You are on page 1of 13


Sibley et al. / RELIABILITY

Reliability and Validity of the Revised Experiences

in Close Relationships (ECR-R) Self-Report Measure
of Adult Romantic Attachment

Chris G. Sibley
Ronald Fischer
James H. Liu
Victoria University of Wellington, New Zealand

Three studies examine the psychometric properties (i.e., the test- self, and attachment-related avoidance, or a model of
retest reliability, convergent, and discriminant validity) of others (Bartholomew & Horowitz, 1991; Brennan et al.,
Fraley, Waller, and Brennans Revised Experiences in Close 1998; Griffin & Bartholomew, 1994a, 1994b). According
Relationships (ECR-R) self-report measure of romantic attach- to Fraley and Shaver (2000) attachment-related anxiety
ment anxiety (model of self) and avoidance (model of others). reflects an individuals predisposition toward anxiety
Longitudinal analyses suggest that the ECR-R provided highly and vigilance concerning rejection and abandonment,
stable indicators of latent attachment during a 3-week period whereas the avoidance dimension corresponds to dis-
(85% shared variance). Hierarchical linear modeling analyses comfort with closeness and dependency or a reluctance
further validated the ECR-R, suggesting that it explained to be intimate with others (pp. 142-143).
between 30% to 40% of the between-person variation in social Although most researchers now agree that the attach-
interaction diary ratings of attachment-related emotions experi- ment behavioral system is composed of these two funda-
enced during interactions with a romantic partner and only mental dimensions, internal working models, or cogni-
5% to 15% of that in interactions with family and friends. tive subsystems (Crowell, Fraley, & Shaver, 1999; Fraley &
Guidelines are offered regarding the conditions where highly reli- Spieker, 2003; Mikulincer & Shaver, 2003), various
able and precise measures of romantic attachment, such as the researchers have raised concerns regarding how the
ECR-R, are deemed necessary and where shorter, albeit slightly dimensions may be most accurately operationalized
less reliable measures, such as Bartholomew and Horowitzs using self-report measures (Fraley, Waller, & Brennan,
Relationship Questionnaire, may also be viable. 2000; Kurdek, 2002; see J. A. Feeney, 2002a; Shaver &
Mikulincer, 2002, for a discussion). Such concerns are
especially pertinent given the variety of alternative mea-
Keywords: romantic attachment; measurement; validity; reliability; sures of anxiety and avoidance that have been developed
psychometrics throughout the years (e.g., Bartholomew & Horowitz,
1991; Brennan et al., 1998; Collins & Read, 1990; Griffin
The attachment literature has documented a prolific & Bartholomew, 1994b; J. A. Feeney, Noller, &
Hanrahan, 1994; Fraley et al., 2000; Simpson, Rholes, &
increase in the number of self-report measures of adult
attachment published in the years following Hazan and Nelligan, 1992). Detailed research assessing the reliabil-
Shavers (1987) seminal adaptation of Ainsworths infant ity and validity of many of these different measures
typology (Ainsworth, Blehar, Waters, & Wall, 1978). Con- remains limited however (e.g., Brennan et al., 1998;
sistent with Bowlbys (1969/1982, 1973, 1980) original
theorizing, research during the past two decades has Authors Note: This research was supported by a research grant from
converged on a definition of adult attachment based on the Foundation for Research Science and Technology, New Zealand.
This article comprises part of Chris G. Sibleys doctoral dissertation,
two primary dimensions (Bartholomew & Horowitz,
completed under the supervision of James H. Liu and Ronald Fischer.
1991; Brennan, Clark, & Shaver, 1998; Griffin &
PSPB, Vol. 31 No. 11, November 2005 1524-1536
Bartholomew, 1994a). These two dimensions are thought DOI: 10.1177/0146167205276865
to reflect attachment-related anxiety, or a model of the 2005 by the Society for Personality and Social Psychology, Inc.


Kurdek, 2002). In their comprehensive review of the demonstrated in their analyses of previous attachment
area, Mikulincer and Shaver (2003) emphasized the scales, it is therefore not uncommon for a scale to accu-
need for further research addressing this issue when rately differentiate between individuals who score in the
they stated that high end of a trait range but fail to adequately differenti-
ate between individuals who score in the moderate or
much important work remains to be done in the area of low ends of the range. Undesirable response properties
measuring attachment phenomena. The self-report of this type can in turn generate misleading inferences
measures on which our research is based were created concerning individual differences in trait stability and
through a combination of research on infants, intuition, change (Fraley et al., 2000, p. 364). Consistent with this
psychometrics, and convenience. . . . They were all based concern, comparison of the test information functions
on item writers exposure to the attachment literature, of the anxiety and avoidance subscales of the ECR-R with
but in most cases were not designed component-by-
those of other popular scales from which its items were
component with a coherent theoretical framework in
garnered, such as the ECR and AAQ, demonstrated that
mind. Given their relative crudeness, it is remarkable
how systematic and cumulative our research findings the ECR-R provided substantially more precise estimates
have been. (pp. 140-141) of latent attachment across the entire trait range (Fraley
et al., 2000). Simulation analyses conducted by Fraley
et al. further validated the ECR-R, confirming that its
The present research details three studies that seek to
improved item parameters yielded markedly more
address the aforementioned concerns by assessing the
stable test-retest estimates than those provided by the
reliability, temporal stability, construct, discriminant,
aforementioned measures.
and convergent validity of Fraley et al.s (2000) recently
Subsequent research supports Fraley et al.s (2000)
developed 36-item Revised Experiences in Close Rela-
initial work and suggests that the ECR-R also displays
tionships (ECR-R) measure of romantic attachment.
adequate classical psychometric properties (e.g., Sibley
Comparable analyses of Bartholomew and Horowitzs
& Liu, 2004). Elaborating on Fraley et al.s exploratory
(1991) widely used Relationship Questionnaire (RQ)
factor analyses, Sibley and Liu (2004) reported that the
are also reported. Given its brevity and extensive use in
ECR-R provided reliable and replicable self-report mea-
previous research, we argue that the RQ provides a
benchmark against which the psychometric properties sures of romantic attachment anxiety and avoidance.
of more lengthy and comprehensive scales, such as the Consistent with Fraley et al.s simulation estimates, latent
ECR-R, may be evaluated. indicators of the ECR-R anxiety and avoidance subscales
Fraley et al. (2000) developed the ECR-R in an displayed test-retest correlations in the low .90s during a
attempt to foster consensus regarding the most appro- 6-week period (Sibley & Liu, 2004). These results suggest
priate method for assessing self-reported adult romantic that the ECR-R provides more stable test-retest estimates
attachment. The ECR-R items were selected from an of romantic attachment-related anxiety and avoidance
exhaustive set of more than 300 attachment items previ- than those reported in previous research during compa-
ously collated by Brennan et al. (1998). Fraley et al. rable time periods using earlier measures of attachment
(2000) analyzed these items using an innovative combi- (e.g., Collins & Read, 1990; Davila & Sargent, 2003).
nation of classical psychometric techniques, such as fac- The Present Research
tor analysis, and item response theory analysis. The ECR-
R therefore contains many similar and in some cases Elaborating on preliminary research in this area (e.g.,
identical items to those used in previous scales devel- Sibley & Liu, 2004), three studies are detailed that exam-
oped using more traditional psychometric methods, ine the ECR-Rs performance across a wide range of clas-
such as the ECR questionnaire (Brennan et al., 1998) sical psychometric criteria. Study 1 examines the tempo-
and the Adult Attachment Questionnaire (AAQ; ral stability and factor structure of the ECR-R. Study 2
Simpson et al., 1992). then uses Confirmatory Factor Analysis (CFA) to more
The ECR-R differs from these earlier scales, however, formally examine the factor structure of the ECR-R.
in that the combination of items included in the scale Finally, the third study examines the ECR-R using a dif-
provide a series of item discrimination values more ferent but equally important criterion set, that of conver-
evenly distributed across the entire trait ranges of anxi- gent and discriminant validity. Specifically, Study 3 uses
ety and avoidance. Previous multi-item attachment an experience sampling methodology to examine the
scales in contrast tend to contain series of items that are proportions of variance in general and attachment-
highly intercorrelated (indicating that the items mea- specific emotions experienced during social interac-
sure the same construct) but that say nothing about the tions involving a romantic partner, family member, or
scales abilities to accurately differentiate underlying platonic friend predicted by the ECR-R. RQ measures of
trait levels across the trait range. As Fraley et al. (2000) romantic anxiety and avoidance are also included in

Studies 1 and 3 to provide a baseline comparison against how accurately each prototype described their overall
which the ECR-R may be judged. experiences in romantic and love relationships on a
scale ranging from 1 (does not describe me at all) to 9
STUDY 1 (describes me very well). Following the procedure outlined
by Griffin and Bartholomew (1994b), ratings of attach-
According to Bowlby (1969/1982, 1973, 1980), work- ment anxiety and avoidance were then calculated using
ing models of attachment are dynamic cognitive struc- the formulae described in Table 1.
tures that are likely to change relatively slowly over time Fraley et al.s (2000) ECR-R contained 18 Likert-type
through the incorporation and adaptation of new infor- items assessing romantic attachment anxiety (M = 2.16,
mation and experiences (Fraley, 2002). However, as SD = 1.08) and 18 items assessing romantic attachment
Baldwin and Fehr (1995) noted, the considerable levels avoidance (M = 2.06, SD = 1.13; shown in Table 1). Items
of short-term instability observed in both earlier categor- were rated on a scale ranging from 0 (strongly disagree) to
ical measures of attachment style (e.g., Pistole, 1989) 6 (strongly agree). Participants were instructed to please
and more recently developed continuous measures of take a moment to think about your overall experiences
attachment-related anxiety and avoidance raise con- in romantic/love relationships, including both your pre-
cerns about the degree to which such changes may be vious and current relationship experiences. Please
caused by (a) measurement error or (b) unexplained answer the following questions with these experiences in
but genuine change in working models (e.g., Lopez & mind when completing both the ECR-R and RQ.
Gormley, 2002; Pierce & Lydon, 2001; Simpson, Rholes, People who completed only the first testing phase did
Campbell, & Wilson, 2003). not differ from those who completed both phases on any
Elaborating on Sibley and Liu (2004), Study 1 exam- demographics or measures (Fs < 1).
ined the test-retest stability of the ECR-R and RQ across a
3-week period. Consistent with previous research (e.g., Results
B. C. Feeney & Collins, 2001; Simpson, Rholes, &
Phillips, 1996), it was first hypothesized that ECR-R and Preliminary analyses. ECR-R measures of anxiety and
RQ measures of attachment anxiety and avoidance avoidance were strongly positively correlated, r(298) =
would load on the same factors, indicating that both .48, p < .001. RQ measures of anxiety and avoidance, in
scales assess attachment-related anxiety and avoidance. contrast, were only weakly positively correlated, r(298) =
Given that Fraley et al. (2000) reported that ECR-R mea- .20, p < .001. ECR-R and RQ measures of anxiety during
sures of anxiety and avoidance displayed levels of stabil- Time 1 were moderately positively correlated, r(298) =
ity equivalent to rs between .94 and .95 under simulated .60, p < .001, as were measures of avoidance, r(298) = .62,
conditions where the underlying constructs were p < .001. These items were then entered into a principal
assumed to be perfectly stable, Hypothesis 2 predicted components exploratory factor analysis with direct
that latent ECR-R indices of these two constructs would oblimin rotation. The eigenvalues displayed a steeply
display similar levels of stability when examined across a decreasing trend that leveled out after the second value
relatively short 3-week time period. (13.15, 4.97, 1.55, 1.48, 1.17, and 1.12), thus supporting
a two-factor solution that explained 48% of the scale vari-
Method ance. As can be seen in Table 1, the ECR-R items assess-
ing attachment avoidance and the RQ measure of avoid-
Participants. Participants were undergraduate stu-
ance derived from prototype ratings loaded on the first
dents who received partial course credit for participa-
factor, whereas ECR-R and RQ measures of anxiety
tion. Three hundred people (67% female; mean age =
loaded on the second factor.
22) participated during Phase 1 (51% were involved in a
romantic relationship). One hundred and seventy-two To more explicitly test whether the ECR-R and RQ
of these people (72% female) also participated during assessed the same pair of attachment dimensions, com-
the second testing phase (47% of whom were posite anxiety and avoidance scores were created for
romantically involved). both the ECR-R and RQ, and these four indices were
then analyzed using a second exploratory factor analysis
Procedure and materials. Participants completed a sur- with direct oblimin rotation. This analysis yielded two
vey containing both the RQ (Bartholomew & Horowitz, factors with eigenvalues greater than 1 that explained
1991) and ECR-R (Fraley et al., 2000) during two ses- 83% of the variance. ECR-R and RQ composite measures
sions 3 weeks apart. Participants questionnaires were of attachment-related anxiety both loaded solely on the
matched using confidential identification numbers. first factor (loadings = .92 and .86, respectively), whereas
Bartholomew and Horowitzs (1991) RQ contained composite measures of attachment-related avoidance
descriptions of four attachment prototypes (secure, dis- both loaded solely on the second factor (loadings = .74
missing, preoccupied, and fearful). Participants rated and .98, respectively).

TABLE 1: Factor Loadings for the RQ and ECR-R Measures of Attachment-Related Anxiety and Avoidance (Study 1)

Measure Factor 1 Factor 2

RQ model of others (dismissing + fearful prototypes) (secure + preoccupied prototypes) .78

ECR-R attachment-related avoidance ( = .94)
I prefer not to be too close to romantic partners .77
I prefer not to show a partner how I feel deep down .75
I usually discuss my problems and concerns with my partner (r) .73
I tell my partner just about everything (r) .73
I get uncomfortable when a romantic partner wants to get very close .73
It helps to turn to my romantic partner in times of need (r) .70
I am nervous when partners get too close to me .70
Its easy for me to be affectionate with my partner (r) .70
Its not difficult for me to get close to my partner (r) .69
I am very comfortable being close to romantic partners (r) .68
I talk things over with my partner (r) .67
I feel comfortable sharing my private thoughts and feelings with my partner (r) .67
I dont feel comfortable opening up to romantic partners .66
I find it relatively easy to get close to my partner (r) .65
I find it easy to depend on romantic partners (r) .65
I feel comfortable depending on romantic partners (r) .65
I find it difficult to allow myself to depend on romantic partners .54
My partner really understands me and my needs (r) .49

RQ model of self (preoccupied + fearful prototypes) (secure + dismissing prototypes) .54

ECR-R attachment-related anxiety ( = .93)
Im afraid that I will lose my partners love .88
I often worry that my partner will not want to stay with me .86
I worry that romantic partners wont care about me as much as I care about them .82
I often worry that my partner doesnt really love me .78
I often wish that my partners feelings for me were as strong as my feelings for him or her .74
When my partner is out of sight, I worry that he or she might become interested in someone else .72
When I show my feelings for romantic partners, Im afraid they wont feel the same way about me .66
I worry a lot about relationships .65
I find that my partner(s) dont want to get as close as I would like .65
My desire to be very close sometimes scares people away .63
I worry that I wont measure up to other people .62
It makes me mad that I dont get the affection and support I need from my partner .60
I rarely worry about my partner leaving me (r) .57
My romantic partner makes me doubt myself .56
Im afraid that once a romantic partner gets to know me, he or she wont like who I really am .52
I do not often worry about being abandoned (r) .52
Sometimes romantic partners change their feelings about me for no apparent reason .50
My partner only seems to notice me when Im angry .49

NOTE: (r) indicates that an item is reverse scored. Item loadings of less than .30 are not reported. RQ = Relationship Questionnaire; ECR-R = Re-
vised Experiences in Close Relationships questionnaire.

Overview of the models. The 3-week test-retest stability of a single indicator of attachment-related anxiety and
the RQ and ECR-R measures of anxiety and avoidance avoidance, RQ measures were entered as observed
were examined by estimating a path between Time 1 and variables.
Time 2 measures of the same variables. Latent indicators Stability of the ECR-R. Analyses of the initial model in
of ECR-R measures of anxiety and avoidance at each which Time 1 latent ECR-R avoidance ( = .90, R2 = .84)
time period were assessed using three parcels of six ran- and anxiety ( = .92, R2 = .85) loaded on their respective
domly selected items each (see Bandalos & Finney, 2001, Time 2 measures indicated high levels of stability. Time 1
for a discussion of item parceling). The estimation of RQ measures were also included in this model but did
weighed latent variables using structural equation mod- not load on any other variables.
eling has considerable advantages over the use of the Hu and Bentler (1999) argued that it is important to
observed means as it reduces the effects of measurement consider both the Standardized Root Mean square
error, thus providing a more accurate appraisal of the Residual (SRMR; a residual-based fit index) and one or
underlying constructs. Given that the RQ provided only more indexes of comparative fit, such as the Root Mean

Square Error of Approximation (RMSEA), Goodness of

Fit Index (GFI), Comparative Fit Index (CFI), or Non-
Normed Fit Index (NNFI), when considering the overall
adequacy of a model. GFI, CFI, and NNFI indices of
more than .95 and RMSEA and SRMR values of less than
.06 and .08, respectively, are indicative of good fitting
models (Hu & Bentler, 1999). According to these crite-
ria, the initial (Step 1) model provided an acceptable fit
to the data, 2(71, n = 172) = 100.47, GFI = .92, NNFI =
.99, CFI = .99, SRMR = .03, RMSEA = .05.
RQ measures of anxiety and avoidance were then
specified as additional predictors of their respective
Time 2 ECR-R counterparts (see Figure 1). Consistent Figure 1 Model examining the stability of the ECR-R measures of
anxiety and avoidance for 3 weeks.
with predictions, Time 1 RQ measures failed to predict NOTE: For simplicity, manifest variables and the paths from manifest
additional variance in ECR-R measures of attachment, to latent variables are not shown. Dashed lines indicate nonsignificant
R2 < .01. This revised (Step 2) model, in which Time 1 paths. Standard errors are displayed in parentheses. ECR-R = Revised
Experiences in Close Relationships questionnaire; RQ = Relationship
RQ measures were included as additional predictors, Questionnaire.
provided fit indices that were comparable with the ear- **p < .01.
lier model at Step 1, 2(69, n = 172) = 99.44, GFI = .92,
NNFI = .99, CFI = .99, SRMR = .03, RMSEA = .05. The two
models were not significantly different according to a
chi-square difference test, 2d.ff(2) = 1.03, p > .05.
Stability of the RQ. A similar series of analyses were con-
ducted examining the stability of the RQ. Analyses of the
model in which Time 1 RQ measures of avoidance ( =
.71, p < .01, R2 = .50) and anxiety ( = .67, p < .01, R2 = .45)
loaded on their respective Time 2 measures indicated
low to moderate levels of stability.1 Time 1 latent ECR-R
anxiety and avoidance were also included in this model
but did not load on any other constructs. Fit indices for
this initial (Step 1) model approached acceptable levels,
2(31, n = 172) = 62.56, GFI = .93, NNFI = .98, CFI = .98,
SRMR = .07, RMSEA = .08. Figure 2 Model examining the stability of the RQ measures of anxi-
ety and avoidance for 3 weeks.
A second model (displayed in Figure 2), in which NOTE: For simplicity, manifest variables and the paths from manifest
Time 1 latent ECR-R anxiety and avoidance were allowed to latent variables are not shown. Standard errors are displayed in pa-
to load on their respective Time 2 RQ counterparts, was rentheses. RQ = Relationship Questionnaire; ECR-R = Revised Experi-
ences in Close Relationships questionnaire.
then examined (Step 2). As shown in Figure 2, Time 1 **p < .01.
latent ECR-R anxiety and avoidance predicted 5% and
6% of the variance in their respective RQ measures in fit of either the ECR-R (Figure 1) or RQ (Figure 2) mod-
addition to that already predicted by the RQ at Time 1. els. Alternative models in which Time 1 measures of
This revised model provided an excellent fit to the data, anxiety loaded on Time 2 avoidance and vice versa were
2(29, n = 172) = 28.99, GFI = .97, NNFI = 1.00, CFI = 1.00, also examined, but they provided markedly worse fit
SRMR = .03, RMSEA < .01. A chi-square difference test
indices than the hypothesized models in all cases
further indicated that this revised model fit the data sig-
(SRMRs > .15).
nificantly better than the first model, in which only RQ
measures were allowed to load on their respective Time Discussion
2 RQ counterparts, 2d.ff(2) = 33.57, p < .05.
For simplicity, the manifest variables and paths from Consistent with previous research, these results indi-
latent to manifest indicators are not shown in Figures 1 cated that the ECR-R and RQ assessed the same pair of
and 2. However, all paths from latent to manifest ECR-R attachment dimensions (e.g., Griffin & Bartholomew,
indicators in both of these models were highly signifi- 1994b). Consistent with Sibley and Liu (2004), both the
cant (s > .85, zs > 18.00). Post hoc model modification anxiety and avoidance subscales of the ECR-R displayed
indices using the Lagrange multiplier indicated that between 84% and 85% shared variance (equivalent to rs
there were no other paths that would have improved the of .90 to .92) for a 3-week period. RQ measures of these

same constructs displayed roughly two thirds this level of Procedure and materials. Participants completed ECR-R
stability, that is, 50% shared variance (equivalent to rs of measures of romantic attachment anxiety ( = .93, M =
.70) for this same period. These estimates of the RQs sta- 2.06, SD = 1.00) and avoidance ( = .94, M = 1.95, SD =
bility are similar to those of other self-report measures .99; Fraley et al., 2000), which followed the procedures
assessed across similar time frames (rs between .60 and outlined in Study 1.
.70; e.g., Davila, Karney, & Bradbury, 1999; J. A. Feeney, Results
Alexander, Noller, & Hohaus, 2003; Lopez & Gormley,
2002). The ECR-R in contrast displayed more stable esti- CFA was used to examine whether the ECR-R ade-
mates of global romantic attachment than those pro- quately fit the hypothesized two-factor solution repre-
vided by other self-report measures assessed across com- senting attachment-related anxiety and avoidance.
parable time periods (cf. Davila & Sargent, 2003). These analyses used the partial disaggregation proce-
The ECR-R stability estimates reported here and by dure outlined by Bagozzi and Heatherton (1994), in
Sibley and Liu (2004) were comparable to those esti- which items assessing anxiety and avoidance were each
mated by Fraley et al. (2000) under simulated conditions parceled into six groups of three randomly selected items.
in which the underlying constructs were perfectly stable The hypothesized two-factor solution in which the six
parcels assessing avoidance loaded on one latent factor
(equivalent to rs of .94 to .95). Thus, although it is possi-
(s > .84, zs > 22.29) and the six parcels assessing anxiety
ble that some of the instability observed in the ECR-R
loaded on a second latent factor (s > .78, zs > 19.90) pro-
across this 3-week period may have been caused by mean-
vided an excellent fit to the data, 2(53, n = 478) = 142.26,
ingful change in working models of attachment, we sus-
GFI = .95, NNFI = .98, CFI = .98, SRMR = .04, RMSEA =
pect that the majority of the instability observed here was
.06. Anxiety and avoidance were moderately positively
most likely caused by measurement error akin to that
correlated ( = .45, z = 11.35).
demonstrated in Fraley et al.s simulation analyses. An alternative single-factor solution in which parcels
assessing anxiety and avoidance loaded on a single latent
STUDY 2 factor was also considered, 2(54, n = 478) = 3,623.12,
GFI = .44, NNFI = .53, CFI = .61, SRMR = .23, RMSEA =
Study 2 sought to elaborate on these results using
.37. This model provided a significantly poorer fit than
CFA. Relatively few studies have examined the factor
the hypothesized two-factor solution, 2d.ff(1) = 3,480.86,
structure of earlier attachment scales using CFA, which
p < .001.
formally tests how well the data fit a hypothesized factor
structure (e.g., Baeckstroem & Holmes, 2001; Discussion
Lafontaine & Lussier, 2003; Sanford, 1997). As Kurdek
These results provide important additional informa-
(2002) recently commented, for example, although tion on the psychometric properties of the ECR-R.
attachment style variables assessed by multiple items have Together with Study 1 and previous research by Sibley
appeared with increasing frequency in the close relation- and Liu (2004), they show that the ECR-R accurately fits
ships literature, their psychometric properties have not the hypothesized two-factor solution representing
been extensively critically evaluated (pp. 827-828). dimensions of attachment anxiety and avoidance.
Such comments are particularly true of the ECR-R.
We are aware of only one study previously assessing the STUDY 3
factor structure of the ECR-R using techniques such as
CFA (i.e., Sibley & Liu, 2004). This issue is especially per- Study 3 examined the proportions of variance in gen-
tinent given the relatively lenient item loading criterion eral and attachment-specific emotions in social interac-
used by Fraley et al. (2000) when selecting the initial tions involving a romantic partner, family member, or
ECR-R item pool and the use of item response theory close friend predicted by ECR-R measures of romantic
attachment. Event sampling methods, such as Wheeler
analysis, which assessed anxiety and avoidance as sepa-
and Nezleks (1977) social interaction diary, provide a
rate unidimensional constructs when selecting the final
relatively accurate way of assessing a range of emotions
scale items.
in social interactions across a broad range of naturally
Method occurring contexts while avoiding retrospective biases
inherent in a variety of other self-report designs (Reis &
Participants. Participants were 478 undergraduate stu- Gable, 2000). Diary methods of this type thus provide
dents (70% female, mean age = 20) who participated for one reliable external criterion that may be used to assess
partial course credit. Forty-five percent of participants the convergent and discriminant validity of measures of
were involved in a romantic relationship, and an addi- attachment referring to a given relationship domain
tional 5% were married. (Tidwell, Reis, & Shaver, 1996).

Attachment-related anxiety and avoidance have been ( = .94, M = 1.75, SD = 1.01) along with the RQ
shown to predict a range of emotions and subjective per- (Bartholomew & Horowitz, 1991), as outlined in Study 1.
ceptions of others measured using social interaction These measures were standardized as the ECR-R and RQ
diary procedures, such as intimacy and overall percep- were scored on different metrics.
tions of interaction quality (e.g., Bradford, Feeney, & As a validation check, two semantic differentials simi-
Campbell, 2002; J. A. Feeney, 2002b; Pierce & Lydon, lar to those used in the social interaction diary to assess
2001). These effects are relatively consistent across vari- state-based attachment anxiety and avoidance were also
ous measures of attachment, such as the RQ and ECR. included in the retrospective survey. These two items
Pierce and Lydon (2001), for example, reported that referred to the overall perception of ones relationships
global and relationship-specific RQ measures of attach- (similar to the items included in the ECR-R and RQ),
ment anxiety and avoidance (referred to in their whereas the diary versions of these same items referred
research as models of the self and other) predicted 18% to ones perceptions of a given social interaction (see
and 17% of the variance in experiences of social interac- below for details). The general-worded version of the
tions with close others, respectively. Similar trends have item assessing anxiety ranged from 0 = I am not worried
also been observed using less reliable and precise cate- about my romantic partners feelings toward me to 6 = I worry
gorical measures of attachment style (e.g., Kafetsios & about my romantic partners feelings toward me. The general-
Nezlek, 2002; Pietromonaco & Feldman Barrett, 1997; worded item assessing avoidance was anchored by the
Tidwell et al., 1996; see Fraley & Waller, 1998, for critique endpoints I am comfortable with the level of emotional close-
of earlier categorical measures of attachment). ness with my romantic partner (0) and I am uncomfortable
Study 3 examined three diary ratings: a standard mea- with the level of emotional closeness with my romantic partner
sure of enjoyment widely used in previous diary research (6). These items were based on published definitions of
(e.g., Tidwell et al., 1996) and two newly developed items attachment-related anxiety and avoidance (e.g.,
that assess state-based levels of attachment-related anxi- Bartholomew & Horowitz, 1991; Brennan et al., 1998;
ety and avoidance experienced during social interac- Mikulincer & Shaver, 2003).
tions. It was first hypothesized that ECR-R measures of
romantic attachment anxiety and avoidance would pre- Social interaction diary. Immediately following comple-
dict a substantial portion of the variance in respective tion of these measures, participants were administered a
levels of anxiety- and avoidance-related emotions experi- variant of Wheeler and Nezleks (1977) social interac-
enced during social interaction with a romantic partner. tion diary, which they completed for the next 2 weeks.
Given that the ECR-R was worded to refer to attachment Consistent with Wheeler and Nezlek (1977), partici-
representations of the romantic relationship domain, it pants were asked to complete a diary entry for every
was further hypothesized that ECR-R measures would be social interaction 10 min or longer in length. A detailed
only weakly, if not entirely unrelated, to comparable tutorial outlined instructions for completing the diary
diary measures of participants social interactions with accurately. The importance of updating the diary as
their most frequently recoded family member and close often as possible with a minimum of two or three times a
friend. Failure to support this hypothesis could raise day to improve the accuracy of ratings was stressed. Par-
concerns regarding the discriminant validity of the ECR- ticipants were also contacted after the 1st week to assess
R as a measure of ones models of romantic attachment. their performance and answer any additional queries.
Consistent with Study 1, RQ measures of romantic anxi-
Each social interaction diary record form contained
ety and avoidance were also included in these analyses to
the standard descriptive measures outlined by Wheeler
provide a baseline estimate of the variance predicted by
and Nezlek (1977; e.g., date, time, gender, and initials of
this shorter and more easily administered measure of
each interactant). Participants also indicated their rela-
tionship to the interactants (e.g., romantic partner, fam-
Method ily member, or friend). If the interaction was with two or
more people, then participants were instructed to list
PARTICIPANTS the interactants in order according to the amount of
Participants were 82 (74% female; mean age = 21) time that they spent with each person during the interac-
undergraduate students involved in a romantic relation- tion. If a person fulfilled multiple relationship roles,
ship who participated for partial course credit. then participants were instructed to code the relation-
ship as the more interpersonally close of the two options
(e.g., if a person fulfilled the roles of both a romantic
Retrospective Survey. Participants first completed the partner and a friend or family member, then they were
ECR-R (Fraley et al., 2000) measures of romantic attach- instructed to code them as a romantic partner). The
ment avoidance ( = .91, M = 2.03, SD = 1.16) and anxiety friend and family member most frequently recorded by

each participant were identified using a unique set of Bryk, 2002). As Nezlek (2003) noted, in one sense, HLM
initials recorded for each interactant. may be thought of as calculating a separate regression
Consistent with the general-worded versions of the slope for each persons social interactions with their
semantic differentials described earlier, social interac- romantic partner (i.e., within persons) and then treating
tion anxiety was rated on a scale ranging from 1 = I was these slopes as the dependent variable in comparisons
not worried about this persons feelings toward me to 7 = I wor- across persons. However, HLM is superior to traditional
ried about this persons feelings toward me. Avoidance during ordinary least squaresbased analyses of this type, as it
interactions was anchored by the endpoints I was comfort- provides methods for simultaneously modeling the
able with the level of emotional closeness (1) and I was uncom- error involved with sampling observations at multiple
fortable with the level of emotional closeness (7). Consistent levels, that is, both the between-person error (Level 2)
with Wheeler and Nezlek (1977), enjoyment was and the betweensocial interaction error (Level 1; see
anchored by the endpoints (1) very unpleasant to (7) very Raudenbush & Bryk, 2002, for technical discussion of
pleasant. These items referred to the primary interactant this issue).
of each interaction. Higher scores indicated higher lev-
els of anxiety, avoidance, and enjoyment, respectively.
As a validity check, all participants were contacted by Preliminary analyses. ECR-R measures of romantic
an independent research assistant 1 month after com- attachment anxiety and avoidance were strongly posi-
pleting the diary and personally handed a declaration tively correlated, r(80) = .51, p < .001. RQ measures of
that they were asked to sign and post back to the anxiety and avoidance in contrast were orthogonal,
researchers using a postage paid envelope if and only if r(80) = .04, p = .80. ECR-R and RQ measures of anxiety
they had maintained the social interaction diary regu- were moderately positively correlated, r(80) = .69, p <
larly (i.e., a minimum of once per day) and accurately .001, as were measures of avoidance, r(80) = .45, p < .001.
(i.e., made sure not to miss any interactions). The decla- Validation analyses were performed assessing the
ration stressed that the researchers had absolutely no relationship between the ECR-R items and two semantic
interest in identifying individuals and simply wanted to differentials similar to those used in the diary that were
make sure that the data were as accurate as possible for reworded to refer to overall relationship perceptions.
research purposes. Thus, if a signed declaration was not The ECR-R and RQ items were entered along with the
received, their data would be excluded from all analyses. general-worded versions of these two items into an
Eleven people (4 male and 7 female) failed to return this exploratory factor analysis with direct oblimin rotation.
declaration and were therefore removed from the sam- The results supported a two factor solution, with eigen-
ple (this yielded the final sample of 82 participants on values and ECR-R and RQ item loadings similar to those
which analyses were based). reported in Study 1. The semantic differential assessing
anxiety loaded at .86 on the ECR-R and RQ anxiety factor
and at .01 on the avoidance factor. The semantic differen-
Of the total 6,347 social interactions recorded by the tial assessing avoidance in contrast loaded at .62 on the
82 participants included in the study, a romantic partner ECR-R and RQ avoidance factor and at .12 on the anxiety
was coded as the primary interactant in 1,954. Partici- factor. These results suggest that the semantic differen-
pants most frequently recorded friend was coded as the tial items used in the diary assessed the same dimensions
primary interactant in 587 interactions, and their most of attachment measured by the ECR-R and RQ.
frequently recorded family member was coded as the Overview of the model. On average, each participant
primary interactant in 512. The data therefore consisted recorded 23.83 social interactions in which their roman-
of the 3,053 social interactions in which one of these tic partner was the primary interactant (77% of these
three persons was recorded as the primary interactant. were dyadic interactions; the remaining 23% were group
These data may be conceived of as hierarchical or multi- interactions that involved nonromantic others). Partici-
level in nature. The first (i.e., lower) level comprised the pants also reported an average of 6.24 and 7.16 interac-
multiple social interactions with a romantic partner, tions with their most frequently recorded family mem-
family member, or friend recorded by each participant, ber and close friend (53% and 62% of these interactions
whereas the second (i.e., higher) level referred to each were dyadic, respectively). Eighty-nine percent of partic-
participant. In this sense, social interactions (Level 1) ipants reported that their most frequently recorded
were nested within persons (Level 2). friend was of the same gender. The majority (82%) of
Current research suggests that such data structures participants reported that their most frequently
may be most appropriately analyzed using Multilevel recorded family member was a parent.
Random Coefficient Modelling, often referred to as Diary ratings of interactions with each of these three
hierarchical linear modeling (HLM; Raudenbush & persons were treated as Level 1 (social interaction level)

variables that were nested within participants. ECR-R HLM analyses. Initial analyses of the baseline intercept
and RQ measures of participants global romantic model (yij = 0j + rij) showed that 32% and 27% of the vari-
attachment-related anxiety and avoidance were in turn ance in diary ratings of anxiety and avoidance was at the
treated as Level 2 (person-level) variables. Following between-person level, respectively. In contrast, only 18%
Nezlek (2001, 2003), interactions with each of these of the variance in enjoyment in social interaction was
three persons were represented by three dummy-coded attributable to between-person differences. These
(0, 1) variables, each of which indicated the presence or results suggest that attachment-related emotions experi-
absence of one of the three (mutually exclusive) catego- enced during social interactions with well-known others
ries. Following HLM conventions, the interaction-level may be more trait like than more general positive
(Level 1), no intercept model may thus be expressed as emotions, such as enjoyment.
follows: The coefficients from Equation 2.1 and the cumula-
tive variance predicted at Steps 1 and 2 for social interac-
yij = 1j (romantic partner) + 2j (family member) + tions with a romantic partner, family member, and close
3j (close friend) + rij (1.0) friend are presented in Table 2. These coefficients were
functionally equivalent to unstandardized regression
coefficients and may be interpreted along similar lines.
where yij was the social interaction diary rating of interest As shown in Table 2, the RQ measures of romantic
(subscripted i; i.e., anxiety, avoidance, enjoyment) by attachment (entered at Step 1) accounted for 10% to
each participant (subscripted j), and rij represented er- 15% of the between-person variance in feelings of avoid-
ror. 1j, 2j, and 3j were random coefficients represent- ance, anxiety, and enjoyment experienced during social
ing the mean of yij across social interactions with each of interactions with a romantic partner.2 The ECR-R mea-
the three different persons. sures of romantic attachment accounted for an addi-
The relationship between Level 1 measures of anxi- tional 15% to 20% of the variance in these respective
ety, avoidance, and enjoyment in social interactions with measures when entered at Step 2. As expected, ECR-R
a romantic partner, family member, or close friend and anxiety predicted attachment-related anxiety experi-
Level 2 RQ and ECR-R measures of participants global enced during social interactions ( = .46). ECR-R avoid-
romantic anxiety and avoidance was assessed using two ance predicted attachment-related avoidance experi-
nested models. This procedure is similar to standard enced during social interactions at a comparable
hierarchical multiple regression involving multiple magnitude ( = .51). Enjoyment experienced during
steps. RQ measures of attachment anxiety and avoidance social interactions was also predicted primarily by ECR-R
were entered at Step 1. The Level 2 ECR-R measures of avoidance ( = .44). Across all three of these variables,
these same constructs were then entered at Step 2. The the variance explained by the RQ at Step 1 was also
Step 1 and Step 2 models for the assessment of interac- explained entirely by the ECR-R. Overall then, the ECR-
tions involving a romantic partner, for example, may be R accounted for a sizable portion of the between-person
expressed as follows: variation in the quality of social interactions with ones
romantic partner.
Step 1 1j = 00 + 01(RQ anxiety) + The predictive utility of the ECR-R and RQ were also
02 (RQ avoidance) + u0j (2.0) examined in social interactions involving a family mem-
ber and close friend. As shown in Table 2, the RQ pre-
Step 2 1j = 00 + 01 (RQ anxiety) + dicted between 5% and 10% of the variance in diary rat-
02 (RQ avoidance) + 03 (ECR-R anxiety) + ings of avoidance, anxiety, and enjoyment experienced
04 (ECR-R avoidance) + u0j (2.1) during social interactions with a family member. The
ECR-R predicted between 0% and 10% additional vari-
ance when entered at Step 2. The variance in ratings of
where 1j was the random coefficient representing the social interactions with a family member was also
mean of yij (i.e., diary ratings of anxiety, avoidance, or en- explained by the RQ at Step 1 and was explained entirely
joyment) in interactions with ones romantic partner, 00 by the ECR-R.
represented the intercept, 01 represented the effect of The RQ measure of romantic anxiety also predicted
RQ anxiety, 02 represented the effect of RQ avoidance, roughly 14% of the variance in ratings of social interac-
03 represented the effect of ECR-R anxiety, 04 repre- tions with a close friend. The ECR-R, however, failed to
sented the effect of ECR-R avoidance, and u1j repre- predict additional variance in this case and tended to
sented error. Identical Level 2 models examined the ef- account for slightly less variance in ratings of both anxi-
fects of the ECR-R and RQ in social interactions with a ety and enjoyment experienced during such interac-
family member (2j) or close friend (3j). tions (across all three ratings, the ECR-R predicted less

TABLE 2: Coefficients and Cumulative Variance in Social Interaction Diary Ratings Explained by the RQ (Step 1) and ECR-R (Step 2) Measures
of Global Romantic Attachment (Study 3)

Avoidance Anxiety Enjoyment

Cumulative Cumulative Cumulative
a a a
Measure Coefficients Variance (%) Coefficients Variance (%) Coefficients Variance (%)

Social interactions with romantic partner

Step 1 RQ anxiety .04 .01 .02
Step 1 RQ avoidance .12 14.53 .14 21.47 .10 10.70
Step 2 ECR-R anxiety .09 .46** .11
Step 2 ECR-R avoidance .51** 32.51 .29 41.36 .44** 29.67
Social interactions with family member
Step 1 RQ anxiety .09 .02 .03
Step 1 RQ avoidance .39* 11.41 .20 5.88 .14 10.59
Step 2 ECR-R anxiety .20 .34* .09
Step 2 ECR-R avoidance .07 11.48 .02 18.28 .29 16.39
Social interactions with platonic friend
Step 1 RQ anxiety .01 .17 .08
Step 1 RQ avoidance .05 0.01 .09 14.10 .08 3.31
Step 2 ECR-R anxiety .02 .15 .12
Step 2 ECR-R avoidance .25 2.33 .03 13.82 .06 0.01

NOTE: Analyses were based on 3,053 observations collected from 82 participants. RQ = Relationship Questionnaire; ECR-R = Revised Experiences
in Close Relationships questionnaire.
a. coefficients displayed for Step 2.
*p < .05. **p < .001.

than 5% of the variance in social interactions with a times as much variance as it did in interactions with a
platonic friend). platonic friend.
The number of people present during each social Overall then, these results indicate that the ECR-R dis-
interaction was entered as an additional Level 1 predic- played suitable convergent and discriminant validity as a
tor but failed to alter any of the reported analyses. Simi- measure of attachment representations of the romantic
larly, the entry of friends gender (opposite or same) did relationship domain. This suggests that romantic attach-
not alter any of the reported analyses. ment, as operationalized by the ECR-R, reflects variation
in relationship-level models of romantic relationships
rather than variation in more global interpersonal dispo-
As expected, ECR-R measures of romantic attach- sitions that encompass other relationship types that
ment anxiety and avoidance predicted sizable portions often also constitute attachments (Trinke & Bartholo-
of the variance in respective diary ratings of anxiety and mew, 1997). Comparison of trends in the cumulative
avoidance experienced during social interactions with a variance predicted by the RQ in contrast suggested that
romantic partner (rs equivalent to roughly .50). Given versions of the RQ administered with the same instruc-
the ECR-Rs improved reliability and test information tions (i.e., to answer the questions with ones experi-
functions (Fraley et al., 2000; Sibley & Liu, 2004), it is not ences in romantic and love relationships in mind) may
surprising that this multi-item measure predicted nota- not discriminate between models of romantic and non-
bly more between-participant variance in social interac- romantic relationship domains to the same degree.
tion ratings involving a romantic partner (30% to 40%)
than the RQ (10% to 20%). Furthermore, ECR-R mea- GENERAL DISCUSSION
sures were only weakly, and for the most part
nonsignificantly, related to diary ratings of anxiety, There has been considerable debate during the past
avoidance, and enjoyment in social interactions with a decade regarding the ways in which self-report measures
family member or close friend. Comparison of the may most validly and reliably assess working models of
cumulative variance in ratings of social interactions attachment. The present research contributed to this lit-
across these three relational contexts further indicates erature by documenting the short-term temporal stabil-
that the ECR-R predicted more than twice as much vari- ity (Study 1), factor structure (Study 2), and convergent
ance in avoidance, anxiety, and enjoyment in social and discriminant validity (Study 3) of Fraley et al.s
interactions with a romantic partner as it did variance in (2000) recently developed ECR-R self-report measure of
interactions with a family member and more than 3 to 4 adult romantic attachment.

As various researchers have noted, ones choice of researchers using the RQ may be relatively confident
measurement instrument depends on numerous fac- that the significant effects observed under such condi-
tors, including brevity and convenience on one hand tions would also have been observed using more recent
and reliability and comprehensiveness on the other measures of attachment. As indicated in Study 3, the RQ
(e.g., Pierce & Lydon, 2001; Overall, Fletcher, & Friesen, may, however, attenuate estimates of effect size in some
2003). Our results indicated that the extensive multi- cases, thus increasing the chances that small but genuine
item ECR-R and briefer RQ assessed the same pair of effects may go unnoticed (i.e., the RQ may be more
working models. The ECR-R and RQ differ somewhat in prone to Type II than Type I errors, at least when the
their short-term temporal stability and convergent and attachment models and outcomes being considered are
discriminant validity, however. In closing, we therefore of similar domains). Most notably, the RQ may be consid-
seek to offer some practical guidelines for judging when ered a viable alternative to more recent multi-item mea-
highly reliable and precise measures of global romantic sures when time constraints, survey length, or item
attachment, such as the ECR-R, may be deemed neces- repetition are pressing concerns.
sary and where shorter, albeit slightly less reliable mea- One caveat worth noting is that the RQ was adminis-
sures, such as the RQ, may also be considered viable. tered prior to the ECR-R in Studies 1 and 3 to avoid
Our results indicate that the ECR-R is preferable in artifactually inflating the reliability of the RQ because of
conditions where one wishes to examine (a) subtle possible priming effects. Estimates of the stability and
attachment effects with limited power or relatively small predictive validity of the RQ should thus be of a compa-
expected effect sizes and/or (b) use analyses or designs rable magnitude to the majority of previous research, in
that may exacerbate potential measurement error. which it was the sole measure of attachment used (cf.
Researchers intending to examine longitudinal changes Griffin & Bartholomew, 1994a; Pierce & Lydon, 2001).
in romantic attachment, for instance, are particularly Furthermore, if priming attachment-related cognitions
encouraged to use the ECR-R as the scale was designed does indeed improve the validity of subsequently com-
explicitly to reduce error encountered in the longitudi- pleted items, that is, if participants do indeed experi-
nal designs (Fraley et al., 2000). Indeed, Study 1 demon- ence a warm-up phase when completing items assessing
strated that not only did the ECR-R display markedly attachment, then such effects should naturally be attenu-
more stable estimates of romantic attachment than the ated in multi-item scales, such as the ECR-R.3
RQ (85% vs. 50% shared variance), but the inclusion of In sum, then, it appears that the ECR-R provides one
Time 1 ECR-R measures significantly improved the sta- of, if not the, most appropriate self-report measure of
bility estimates of RQ measures by roughly 5%. This find- adult romantic attachment currently available. However,
ing indicates that a portion of the instability observed similarly comprehensive measures assessing different
when global models of romantic attachment were mea- aspects of peoples attachment representations (e.g.,
sured using the RQ may be due to measurement error their models of family and friends) or the superordinate
inherent in the RQ. Similarly, we recommend that models theorized by Collins and Read (1994) to apply
researchers wishing to examine the interactive (i.e., to a wide range of relationships and situations (p. 58)
moderated) effects of romantic attachment avoidance have yet to be developed. Until such a time, one of the
and anxiety should consider using the ECR-R where pos- greatest conceptual strengths of many earlier measures,
sible as analyses of this type are also known to exacerbate most notably the RQ, is that they may be easily reworded
potential measurement error (McClelland & Judd, to assess models of different persons or different rela-
1993). Caveats of this type should also be kept in mind tionship domains. When it comes to assessing models of
when drawing conclusions from previous research that romantic relationships, however, our results clearly indi-
assessed attachment using less reliable measures, such as cate that the ECR-R more accurately discriminates
the RQ, and incorporated designs or analyses that may between attachment-related emotions experienced in
exacerbate potential measurement error (e.g., Davila & interactions with romantic and nonromantic partners. It
Cobb, 2003; Pierce & Lydon, 2001). remains to be seen, however, whether the ECR-R may be
On the other hand, the RQ may be considered an suitably reworded to assess different aspects of
appropriate alternative when (a) it is administered con- (nonromantic) working models or whether entirely new
currently with other variables of interest, (b) one has a scales are necessary. Furthermore, it appears that the
clear expectation of the expected effect sizes (which ECR-R provides a moderately positively correlated mea-
should be moderate to large), and (c) the relationship sure of anxiety and avoidance, which may be somewhat
domain being assessed matches the outcome domain of inconsistent with recent models of attachment system
interest (e.g., if one is assessing models of romantic rela- dynamics that operationalize anxiety and avoidance as
tionships, one should focus on outcomes related to orthogonal cognitive subsystems (Mikulincer & Shaver,
romantic relationships). Our results indicate that 2003). We encourage future researchers to address these

issues when developing and validating similarly compre- difference and life stress models of attachment change. Personality
and Social Psychology Bulletin, 29, 859-870.
hensive measures of the various global and specific Davila, J., Karney, B. R., & Bradbury, T. N. (1999). Attachment change
models of different relationship types theorized to processes in the early years of marriage. Journal of Personality and
underlie cognitive representations of attachment. Social Psychology, 76, 783-802.
Davila, J., & Sargent, E. (2003). The meaning of life (events) predicts
changes in attachment security. Personality and Social Psychology
NOTES Bulletin, 29, 1385-1395.
Feeney, B. C., & Collins, N. L. (2001). Predictors of caregiving in adult
1. Test-retest correlations indicated that the secure (r = .55), preoc- intimate relationships: An attachment theoretical perspective.
cupied (r = .64), dismissing (r = .64), and fearful (r = .76) prototypes Journal of Personality and Social Psychology, 80, 972-994.
from which RQ indices of attachment-related anxiety and avoidance Feeney, J. A. (2002a). Attachment-related dynamics: What can we
were derived displayed comparable levels of stability. learn from self-reports of avoidance and anxiety? Attachment and
2. Additional analyses examining the predictive utility of the four Human Development, 4, 193-200.
RQ attachment prototypes indicated that on average, the four proto- Feeney, J. A. (2002b). Attachment, marital interaction, and relation-
types predicted comparable amounts of variance in diary ratings of ship satisfaction: A diary study. Personal Relationships, 9, 39-55.
social interactions to those predicted by RQ indices of attachment- Feeney, J. A., Alexander, R., Noller, P., & Hohaus, L. (2003). Attach-
related anxiety and avoidance. ment insecurity, depression, and the transition to parenthood.
3. To assess the possible effects of item order, each ECR-R item was Personal Relationships, 10, 475-493.
entered into a separate HLM analysis predicting diary ratings. Results Feeney, J. A., Noller, P., & Hanrahan, M. (1994). Assessing adult
indicated that the predictive utility of individual items assessing anxiety attachment: Developments in the conceptualization of security
and avoidance did not vary systematically according to the order in and insecurity. In M. B. Sperling & W. H. Berman (Eds.), Attach-
which participants completed the items. ment in adults: Clinical and developmental perspectives (pp. 128-152).
New York: Guilford.
Fraley, R. C. (2002). Attachment stability from infancy to adulthood:
REFERENCES Meta-analysis and dynamic modeling of developmental mecha-
Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Pat- nisms. Personality and Social Psychology Review, 6, 123-151.
terns of attachment: A psychological study of the strange situation. Fraley, R. C., & Shaver, P. R. (2000). Adult romantic attachment: The-
Hillsdale, NJ: Lawrence Erlbaum. oretical developments, emerging controversies, and unanswered
Baeckstroem, M., & Holmes, B. M. (2001). Measuring adult attach- questions. Review of General Psychology, 4, 132-154.
ment: A construct validation of two self-report instruments. Scan- Fraley, R. C., & Spieker, S. J. (2003). Are infant attachment patterns
dinavian Journal of Psychology, 42, 79-86. continuously or categorically distributed? A taxometric analysis of
Bagozzi, R. P., & Heatherton, T. F. (1994). A general approach to rep- strange situation behavior. Developmental Psychology, 39, 387-404.
resenting multifaceted personality constructs: Application to Fraley, R. C., & Waller, N. G. (1998). Adult attachment patterns: A test
state self-esteem. Structural Equation Modeling, 1, 35-67. of the typological model. In J. A. Simpson & W. S. Rholes (Eds.),
Baldwin, M. W., & Fehr, B. (1995). On the instability of attachment Attachment theory and close relationships (pp. 77-114). New York:
style ratings. Personal Relationships, 2, 247-261. Guilford.
Bandalos, D. L., & Finney, S. J. (2001). Item parceling issues in struc- Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An item response
tural equation modeling. In G. A. Marcoulides & R. E. theory analysis of self-report measures of adult attachment. Jour-
Schumacker (Eds.), New developments and techniques in structural nal of Personality and Social Psychology, 78, 350-365.
equation modeling (pp. 269-296). Mahwah, NJ: Lawrence Erlbaum. Griffin, D. W., & Bartholomew, K. (1994a). The metaphysics of mea-
Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among surement: The case of adult attachment. In K. Bartholomew & D.
young adults: A test of a four-category model. Journal of Personality Perlman (Eds.), Advances in personal relationships: Vol. 5. Attachment
and Social Psychology, 61, 226-244. processes in adulthood (pp. 17-52). London: Jessica Kingsley.
Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and Griffin, D. W., & Bartholomew, K. (1994b). Models of the self and
anger. New York: Basic Books. other: Fundamental dimensions underlying measures of adult
Bowlby, J. (1980). Attachment and loss: Vol. 3. Loss: Sadness and depres- attachment. Journal of Personality and Social Psychology, 67, 430-445.
sion. New York: Basic Books. Hazan, C., & Shaver, P. R. (1987). Romantic love conceptualized as an
Bowlby, J. (1982). Attachment and loss: Vol. 1. Attachment. New York: attachment process. Journal of Personality and Social Psychology, 59,
Basic Books. (Original work published 1969) 511-524.
Bradford, S. A., Feeney, J. A., & Campbell, L. (2002). Links between Hu, L.- T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in
attachment orientations and dispositional and diary-based mea- covariance structure analysis: Conventional criteria versus new
sures of disclosure in dating couples: A study of actor and partner alternatives. Structural Equation Modeling, 6, 1-55.
effects. Personal Relationships, 9, 491-506. Kafetsios, K., & Nezlek, J. B. (2002). Attachment styles in everyday
Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report mea- social interaction. European Journal of Social Psychology, 32, 719-735.
surement of adult attachment: An integrative overview. In J. A. Kurdek, L. A. (2002). On being insecure about the assessment of
Simpson & W. S. Rholes (Eds.), Attachment theory and close relation- attachment styles. Journal of Social and Personal Relationships, 19,
ships (pp. 46-76). New York: Guilford. 811-834.
Collins, N. L., & Read, S. J. (1990). Adult attachment, working mod- Lafontaine, M.- F., & Lussier, Y. (2003). Bidimensional structure of
els, and relationship quality in dating couples. Journal of Personality attachment in love: Anxiety over abandonment and avoidance of
and Social Psychology, 58, 644-663. intimacy. Canadian Journal of Behavioral Science, 35, 56-60.
Collins, N. L., & Read, S. J. (1994). Cognitive representations of Lopez, F. G., & Gormley, B. (2002). Stability and change in adult
attachment: The structure and function of working models. In K. attachment style over the first-year college transition: Relations to
Bartholomew & D. Perlman (Eds.), Advances in personal relation- self-confidence, coping, and distress patterns. Journal of Counseling
ships: Vol. 5. Attachment processes in adulthood (pp. 53-90). London: Psychology, 49, 355-364.
Jessica Kingsley. McClelland, G. H., & Judd, C. M. (1993). Statistical difficulties of
Crowell, J. A., Fraley, R. C., & Shaver, P. R. (1999). Measurement of detecting interactions and moderator effects. Psychological Bulle-
individual differences in adolescent and adult attachment. In J. tin, 114, 376-389.
Cassidy & P. R. Shaver (Eds.), Handbook of attachment: Theory, Mikulincer, M., & Shaver, P. R. (2003). The attachment behavioral sys-
research, and clinical applications (pp. 434-465). New York: Guilford. tem in adulthood: Activation, psychodynamics, and interpersonal
Davila, J., & Cobb, R. J. (2003). Predicting change in self-reported processes. In M. P. Zanna (Ed.), Advances in experimental social psy-
and interviewer-assessed adult attachment: Tests of the individual chology (Vol. 35, pp. 53-152). New York: Academic Press.

Nezlek, J. B. (2001). Multilevel random coefficient analyses of event Shaver, P. R., & Mikulincer, M. (2002). Attachment-related
and interval contingent data in social and personality psychology psychodynamics. Attachment and Human Development, 4, 133-161.
research. Personality and Social Psychology Bulletin, 27, 771-785. Sibley, C. G., & Liu, J. H. (2004). Short-term temporal stability and
Nezlek, J. B. (2003). Using multilevel random coefficient modelling factor structure of the Revised Experiences in Close Relationships
to analyze social interaction diary data. Journal of Social and Per- (ECR-R) measure of adult attachment. Personality and Individual
sonal Relationships, 20, 437-469. Differences, 36, 969-975.
Overall, N. C., Fletcher, G. J. O., & Friesen, M. D. (2003). Mapping the Simpson, J. A., Rholes, W. S., Campbell, L., & Wilson, C. L. (2003).
intimate relationship mind: Comparisons between three models Changes in attachment orientations across the transition to par-
of attachment representations. Personality and Social Psychology Bul- enthood. Journal of Experimental Social Psychology, 39, 317-331.
letin, 29, 1479-1493. Simpson, J. A., Rholes, W. S., & Nelligan, J. S., (1992). Support-seeking
Pierce, T., & Lydon, J. E. (2001). Global and specific relational mod- and support-giving within couples in an anxiety-provoking situa-
els in the experience of social interactions. Journal of Personality tion: The role of attachment styles. Journal of Personality and Social
and Social Psychology, 80, 613-631. Psychology, 62, 434-446.
Pietromonaco, P. R., & Feldman Barrett, L. (1997). Working models Simpson, J. A., Rholes, W. S., & Phillips, D. (1996). Conflict in close
of attachment and daily social interactions. Journal of Personality relationships: An attachment perspective. Journal of Personality and
and Social Psychology, 73, 1409-1423. Social Psychology, 71, 899-914.
Pistole, M. C. (1989). Attachment in adult romantic relationships: Tidwell, M.- C. O., Reis, H. T., & Shaver, P. R. (1996). Attachment,
Style of conflict resolution and relationship satisfaction. Journal of attractiveness, and social interaction: A diary study. Journal of Per-
Social and Personal Relationships, 6, 505-510. sonality and Social Psychology, 71, 729-745.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Trinke, S. J., & Bartholomew, K. (1997). Hierarchies of attachment
Applications to data analysis methods (2nd ed). Thousand Oaks, CA: relationships in young adulthood. Journal of Social and Personal
Sage. Relationships, 14, 603-625.
Reis, H. T., & Gable, S. L. (2000). Event-sampling and other methods Wheeler, L., & Nezlek, J. (1977). Sex differences in social participa-
for studying everyday experience. In H. T. Reis & C. M. Judd
tion. Journal of Personality and Social Psychology, 35, 742-754.
(Eds.), Handbook of research methods in social and personality psychol-
ogy (pp. 190-222). New York: Cambridge University Press.
Sanford, K. (1997). Two dimensions of adult attachment: Further val- Received September 16, 2004
idation. Journal of Social and Personal Relationships, 14, 133-143. Revision accepted March 1, 2005