which they agree with each statement on a rating scale
tracting the sum of the ratings on the two attachment ranging from 1 (not at all) to 9 (extremely). Responses to
styles with a negative model of self (preoccupied and the items comprising each subscale were averaged.
fearful). Likewise, model of other was calculated by sum- Because scores on the three subscales were highly inter-
ming the ratings of the two attachment styles with a posi- correlated (rs ranged from .84 to .90), we combined
tive model of other (secure and preoccupied) and sub- them to create a single romantic love index. The mean
tracting the sum of the ratings of the two attachment romantic love score in our sample was 7.38 (SD = 1.50)
styles with a negative model of other (dismissing and and the alpha for the combined scale was .98.
fearful).2 Using this procedure, mental model scores can
Psychological adjustment measures. A short form of the
range from –12, indicating a very negative mental
Beck Depression Inventory (BDI) (Beck, 1976) was used
model, to +12, indicating a very positive mental model.
to assess depression. The short form of the BDI is a 13-
Scores on the partner-specific mental models (M = 4.56,
item Guttman scale; for each item, participants are given
SD = 4.07, for model of self; M = 1.78, SD = 3.26, for model
a choice of four statements of decreasing intensity, and
of other) were slightly more positive than the respective
their score is based on the strongest statement they
scores on the general mental models (M = 1.96, SD =
endorse in each set. For example, a participant would
3.34, for model of self; M = .81, SD = 3.9, for model of
earn a score of 3 for selecting “I can’t do any work at all,”
other), although the two forms of both mental models
a 2 for “I have to push myself very hard to do anything,” a
were significantly correlated (general/specific models
1 for “It takes extra effort to get started at doing some-
of self, r = .41, p < .001; general/specific models of other,
thing,” and a 0 for “I can work about as well as before.”
r = .25, p < .01).
The mean total score on this scale (out of a possible score
Relationship-specific outcome measures. Relationship sat- of 36) was 3.5 (SD = 4.6), with a highly skewed distribu-
isfaction was assessed using a scale from Hendrick tion (Mdn = 2, mode = 0, range = 0 to 25, 84% of the total
(1988) that consists of seven statements, each rated on a scores were lower than 5), indicating that most partici-
7-point rating scale. Sample items include “How good is pants were not depressed. Cronbach’s alpha for the BDI
your relationship compared to most?” or “How many was .88.
problems are there in your relationship?” We obtained a We assessed life satisfaction using a measure by
mean of 5.7 in our sample (SD = 1.14), indicating rela- Diener, Emmons, Larson, and Griffin (1985). Partici-
tively high satisfaction. The reliability coefficient was .92. pants rated the extent of their agreement to five items
The Inclusion of Other in the Self (IOS) scale (Aron, (e.g., “I am satisfied with my life”) on scales ranging from
Aron, & Smollan, 1992) is a single-item, pictorial scale. 1 (strongly disagree) to 7 (strongly agree). The ratings were
There are seven Venn diagrams, each with two circles in
then averaged across items. The mean score for our sam-
varying degrees of overlap, ranging from totally independ-
ple was 5.0 (SD = 1.3), with a reliability coefficient of .88.
ent/nonoverlapping to mutually inclusive. One circle in
The Positive Well-Being scale is an abbreviated ver-
each pair is labeled “self,” and the other is labeled
sion of a scale by Ryff (1989; see Major, Zubek, Cooper,
“other.” The participant is instructed to select the pair of
Cozzarelli, & Richards, 1997, for more information on
circles that best describes his or her relationship with his
this scale). This 18-item measure assesses autonomy,
or her romantic partner. The mean on this measure was
environmental mastery, personal growth, positive rela-
5.3 (SD = 1.6) out of a possible 7, indicating that partici-
tions with others, purpose in life, and self-acceptance on
pants felt that their current romantic partner was a
7-point, Likert-type scales. Rather than analyzing
strong part of their self (the self and other circles were
subscales separately, participants’ ratings for all 18 items
approximately 50% to 60% overlapping). Aron et al.
were averaged. Participants were moderately high in
(1992) reported good alternate form (r = .95) and test-
positive well-being (M = 4.8, SD = .57, range = 3.6-6.0).
retest reliability (r = .85) for this measure. Perhaps
The reliability coefficient for this measure was .83.
because of its unusual format, only 83 of our participants
correctly completed this measure. Accordingly, all analy- Mental model covariates. Self-esteem was measured
ses involving the IOS scale are based on a reduced n. using the 10-item Rosenberg Self-Esteem Inventory
Sternberg’s (1988) Triangular Love (TL) scale was (Rosenberg, 1965). This measure consists of items such
used to measure characteristics of participants’ romantic as, “On the whole, I am satisfied with myself,” rated on
relationships. The TL scale consists of three, 15-item scales ranging from 1 (strongly disagree) to 7 (strongly
subscales: Intimacy (e.g., “I receive considerable emo- agree). Items are reverse-scored as necessar y and
tional support from X”), Passion (e.g., “I especially like responses across the 10 items are then averaged. Respon-
physical contact with X”), and Commitment (e.g., “I dents generally reported high self-esteem, with a mean
have confidence in the stability of my relationship with of 5.6 (SD = 1.1), and the scale showed respectable reli-
X”). To complete the scale, participants rate the degree ability (α = .88).
610 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

TABLE 1: Correlation Matrix

Scale 1 2 3 4 5 6 7 8 9 10 11 12

1. General model
of self .— .41*** .07 –.06 .13 .16 –.11 .28** .25** .20* –.02 .13
2. Specific model
of self .— .11 .04 .20* .11 –.19 .52*** .41*** .46*** .44*** .35***
3. General model
of other .— .25** .12 .31** –.18 .11 .12 .01 .10 .04
4. Specific model
of other .— .07 –.03 –.08 .08 .05 .23 .33** .31***
5. Self-esteem .— .18 –.48*** .47*** .45*** .25** .25* .23*
6. Trust .— –.19* .11 .17 .06 .02 .02
7. Depression .— .52*** –.61*** –.29** –.27* –.28**
8. Well-being .— .61*** .49*** .40*** .46***
9. Life satisfaction .— .46*** .38*** .36***
10. Relationship
satisfaction .— .79*** .87***
11. IOS .— .80***
12. Romantic love .—

NOTE: IOS = Inclusion of Other in the Self scale. N for most of the correlations ranges from 105 to 112; however, correlations with the IOS scale
have an N that ranges from 74 to 83.
*p < .05. **p < .01. ***p < .001.

Rosenberg’s Faith in People scale (Rosenberg, 1957) ent variables in this study, sex and relationship duration
was used to assess interpersonal trust. The Faith in Peo- were not included in the analyses that follow.3
ple scale consists of five items asking about the nature of Hierarchical regression analyses were conducted to
humankind (e.g., “If you don’t watch yourself, people determine the degree to which general and partner-
will take advantage of you”). Participants responded to specific mental models of attachment were related to the
each item on a 7-point scale. The scale ranges from 1 relationship-specific and global psychological adjust-
(strongly agree) to 7 (strongly disagree) for three of the ment outcomes. Two hierarchical regression analyses
items; two items have endpoints that correspond more were conducted for each of the outcome variables. In the
directly to the text of the item. For example, participants first of these analyses, general models of self and others
are asked to respond to the question, “Some people say were entered as predictor variables in the first step of the
that most people can be trusted. Others say you can’t be regression, partner-specific models of self and other
too careful in your dealings with people. How do you were entered as predictors in the second step, and the
feel about it?” on a scale ranging from 1 (most people can interactions between the general mental models and
be trusted) to 7 (you can’t be too careful). Responses to the between the specific mental models were entered third
five items were averaged. Responses tended to be mod- (see Tables 2 and 3). In the second regression analysis
erate rather than extreme (M = 3.78, SD = 1.0). The for each outcome variable, the order in which the men-
scale’s reliability coefficient was somewhat low but
tal model variables were entered was reversed: partner-
acceptable (α = .73).
specific models of self and other were entered first, gen-
eral models of self and others were entered second, and
RESULTS the interactions were again entered third (see Tables 2
A correlation matrix including all the variables in the and 3). The first step of these regression analyses allowed
study is presented in Table 1. As we expected, general us to draw conclusions about the relationships between
and partner-specific mental models were modestly cor- the specific and general mental models and the outcome
related for both model of self and model of other variables when only one type of mental model (general
(although the magnitude of this correlation was some- or specific) was used as a predictor variable. The second
what higher for model of self, Fisher’s z = 1.89, p = .058). step allowed us to examine the unique or individual rela-
Participants’ sex and the amount of time they had tionships between the two types of mental models and
been in their current relationship were evaluated as the outcome variables while accounting for the shared
potential covariates. Because neither variable systemati- variance between the general and specific mental mod-
cally covaried with attachment style or any of the depend- els. To clarify interpretation of the results, scores on the
Cozzarelli et al. / MENTAL MODELS OF ATTACHMENT 611

TABLE 2: Hierarchical Regression Analyses Predicting Global Psy- TABLE 3: Hierarchical Regression Analyses Predicting Relation-
chological Adjustment From Partner-Specific and Gen- ship Specific Outcomes From Partner-Specific and Gen-
eral Mental Models of Self and Other eral Mental Models of Self and Other

Well- Life Relationship Romantic


a
Beingb Satisfactionc a
Measure Depression Measure Satisfaction IOSb Lovec

Series 1 Series 1
Step 1 Step 1
General model of self –.09 .27* .24 General model of self .16 –.01 .09
General model of other –.17 .12 .11 General model of other .06 .10 .10
2 2
R change .04 .09** .07* R change .03 .01 .02
Step 2 Step 2
Specific model of self –.15 .49*** .36*** Specific model of self .46*** .55*** .35***
Specific model of other –.04 .09 .03 Specific model of other .22* .30** .30**
2 2
R change .02 .20*** .11** R change .23*** .33*** .19***
Step 3 Step 3
General Self × Other interaction .07 .00 –.04 General Self × Other interaction –.04 .02 .02
Specific Self × Other interaction .16 .05 .02 Specific Self × Other interaction .21* .01 .17
2 2
R change .03 .00 .00 R change .03† .00 .02
2 2
Total R .09 .29*** .18** Total R .29*** .34*** .24***
Series 2 Series 2
Step 1 Step 1
Specific model of self –.18 .52*** .40*** Specific model of self .46*** .44*** .34***
Specific model of other –.07 .10 .04 Specific model of other .21* .32*** .30***
2 2
R change .04 .28*** .17** R change .26*** .30*** .21***
Step 2 Step 2
General model of self –.04 .07 .09 General model of self –.01 –.24* –.04
General model of other –.15 .04 .08 General model of other –.04 .00 .01
2 2
R change .02 .01 .01 R change .00 .04† .00
Step 3 Step 3
General Self × Other interaction .07 .00 –.04 General Self × Other interaction –.04 .02 –.02
Specific Self × Other interaction .16 .05 .02 Specific Self × Other interaction .21* .01 .17
2 2
R change .03 .00 .00 R change .03† .00 .02
2 2
Total R .09 .29*** .18*** Total R .29*** .34*** .24***

NOTE: The regression coefficients are shown for the step at which they NOTE: IOS = Inclusion of Other in the Self scale. The regression coef-
were entered into the equation. ficients are shown for the step at which they were entered into the equa-
a. n = 107. tion.
b. n = 105. a. n = 110.
c. n = 111. b. n = 83.
*p < .05. **p < .01. ***p < .001. c. n = 108.
†p < .10. *p < .05. **p < .01. ***p < .001.
mental models scales were centered prior to these
regression analyses (see Aiken & West, 1991). specific mental model variables was also significantly
related to well-being (R2 = .28) and life satisfaction (R2 =
Psychological Adjustment Outcomes .17). Again, examination of the betas revealed that this
When entered as the first step in the regression analy- effect was due solely to model of self. Specifically, indi-
ses, the block of general mental model variables was sig- viduals with a positive partner-specific model of self were
nificantly related to well-being and life satisfaction (but higher in well-being (β = .52) and life satisfaction (β =
not depression), accounting for 9% and 7% of the vari- .40) than were individuals with a negative partner-
ance in these adjustment measures, respectively. Exami- specific model of self. As was true of the general mental
nation of individual beta weights revealed that in both model, neither the partner-specific model of other nor
cases, this effect was solely attributable to general model the interaction between the two partner-specific mental
of self. Specifically, individuals with a positive general models was related to any of the psychological adjust-
model of self were higher in well-being (β = .27) and ment variables.
higher in life satisfaction (β = .24) than were those with a These analyses suggest that both general and
negative general model of self. (Neither general model partner-specific model of self are related to overall psy-
of others nor the interaction between the two general chological health. However, the significant betas for
mental models was related to any of the psychological partner-specific model of self were typically larger in
adjustment variables.) Interestingly, when entered as the magnitude than the corresponding betas for general
first step in the regression analyses, the block of partner- model of self (.52 vs. .27 and .40 vs. .24), suggesting that
612 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

the specific mental model was the relatively stronger pre-


dictor. This conclusion is buttressed by examination of
the second step of the regression analyses. When the
block of specific mental model variables was entered on
the first step of the regression analyses, the block of gen-
eral mental model variables did not generate a signifi-
cant increase in the variance accounted for in any of the
psychological adjustment variables (see Table 2). How-
ever, when entered in the second step, the block of spe-
cific mental model variables accounted for an additional
20% of the variance in well-being and 11% of the vari-
ance in life satisfaction, above and beyond the variance
accounted for by the block of general mental models
Figure 1 The interaction of partner-specific models of self and other
(see Table 2). Thus, once the variance shared by the two as a predictor of relationship satisfaction.
types of mental models (6% to 8%) was partialed out,
only partner-specific model of self was uniquely related
to the psychological adjustment variables.
eral mental models were not included in the regression
Relationship-Specific Outcomes equation.
Finally, the block of interaction variables was margin-
When entered in the first step of the regression analy- ally significant as a predictor of relationship satisfaction.
ses, the block of partner-specific mental model variables Examination of the betas reveals that this effect was due
was significantly related to all of the relationship-specific to a significant interaction between the specific mental
outcomes (see Table 3), accounting for between 21% models. To probe this interaction, we followed the pro-
and 30% of the variance in these variables. Examination cedures outlined by Aiken and West (1991). Two simple
of the significant beta weights reveals that this effect was regression lines were plotted, showing the regression of
due to the impact of both model of self and model of model of other on relationship satisfaction for individu-
other. Individuals with a more positive partner-specific als with positive and negative models of self. Positive
model of self reported greater relationship satisfaction, model of self and positive model of other were repre-
inclusion of other in the self, and feelings of romantic sented by scores that were 1 standard deviation above 0,
love than did individuals with a negative partner-specific and negative model of self and negative model of other
model of self. The same was true for partner-specific were represented by scores that were 1 standard devia-
model of other, although the magnitudes of these effects tion below 0, with a 0 score representing an individual
were somewhat smaller than was the case for model of whose model was neither positive nor negative. Exami-
self. In contrast, when entered in the first step of the nation of Figure 1 reveals that individuals with positive
regression analyses, the block of general mental model models of both self and other (i.e., secures) were the
variables was not significantly related to any of the most satisfied with their relationships. Specifically,
relationship-specific outcomes (see Table 3). among individuals with a positive model of self, as model
The superiority of the specific mental models over the of other became more positive, relationship satisfaction
general mental models as predictors of relationship- increased, t(104) = 3.84, p < .01, whereas individuals with
specific outcomes can also be seen when both general a negative model of self reported consistent ratings of
and specific mental models are included in the regres- relationship satisfaction regardless of their model of
sion analyses. With the block of general mental model other, t(104) = 1.05, p > .05.
variables entered in the regression equation at the first
Regression Analyses Including Self-Esteem and Trust
step (see Table 3), the addition of the block of specific
mental model variables resulted in highly significant Recall that we had predicted that trust and self-esteem
increases in the amount of variance explained in all of would be more highly correlated with general mental
the relationship-specific outcomes (R2 change ranged models than with partner-specific mental models. A
from .19 to .33). With all of the mental model variables glance at Table 1 reveals that trust was more strongly
included in the regression equation, both partner- related to general model of others (r = .31, p < .01) than
specific model of self and other remained significant as to partner-specific model of other (r = –.01, p > .10). A
predictors of relationship satisfaction, IOS, and roman- Fisher’s z test revealed that the difference between these
tic love. The betas associated with these effects were two correlations was indeed significant (z = 3.41, p < .01).
nearly identical to those that were obtained when gen- Contrary to our predictions, however, general model of
Cozzarelli et al. / MENTAL MODELS OF ATTACHMENT 613

self (r = .13, p > .10) was not more highly correlated with for life satisfaction. Thus, the obtained relationships
self-esteem than was specific model of self (r = .20, p < .05; between specific model of self and these outcome vari-
Fisher’s z = .75, p > .10). ables could not be completely attributed to the overlap
To test the degree to which mental models of attach- between model of self and self-esteem. However, self-
ment are proxy variables for self-esteem and trust, the esteem proved to be somewhat more strongly related to
regressions we described earlier were repeated, control- psychological adjustment than specific model of self.
ling for self-esteem and trust. These variables were Whereas self-esteem was clearly related to psychologi-
entered as covariates in Step 1 of each regression analy- cal adjustment, trust was not related to any of the adjust-
sis. Because these analyses did not alter the conclusions ment indices.
we have already drawn about the relative strengths of the Relationship-specific outcomes. The block of covariates
general versus specific mental models as predictor vari- was a significant predictor of relationship satisfaction
ables, we focus our discussion below on the relationships (R 2 = .06) and a marginal predictor of IOS (R 2 = .06) and
between mental models and self-esteem/trust. romantic love (R 2 = .05). Again, examination of the betas
Psychological adjustment variables. The block of covari- revealed that all of these effects were due to the signifi-
ates was significantly related to all three psychological cant impact of self-esteem. Higher self-esteem was
adjustment variables, explaining between 21% and 24% related to increased relationship satisfaction (β = .23),
of the variance in these measures. Inspection of the increased closeness (β = .25), and greater perceptions of
betas, however, reveals that these significant effects are romantic love (β = .22). Trust was not significantly
due solely to the impact of self-esteem. As might be related to any of the relationship-specific outcomes.
expected, higher self-esteem was related to lower levels Recall that in the regression analyses without the covari-
of depression (β = –.45), higher levels of well-being (β = ates, partner-specific (but not general) mental models
.45), and increased life satisfaction (β = .43). Recall that were significantly related to all of the relationship-
in the analyses without the covariates, when entered in related outcome variables. As we had hypothesized, this
the first step of the regressions, both the general and the was still the case, even when controlling for self-esteem
specific mental model variable blocks were significant as and trust. In fact, the betas for the specific mental mod-
predictors of well-being and life satisfaction. With the els that were obtained in the regressions with the covari-
covariates in the equation, the block of general mental ates included were almost identical to those that were
model variables was only significant in predicting well- obtained when the covariates were omitted. Interest-
being (R 2 change = .05). In contrast, with the covariates ingly, with all of the mental model variables included in
in the equation, the block of specific mental models was the regression equations, self-esteem was no longer sig-
still related to both well-being (R 2 change = .20) and life nificantly related to IOS (β = .11, p > .10) or romantic
satisfaction (R 2 = .10). Thus, as we hypothesized, control- love (β = .14, p > .10) and was only marginally related to
ling for self-esteem and trust reduced the impact of gen- relationship satisfaction (β = .15, p < .10). Thus, in this
eral mental models of attachment in these analyses but case, not only were the effects of specific model of self
did not significantly diminish the impact of partner- not attributable to its overlap with self-esteem but self-
specific mental models. esteem was less strongly related to these outcome vari-
Finally, recall that in the analyses without the covari- ables than was model of self.
ates, when both the general and the specific mental Finally, the interaction between the specific mental
model variables were entered as predictors of psycho- models that was obtained in predicting relationship sat-
logical adjustment, only partner-specific model of self isfaction was still significant when self-esteem and trust
was uniquely related to the outcome variables. With self- were included in the regression equations.
esteem included in the regressions, partner-specific
Summary
model of self was still significant as a predictor of both
well-being and life satisfaction, although the magnitudes In summary, model of self (whether general or
of these betas were somewhat reduced by the presence of partner-specific) was related to both life satisfaction and
the covariates. Interestingly, with all of the mental model positive well-being; however, the magnitudes of these
variables included in the equation, the betas for self- relationships were stronger for the specific mental
esteem were still significant as well. In fact, in two of model. Model of other was generally unrelated to overall
three cases, inspection of the betas suggests that self- psychological adjustment. Specific mental models were
esteem is a slightly stronger predictor than specific much more strongly related to relationship-specific out-
model of self. The betas for self-esteem and specific comes than were general mental models. Both specific
model of self were –.42 (p < .001) and –.09 (p > .10), model of self and specific model of other were related to
respectively, for depression; .35 (p < .001) and .43 (p < a wide variety of relationship-specific outcomes. In addi-
.001) for well-being; and .37 (p < .001) and .29 (p < .01) tion, self-esteem was more strongly related to global psy-
614 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

chological adjustment than was specific model of self, and specific forms of the attachment measure contain
but specific model of self was more strongly related to the same content (comfort with relationships, desire for
outcomes in relationships. Trust was not related to any of intimacy, etc.) but differ only in their level of specificity
the outcome variables. as to the target of attachment, these results clearly sup-
port our contention that measures will be most strongly
DISCUSSION associated when they are assessed at the same level of
specificity. However, it must also be acknowledged that
Multiple Mental Models
feelings of intimacy, and so forth, when assessed with ref-
The results of the current study are consistent with erence to one’s partner, bear a conceptual resemblance
the suggestion that individuals can have different mental to some of the relationship-related outcome variables.
models for different relationships (e.g., Bartholomew & This conceptual overlap may, in part, account for the
Horowitz, 1991; Collins & Read, 1994; Shaver et al., strength of the relationships between these two sets of
1996) and, more specifically, that one’s general attach- variables.4
ment style may not be redundant with one’s style in a par- We had also hypothesized that general mental models
ticular relationship (see also Baldwin et al., 1996). In this would be more strongly related to global psychological
study, we found that individuals’ mental models of their functioning than would specific mental models. Surpris-
current relationship were only modestly correlated with ingly, although both general and specific model of self
their more overarching or general mental models of were related to well-being and life satisfaction, in the
attachment. Presumably, general mental models reflect hierarchical regression analyses, the general mental
an average assessment of the self and others across rela- model was a poorer predictor than the specific mental
tionships, and relationship-specific beliefs about the self model. Indeed, most of the effects of the general model
and one’s partner would logically represent only one were due to shared variance with the more specific
fraction of this overall assessment. It will be interesting model of self measure. Thus, a researcher who uses a
for future research to test more directly the assumption general measure of attachment styles may well find that
that general mental models can be derived from a sum- scores on this measure will be significantly related to
mation of relationship-specific mental models. To exam- global psychological functioning; however, such results
ine this issue, researchers could assess simultaneously may largely reflect the hidden operation of specific men-
participants’ mental models of each of their important
tal models about individual intimates.
attachment figures. Interestingly, in a recent study, Davis
We were surprised by the poor performance of the
et al. (1998) found that global perceptions of social sup-
general mental model variables in this study. The paucity
port could not be derived by summing individuals’ per-
of significant associations involving these variables may
ceptions of support across various specific relationships.
perhaps be a function of the particular dependent mea-
These authors argued instead that global support func-
sures we chose to use. It is possible, for example, that
tions in a trait-like fashion and largely reflects an indi-
general mental models of attachment would relate more
vidual’s overarching tendency to see the self and the sur-
strongly (and uniquely) to global outcomes that are
rounding social world in a positive or negative way.
Perhaps general mental models of attachment function clearly relationship-oriented (e.g., general perceptions
similarly. of social support, loneliness) than to measures of overall
psychological functioning such as the ones we used here.
The Relationships Between Alternatively, the weak performance of the general mod-
Mental Models and the Outcomes Variables els may reflect the fact that the individuals in our sample
We hypothesized that partner-specific mental models were in romantic relationships of some duration. Gener-
would be more strongly related to relationship-related ally, strong relational bonds have been shown to be one
outcomes than would general or overarching mental of the best predictors of happiness and well-being
models (see also Shaver et al., 1996). The results of this (Bowlby, 1979; Hazan & Shaver, 1994; Meyers & Diener,
study clearly corroborate this hypothesis. Specific mod- 1995). For example, studies have consistently shown a
els of both self and other were related to all three of the relationship between being married and better mental
relationship-related outcome measures that we included health and physical well-being (Fincham, 1998). Indeed,
in the study (relationship satisfaction, inclusion of oth- some have argued that relationships are more important
ers in the self, feelings of romantic love), and the interac- than money, career, and religion in contributing to an
tion between the two specific mental models was related overall sense that life is meaningful (Duck, 1998). Thus,
to relationship satisfaction as well. In contrast, general the overall psychological adjustment of those in rela-
mental models were not significantly related to any of tively stable or long-term relationships may become dis-
the relationship-related outcomes. Given that the global proportionately influenced by feelings about their part-
Cozzarelli et al. / MENTAL MODELS OF ATTACHMENT 615

ner. Abstract beliefs or expectations about people in clear overlap between self-esteem and general model of
general may assume a secondary role in such cases. self, in the hierarchical regression analyses we con-
Overall, in this study, there were no instances in which ducted in this study, partner-specific model of self and
the general mental models were more strongly associ- self-esteem had largely independent effects. In fact, the
ated with the outcome variables than were the specific betas for specific model of self were only minimally
mental models. This raises the issue of whether there are reduced when self-esteem was added to the regression
any situations in which general mental models would be analyses. That specific model of self and global self-
expected to be more strongly related to people’s esteem have independent or unique relationships with
thoughts, feelings, and behavior than more other variables is not surprising. Relationship-specific
relationship-specific mental models. Several authors model of self is probably more akin to a domain-specific
have made the point that individuals’ responses in an self-esteem measure than it is to overall or global self-
attachment-relevant situation will be most affected by esteem. That is, global self-esteem has been hypothe-
the mental models that are most accessible to the indi- sized to represent a sum or average of how an individual
vidual in that situation (Baldwin et al., 1996; Shaver et al., feels about himself or herself across a wide variety of
1996). Although general mental models can be concep- domains. How one feels about the self in the context of a
tualized as representations of an individuals most fre- particular relationship (or domain) may simply be one
quently experienced attachment pattern (Baldwin et al., of many contributors to how one feels about the self
1996), specific mental models should be primed when overall.
the individual is thinking about or acting in the context The hierarchical regression analyses also allowed us
of a particular relationship. Thus, it is likely that when to compare the influence of the general mental models,
interacting with others who are well-known, people will specific mental models, and self-esteem. With all three
typically rely on mental models specifically based on sets of variables in the regression equations, specific
those individuals and relationships. General mental
model of self and self-esteem were both independently
models may be most likely to come into play in situations
associated with the psychological adjustment variables;
where an individual is interacting with someone new or
however, the general mental models were not. Specific
does not currently have close relationships with others.
model of self was somewhat more strongly related to
In cases such as these, the individual would not have
well-being than was self-esteem but it was less strongly
available a more specific set of mental models and would
related to life satisfaction and, unlike self-esteem, it was
be expected to rely on generalized past interactions as a
not related to depression. When examining
guide in the current situation. Similarly, if an individu-
relationship-related outcomes, a different picture
al’s friend or partner does something that he or she has
emerged from our data. Specific model of self clearly
never done before (e.g., provoke jealousy, behave in an
unresponsive fashion), general mental models may had the strongest relationships with these outcome vari-
again influence the affected person’s responses. ables. It was related to all of the relationship-specific out-
come variables, whereas self-esteem was only marginally
Construct Overlap related to one (relationship satisfaction) and general
Model of self and self-esteem. We had hypothesized that model of self was not related to any of these dependent
self-esteem would be more highly correlated with gen- measures.
eral model of self than with specific model of self. This Taken together, this evidence suggests that both spe-
hypothesis was not supported; self-esteem was related cific model of self and self-esteem have unique and
at roughly the same magnitude to both forms of model nonoverlapping relationships with overall psychological
of self. adjustment, although these relationships are somewhat
We had also hypothesized that holding self-esteem stronger for self-esteem. Thus, if a researcher’s goal is to
constant would reduce the relationships between gen- predict an individual’s overall psychological functioning
eral model of self and the outcome variables but that this and he or she can only assess a limited number of vari-
would not be the case for partner-specific model of self. ables, using self-esteem as a predictor would likely
Overall, this hypothesis was supported. The few signifi- account for more variance in adjustment than would
cant effects of general model of self were weakened with using model of self. However, if participant time and
the addition of self-esteem to the regression equations. questionnaire space are not an issue, assessing model of
These results are in line with those reported by Cozza- self as well as self-esteem will likely explain some addi-
relli et al. (1998). In their study, the relationships tional variance, at least with regard to global indices of
between general model of self and variables relevant to quality of life. When predicting relationship-related out-
adjustment to abortion were substantially reduced by comes, researchers should clearly use relationship-
partialing out the effects of self-esteem. In contrast to the specific attachment measures.
616 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

Model of others and trust. We had predicted that trust found, in contrast, that model of others was more
would be more highly correlated with general model of strongly related to relationship outcomes than was
others than with specific model of other. This hypothesis model of self. Thus, pending further research, we would
was supported. In fact, the correlation between trust and speculate that model of others is primarily related to how
specific model of other was not significant. We had also people feel about their relationships and that when this
hypothesized that when trust was entered into the mental model is related to overall mental health, it is
regression analyses as a covariate, the effects of general likely to be in the form of an interaction with model of
but not specific model of others would be reduced. self. Model of self appears to not only be related to how
Given that general model of others had no significant individuals feel about their relationships but to have a
effects, the part of the hypothesis relating to these men- strong association with their overall psychological func-
tal models could not be supported. However, as tioning as well.
expected, partner-specific model of other was related to
Caveats and Future Research
the relationship-related outcomes and the magnitudes
of these effects were not reduced by the presence of trust In this article, we sought to examine the relationships
in the regression equations. This was not surprising between mental models of attachment and both psycho-
given that trust and specific model of others were uncor- logical adjustment and relationship-related outcomes.
related and that trust was not related to the relationship- However, the mental models measure that we used,
specific outcome variables. These effects are likely due to although it has been shown to be valid, is clearly not
the fact that our trust measure asked about trusting peo- ideal. Measuring mental models directly would be more
ple in general. The effects of global trust such as this desirable but not necessarily easy, especially given the
would not be expected to reproduce the effects of trust- fact that there seems to be no consensus as to what
ing one’s partner, who could well be perceived as more exactly comprises mental models. Some formulations of
or less trustworthy than people in general. In future mental models (e.g., Collins & Read, 1994; Shaver et al.,
research, it would be informative to use a person-specific 1996) have been quite complex and include things such
trust measure such as the Specific Interpersonal Trust as goals, behaviors, unconscious urges, and psychologi-
Scale (Johnson-George & Swap, 1982) or the Trust Scale cal defenses, whereas the implicit definition of mental
(Rempel et al., 1985) as a covariate. Presumably, such a models derived from the Bartholomew and Horowitz
measure would overlap much more with an individual’s (1991) measure is, of necessity, simpler. Because many of
mental model of his or her partner and might reduce the the predictions that researchers make about the impact
relationships between partner-specific model of other of adult attachment styles are predicated on the work-
and relationship-relevant outcomes. ings of mental models, the field of attachment research
would greatly benefit from both a clarification of the
Model of Self Versus Model of Others
mental models construct and the development of a
Although we had not anticipated it, in our regression direct measure of these models.
analyses, model of self had more numerous and gener- The current study also could have been improved by
ally stronger effects than did model of others. Model of the use of a more diverse group of participants, espe-
self was related to some of the psychological adjustment cially older individuals with more experience in relation-
measures, but this was not the case for model of others or ships. Such individuals would logically have very well-
the interaction between the two mental models. Model developed mental models about their partners and these
of others was related to the relationship-relevant out- models would likely be even more influential than the
comes, but only when assessed at the specific level. Thus, specific mental models held by the participants in our
model of self was related to a wide variety of feelings study. It would also be interesting to see whether the gen-
about the self and intimate relationships but the role of eral mental models of individuals who do not have a
model of others was much more circumscribed. Because romantic partner show stronger associations with overall
few studies have directly compared the correlates or psychological adjustment than the general mental mod-
effects of model of self and model of others, it is difficult els of individuals with a romantic partner (such as our
to know if our results are generalizable. However, some research participants). Finally, we believe that this study
studies do yield evidence that is roughly consistent with is only the necessary first step in exploring the differen-
our findings. For example, Cozzarelli et al. (1998) exam- tial correlates and impact of general and relationship-
ined the relationships between general mental models specific mental models. Future research should extend
and postabortion well-being and distress. Model of self this study and should examine a broader array of out-
was related to both psychological outcomes, but model come variables (e.g., global perceptions of support,
of others was not. (The two mental models did interact in relationship-specific perceptions of trust, loneliness) to
predicting well-being, however.) Lipschultz et al. (1995) more clearly delimit the utility of various types of attach-
Cozzarelli et al. / MENTAL MODELS OF ATTACHMENT 617

ment measures. Such an endeavor should make use of 2. We chose to explore the effects of mental models rather than the
more traditional attachment styles because doing so allowed us to
multiple methods of data collection (e.g., interviews, explicitly contrast the effects of model of self and model of others and
behavioral observation) and should seek to go beyond because this procedure allowed us to examine the construct overlap
the questionnaire methodology that we used here. issue in a clearer manner. However, we also examined the effects of
categorical attachment style on all of our dependent variables. These
Conclusions analyses yielded conclusions that did not differ substantively from
those we report in the body of the article.
3. Although relationship length was not correlated with any of the
Overall, partner-specific mental models were more outcome variables in the study, we thought it possible that this variable
strongly related to a variety of outcome variables than might moderate the associations between general/specific mental
were general mental models, especially when those out- models and the relationship-related outcome variables. To address this
issue, we examined all possible interactions of this type. The interac-
comes were related to the functioning of romantic rela- tion between relationship length and specific model of self was margin-
tionships. In addition, in our regression analyses, the ally associated with love (β = .26, p = .07). The pattern of this interaction
effects of partner-specific model of self were not redun- suggested that although reports of love decreased slightly over time for
dant with those of self-esteem and in some cases were all participants, this effect was more pronounced for individuals with a
positive model of self. No other interactions were significant.
stronger. Although general mental models were signifi- 4. Relatedly, it is possible that the association between the specific
cantly related to a few of the outcome measures we used, mental models and the relationship-related outcome variables is at
self-esteem typically demonstrated stronger associations least partly due to overlap in the items that comprise the two sets of
measures. To address this issue, we compared the items on the attach-
with these variables. Based on this set of results, we strongly ment scales with those comprising the relationship satisfaction and
recommend the use of relationship-specific measures of romantic love measures. The items comprising the relationship satis-
attachment over the more common general measures, faction measure were judged to be nonoverlapping with those com-
prising the attachment measures. However, three of the items in the
especially in cases where the researcher is interested in love scales were judged to be somewhat similar in wording to portions
the functioning or outcomes of a particular relationship. of the attachment measures. Accordingly, we dropped these items and
Overall, our results suggest that general measures of reran all of our analyses. The patterns of results that we report in the
body of this article were completely unchanged by this procedure.
attachment may be of fairly limited utility. They overlap
to some extent with other dispositional measures (e.g.,
Cozzarelli et al., 1998; Griffin & Bartholomew, 1994; REFERENCES
Shaver & Hazan, 1993) and in this study they did not do a
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpret-
particularly good job of explaining how people will ing interactions. Newbury Park, CA: Sage.
feel/behave in a particular relationship. Why have these Aron, A., Aron, E. N., & Smollan, D. (1992). Inclusion of other in the
commonly used measures generated so much in the way self scale and the structure of interpersonal closeness. Journal of Per-
sonality and Social Psychology, 63(4), 596-612.
of research results? Our feeling is that this is likely due to Baldwin, M. W., & Fehr, B. (1995). On the instability of attachment style
the fact that general mental models (or attachment ratings. Personal Relationships, 22, 247-261.
styles) correlate with more specific mental models and Baldwin, M. W., Keelan, M.P.R., Fehr, B., Enns, V., & Koh-Rangarajoo,
E. (1996). Social-cognitive conceptualizations of attachment work-
are related to global dispositional variables that affect ing models: Availability and accessibility effects. Journal of Personality
psychological adjustment. Accordingly, general mental and Social Psychology, 71, 94-109.
models inherently share some variance with other vari- Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among
young adults: A test of a four-category model. Journal of Personality
ables that have been successfully related to a very wide and Social Psychology, 61(2), 226-244.
array of outcomes. But it may well be the case that except Beck, A. T. (1976). Cognitive therapy and the emotional disorders. New York:
for individuals who are not currently involved in a spe- International Universities Press.
Bowlby, J. (1969). Attachment and loss: Vol. I. Attachment. Middlesex, UK:
cific relationship, these global attachment measures Penguin.
would always be of limited utility and that greater Bowlby, J. (1973). Attachment and loss: Vol. II. Separation: Anxiety and
explanatory precision would be achieved by using anger. New York: Basic Books.
Bowlby, J. (1979). The making and breaking of affectional bonds. London:
attachment measures that are more precisely tailored to Tavistock.
the relationship of interest. We also conclude (as does Bowlby, J. (1980). Attachment and loss: Vol. III. Loss. New York: Basic
Baldwin et al., 1996) that conceptualizing attachment Books.
Brannon, R. (1976). Attitudes and the prediction of behavior. In B. Sei-
styles as stable dispositional variables obscures much denberg & A. Snadowsky (Eds.), Social psychology (pp. 145-197).
variability across relationships and that focusing instead New York: Free Press.
on the particular mental models that are accessible or Brennan, K. A., & Bosson, J. K. (1998). Attachment-style differences in
attitudes toward and reactions to feedback from romantic partners:
activated in a given situation may be a more fruitful An exploration of the relational bases of self-esteem. Personality and
approach. Social Psychology Bulletin, 24, 699-714.
Bretherton, I., Biringen, Z., Ridgeway, D., Maslin, C., & Sherman, M.
(1989). Attachment: A parental perspective. Infant Mental Health
NOTES Journal, 10, 203-221.
Bridges, L. J., Connell, J. P., & Belsky, J. (1988). Similarities and differ-
1. Comparisons between the two samples revealed no differences ences in infant-mother and infant-father interaction in the strange
on any of the variables in the study; therefore, subsequent analyses col- situation: A component process analysis. Developmental Psychology,
lapsed across the two groups. 24, 92-100.
618 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

Collins, N. L., & Read, S. J. (1990). Adult attachment, working models, ity better than attachment styles. Paper presented at the annual meet-
and relationship quality in dating couples. Journal of Personality and ing of the American Psychological Society, New York.
Social Psychology, 58, 644-663. Major, B., Zubek, J. M., Cooper, M. L., Cozzarelli, C., & Richards, C.
Collins, N. L., & Read, S. J. (1994). Cognitive representations of attach- (1997). Mixed messages: Implications of social conflict and social
ment: The content and function of working models. In K. Bartholo- support within close relationships for adjustment to a stressful life
mew & D. Perlman (Eds.), Advances in personal relationships (Vol. 5, event. Journal of Personality and Social Psychology, 72(6), 1349-1363.
pp. 53-90). London: Jessica Kingsley. Meyers, D. G., & Diener, E. (1995). Who is happy? Psychological Science,
Cozzarelli, C., Sumer, N., & Major, B. (1998). Mental models of attach- 6, 10-19.
ment and coping with abortion. Journal of Personality and Social Psy- Mikulincer, M. (1998). Attachment working models and the sense of
chology, 74, 453-467. trust: An exploration of interaction goals and affect regulation.
Davis, M. H., Morris, M. M., & Kraus, L. A. (1998). Relationship-specific Journal of Personality and Social Psychology, 74, 1209-1224.
and global perceptions of social support: Associations with well- Mikulincer, M., & Nachshon, O. (1991). Attachment styles and pat-
being and attachment. Journal of Personality and Social Psychology, 74, terns of self-disclosure. Journal of Personality and Social Psychology, 61,
468-481. 321-331.
Diener, E., Emmons, R. A., Larson, R. J., & Griffin, S. (1985). The satis- Pietromonaco, P. R., & Feldman Barrett, L. (1997). Working models of
faction with life scale. Journal of Personality Assessment, 49(1), 71-75. attachment and daily social interactions. Journal of Personality and
Duck, S. (1998). Relating to others. Chicago: Dorsey. Social Psychology, 73, 1409-1423.
Feeney, J. A., & Noller, P. (1990). Attachment style as a predictor of Rempel, J. K., Holmes, J. G., & Zanna, M. P. (1985). Trust in close rela-
adult romantic relationships. Journal of Personality and Social Psychol- tionships. Journal of Personality and Social Psychology, 49, 95-112.
ogy, 58, 281-291. Rosenberg, M. (1957). Occupations and values (pp. 25-35). Glencoe, IL:
Fincham, F. D. (1998). Marital quality. In E. A. Blechman & K. D. Free Press.
Brownell (Eds.), Behavioral medicine and women: A comprehensive Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ:
handbook (pp. 108-112). New York: Guilford. Princeton University Press.
Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: Rothbard, J. C., & Shaver, P. R. (1994). Continuity of attachment across
An introduction to theory and research. Reading, MA: Addison-Wesley. the lifecourse: An attachment-theoretical perspective on personal-
Fox, N. A., Kimmerly, N. L., & Schafer, W. D. (1991). Attachment to ity. In M. B. Sperling & W. H. Berman (Eds.), Attachment in adults:
mother/attachment to father: A meta-analysis. Child Development, Theory, assessment, and treatment (pp. 31-71). New York: Guilford.
62, 210-225. Ryff, C. D. (1989). Happiness is everything, or is it? Explorations on the
Griffin, D., & Bartholomew, K. (1994). Models of the self and other: meaning of psychological well-being. Journal of Personality and Social
Fundamental dimensions underlying measures of adult attach- Psychology, 57, 1069-1081.
ment. Journal of Personality and Social Psychology, 67(3), 430-445. Shaver, P. R., Collins, N., & Clark, C. L. (1996). Attachment styles and
Hazan, C., & Shaver, P. (1987). Conceptualizing romantic love as an internal working models of self and relationship partners. In G.J.O.
attachment process. Journal of Personality and Social Psychology, 52, Fletcher & J. Fitness (Eds.), Knowledge structures in close relationships:
511-524. A social psychological approach (pp. 25-61). Mahwah, NJ: Lawrence
Hazan, C., & Shaver, P. R. (1994). Attachment as an organizational Erlbaum.
framework for research on close relationships. Psychological Inquiry, Shaver, P. R., & Hazan, C. (1993). Adult romantic attachment: Theory
5, 1-22. and evidence. In K. Bartholomew & D. Perlman (Eds.), Advances in
Hendrick, S. S. (1988). A generic measure of relationship satisfaction. personal relationships (Vol. 4, pp. 29-70). London: Jessica Kingsley.
Journal of Marriage and the Family, 50, 93-98. Simpson, J. A. (1990). The influence of attachment style on romantic
Johnson-George, C., & Swap, W. C. (1982). Measurement of specific relationships. Journal of Personality and Social Psychology, 59, 971-980.
interpersonal trust: Construction and validation of a scale to assess Simpson, J. A., Rholes, W. S., & Nelligan, J. S. (1992). Support seeking
trust in a specific other. Journal of Personality and Social Psychology, 43, and support giving within couples in an anxiety-provoking situa-
1306-1317. tion: The role of attachment styles. Journal of Personality and Social
Kirkpatrick, L. A., & Davis, K. E. (1994). Attachment style, gender, and Psychology, 62, 434-446.
relationship stability: A longitudinal analysis. Journal of Personality Sternberg, R. J. (1988). The triangle of love: Intimacy, passion, commitment.
and Social Psychology, 66, 502-512. New York: Basic Books.
Kobak, R. R., & Hazan, C. (1991). Attachment in marriage: The effects Sumer, N., Cozzarelli, C., & Burns, M. (1996, May). A direct measure of
of security and accuracy of working models. Journal of Personality and mental models of attachment. Poster presented at the 68th annual con-
Social Psychology, 60, 861-869.
vention of the Midwestern Psychological Association, Chicago.
Lamb, M. E. (1978). Qualitative aspects of mother-infant and father-
infant attachments. Infant Behavior and Development, 1, 265-275.
Lipschultz, W., Wall, J. J., Scotilla, S., Bylsma, W. H., & Cozzarelli, C. Received July 6, 1998
(1995, June). Attachment self and other models predict relationship qual- Revision accepted February 3, 1999

You might also like