Content uploaded by Henrik Daae Zachrisson
Author content
All content in this area was uploaded by Henrik Daae Zachrisson
Content may be subject to copyright.
This article was downloaded by: [Universitetsbiblioteket i Oslo]
On: 25 November 2011, At: 05:45
Publisher: Psychology Press
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,
UK
European Journal of
Developmental Psychology
Publication details, including instructions for authors
and subscription information:
http://www.tandfonline.com/loi/pedp20
Factor structure of the Child
Attachment Interview
Henrik Daae Zachrisson a b , Espen Røysamb a c , Brit
Oppedal a & Stuart T. Hauser a b
a Norwegian Institute of Public Health, Division of
Mental Health, Department of Child and Adolescent
Mental Health, Oslo, Norway
b Judge Baker Children's Center & Harvard Medical
School, Cambridge, MA, USA
c Department of Psychology, University of Oslo, Oslo,
Norway
Available online: 21 Nov 2011
To cite this article: Henrik Daae Zachrisson, Espen Røysamb, Brit Oppedal & Stuart T.
Hauser (2011): Factor structure of the Child Attachment Interview, European Journal of
Developmental Psychology, 8:6, 744-759
To link to this article: http://dx.doi.org/10.1080/17405629.2011.631293
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.tandfonline.com/page/terms-
and-conditions
This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan, sub-
licensing, systematic supply, or distribution in any form to anyone is expressly
forbidden.
The publisher does not give any warranty express or implied or make any
representation that the contents will be complete or accurate or up to
date. The accuracy of any instructions, formulae, and drug doses should be
independently verified with primary sources. The publisher shall not be liable
for any loss, actions, claims, proceedings, demand, or costs or damages
whatsoever or howsoever caused arising directly or indirectly in connection
with or arising out of the use of this material.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
Factor structure of the Child Attachment Interview
Henrik Daae Zachrisson
1,2
, Espen Røysamb
1,3
, Brit Oppedal
1
,
and Stuart T. Hauser
1,2
1
Norwegian Institute of Public Health, Division of Mental Health,
Department of Child and Adolescent Mental Health, Oslo, Norway
2
Judge Baker Children’s Center & Harvard Medical School, Cambridge,
MA, USA
3
Department of Psychology, University of Oslo, Oslo, Norway
The aim of this study was to specify and test the factor structure of the Child
Attachment Interview (CAI), by means of confirmatory factor analysis (CFA).
The CFA provides a continuous measure of attachment, and also an
examination of the construct validity of this continuous approach. Analyses
included 150 children aged 9–13, recruited from schools in Norway. The one
factor-model ‘‘Security–Dismissal’’, as well as a model including the
supplementary factor ‘‘Preoccupation–Idealization’’, fitted the data well.
Individuals’ scores based on these factors converged with the categorical
attachment classifications, but added more information about individual
differences within categories. The findings support the construct validity of this
continuous approach to the CAI.
Keywords: Attachment dimension; Child Attachment Interview; Confirmatory
factor analysis; Validation.
Methodologists have long argued against dichotomization of continuous
measures in psychology (e.g., DeCoster, Iselin, & Gallucci, 2009;
MacCallum, Zhang, Preacher, & Rucker, 2002). In attachment research,
well-established measures like the Infant Strange Situation (ISS; Ainsworth,
Blehar, Waters, & Wall, 1978) and the Adult Attachment Interview
Correspondence should be addressed to Henrik Daae Zachrisson, Norwegian Institute of
Public Health, Division of Mental Health, Department of Child and Adolescent Mental Health,
PO Box 4404 Nydalen, 0403 Oslo, Norway. E-mail: heza@fhi.no
Parts of the results were presented as Pecha Kucha at the Society for Research in Child
Development Biannual conference in Denver, CO, April 2009.
EUROPEAN JOURNAL OF DEVELOPMENTAL PSYCHOLOGY
2011, 8 (6), 744–759
Ó2011 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business
http://www.psypress.com/edp http://dx.doi.org/10.1080/17405629.2011.631293
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
(AAI; see Hesse, 2008, for a review) are rated by trained coders on a number
of continuous scales, ultimately arriving at a categorical classification of
attachment. Although researchers have suggested dimensional approaches
to both the ISS (e.g., Ainsworth et al., 1978; Cummings, 1990) and the AAI
(e.g., Kobak, Cole, Ferenzgillies, Fleming, & Gamble, 1993), continuous
approaches to attachment have been most common among the question-
naire-based adult attachment measures (Crowell, Fraley, & Shaver, 2008).
Advances in the measurement of attachment in middle childhood
(Dwyer, 2005), have included continuous approaches in questionnaires
(Kerns, Aspelmeier, Gentzler, & Grabill, 2001), and interviews (Shmueli-
Goetz, Target, Fonagy, & Datta, 2008; Steele & Steele, 2005). Although new
attachment measures have raised concerns about validation issues (e.g.,
Laible, 2005), advances in psychometrics and statistics have to a limited
extent been applied. We aimed to specify and test a measurement model of a
continuous approach to the CAI by means of a Confirmatory Factor
Analysis (Brown, 2006), and thereby test the construct validity of this
approach (John & Benet-Martinez, 2000).
The Child Attachment Interview
The CAI builds upon the AAI, but the interview and the coding manual is
designed to measure representations of current attachment relationships of
children aged 8–13. In the CAI quality of the discourse is considered to be
indicative of attachment representations. The interview is rated on a number
of subscales, allowing the coder to classify it into one of the attachment
categories secure, dismissing, and preoccupied. The secure pattern reflects
undistorted affective and cognitive information about the availability of the
attachment figures, and their capacity to provide safety and comfort (see
Crowell et al., 2008, for an overview of attachment patterns). The dismissing
and preoccupied patterns reflect representations of the attachment figures as
being unpredictable and/or unavailable, and strategies to cope with this
experience. This includes for the dismissing category idealization or
dismissal attachment figures in order to avoid experiences of them being
rejecting and unavailable, for the preoccupied category angry or passive
blame the attachment figures for their shortcomings. Classification into
the disorganized category (with no coherent strategy to cope with unpre-
dictable or unresponsive attachment figures) is based on a qualitative judge-
ment, and is not anchored in the rating scales. Further details about the
interview and the coding procedure are presented in the method section below.
Classification based on the CAI has been related to theoretically
predictable aspects of child functioning. They are overlapping with the
capacity for mentalizing (Humfress, O’Connor, Slaughter, Target, &
Fonagy, 2002), and the secure classification has been associated with
FACTOR STRUCTURE OF THE CAI 745
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
aspects of emotion regulation (Borelli, Crowley et al., 2010). Finally,
disorganized attachment has been associated with higher levels of self-
reported depressive symptoms and shyness, as well as parental reports of
social anxiety, inattention, and thought problems (Borelli, David, Crowley,
& Mayes, 2010).
Shmueli-Goetz et al. (2008) suggested that the CAI provides a continuous
measure of attachment as a supplement to the categories, however without
specifying how this should be constructed. In this paper, we will argue that a
CFA model based on the subscales of the CAI is a valuable approach to
construct this continuous measure.
Attachment dimensions
There are many reasons not to categorize a continuous measure. It leads to
‘‘loss of information about individual differences, loss of effect size and power,
the occurrence of spurious main effects and interactions, risks of overlooking
nonlinear effects, and problems in comparing and aggregating findings across
studies’’ (MacCallum et al., 2002, p. 29). DeCoster et al. (2009) demonstrated
that a continuous measure is preferable under most circumstances, and that
a categorical measure never performs substantially better. Consequently,
attachment research will benefit from taking a continuous approach.
The idea of attachment dimensions is not new. In the initial description of
the ISS, Ainsworth et al. (1978) described dimensions of security to
insecurity and avoidance to conflict. Cummings (1990) suggested one
dimension of felt security in the ISS. Another example is Kobak et al’s
(1993) Q-sort procedure for the AAI, including two dimensions; security to
insecurity/anxiety and hyperactivating insecurity to deactivating insecurity.
There is empirical support for a dimensional approach to the attachment
construct. Using a taxometric technique (Waller & Meehl, 1998), the latent
structure of attachment was found to be continuous, not categorical, when
measured in infants with the ISS (Fraley & Spieker, 2003a) and in adults
with the AAI (Roisman, Fraley, & Belsky, 2007).
A confirmatory factor model of the CAI
These taxometric studies (Fraley & Spieker, 2003a; Roisman et al., 2007)
found two dimensions underlying variation in attachment organization. In
infants, these were ‘‘Proximity-Seeking versus Avoidant Strategies’’, and
‘‘Angry and Resistant Strategies’’ (Fraley & Spieker, 2003a). The authors
argued that these dimensions remain throughout development (Fraley &
Spieker, 2003b). In adults, they found a similar dimension from security to
dismissal, and that there may be a second dimension reflecting active versus
passive forms of preoccupation (Roisman et al., 2007).
746 ZACHRISSON ET AL.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
Expanding on these conceptualizations, we hypothesize a two-dimen-
sional model, capturing the essence of the secure, dismissing, and
preoccupied categories. The first dimension is adapted from Fraley and
Spieker (2003a) and Kerns et al. (2001), ranging from high levels of security
to high levels of dismissal. The Security–Dismissal dimension reflects the
logic inherent in coding of the CAI (Shmueli-Goetz, Target, Fonagy, &
Datta, 2004), that an individual rated high on indicators of security is rated
low on indicators of dismissal. This ‘‘Security–Dismissal’’ factor reflects the
two most prevalent attachment categories as opposites on a continuum, and
may be considered a one-dimensional measure of attachment security in
middle childhood, because the preoccupied pattern is very uncommon in the
normal population (Shmueli-Goetz et al., 2008).
In order to account for the preoccupied pattern, a second factor could be
added as a supplement. Following models of attachment in both infants
(Fraley & Spieker, 2003b) and adults (Kobak et al., 1993), we hypothesized
a dimension ranging from high levels of preoccupation to high levels of
dismissal, thus differentiating between forms of insecurity. This ‘‘Preoccu-
pied–Dismissal’’ dimension is also based on the logic of CAI coding, where
the two insecure patterns are mutually exclusive. Furthermore, moderate to
high ratings on the ‘‘Preoccupied anger’’ scales are relatively independent of
ratings on the scales indicative of attachment security (Shmueli-Goetz et al.,
2004), and we therefore assume this factor to be uncorrelated with the
‘‘Security–Dismissal’’ factor. The two-dimensional model, and its hypothe-
sized relation to the attachment categories, is illustrated in Figure 1.
Figure 1. Theoretically derived dimensional model of the attachment construct to be tested with
Confirmatory Factor Analysis. The horizontal double-arrow illustrates the ‘‘Security–
Dismissal’’ dimension. The vertical double-arrow illustrates the ‘‘Preoccupied–Dismissal’’
dimension.
FACTOR STRUCTURE OF THE CAI 747
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
Inspired by Cummings (1990), we examined the convergence of the
dimensional and the categorical approach to measuring attachment.
The dimensional approach should maintain the information inherent in
the categories, and in addition provide information about differences
between individuals within the categories.
To summarize, the aim of this study was to specify and test two separate
factor models of the CAI, a one-dimensional and a two-dimensional. We
specified the one-dimensional model to distinguish between attachment
security and dismissal, the ‘‘Security–Dismissal’’ factor. We supplemented
this model with a second factor, distinguishing between attachment
preoccupation and dismissal, the ‘‘Preoccupied–Dismissal’’ factor. We also
examined whether the models converge with the traditional attachment
classifications.
METHOD
Participants and procedure
Participants were recruited from five public schools in the city of Bergen,
Norway, as part of the project ‘‘Youth, Culture, and Competence’’ (YCC)
at the Norwegian Institute of Public Health. Schools with larger proportions
of ethnic minority children were selected for the study. All students in 4th,
5th, 6th, and 7th grade (N¼496) and their mothers were invited to the YCC
in Bergen, through a letter to the parents. Consent forms were returned for
333 (67%) students. These students and parents were invited to complete
questionnaires, and the children were invited to an interview, including the
CAI, and of these 136 (41%) accepted. The interviewers were trained and
supervised by the first author, who is a clinical psychologist and certified
CAI coder. Compared to the total study sample, a higher proportion of
girls, w
2
(1) ¼4.92, p5.05, and children with immigrant backgrounds,
w
2
(1) ¼4.42, p5.05, participated in the second phase. The only socio-
economic indicator available for all participants in the study was children’s
perception of economic hardship in the family. This did not differ between
those attending and not attending the second phase. Interviews conducted as
part of the pilot study for YCC (n¼27) are also included in the analyses.
Children in the pilot data were on average one year younger than children in
the total sample (t¼73.37, p5.001), but did not otherwise differ on
demographic variables. Thirteen interviews could not be used due to
technical problems with the recorder, etc., or because some children with
immigrant backgrounds were considered by the coders insufficiently fluent
in Norwegian to respond adequately. The total number of participants
included in this study were 150 children (60% girls) of 9–13 years of age
(M¼11.7 years, SD ¼1.3 years). Forty-nine (33%) of these children
748 ZACHRISSON ET AL.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
reported a national background other than Norwegian (12 from Europe/
USA, 24 from the Middle East/Central Asia, 5 from East Asia, 6 from Latin
America, and 1 from West Africa). The highest level of education for either
of the parents was: 54 (36%) more than four years at college or university;
39 (26%) less than four years at college or university; 15 (10%) senior high
school; and 10 (7%) secondary school. Information about parents’ level of
education was not available for 32 (21%) of the children. Twenty-nine
(20%) children reported having divorced parents, of these, six lived in single-
parent homes. The study was carried out according to the directions of the
National Committees for Research Ethics in Norway (www.etikkom.no/
English).
Measures
The Child Attachment Interview (CAI; Shmueli-Goetz et al., 2004, 2008)
consists of an interview protocol and a manual for analysing the interview.
The interview takes about 30 minutes during which children (aged 7 to 13)
are asked about attachment-related experiences with attachment figures.
The interview protocol is directed toward general (semantic) and specific
(episodic) descriptions of the relationships with attachment figures. The
interview is videotaped and subsequently transcribed verbatim for coding.
Coding is done according to the manual (Shmueli-Goetz et al., 2004),
based on transcripts and video recordings of the interviews. The child’s
responses are rated on eleven subscales, each rated from 1–9 (low through
high). The subscales are selected to cover the main aspects of the attachment
construct. The rationale for each subscale is described by Shmueli-Goetz
et al. (2008). The CAI subscales are: Emotional openness; Balance of
references; Use of examples; Preoccupied anger with respect to mother and
father; Idealization with respect to mother and father; Dismissal with
respect to mother and father; Resolution of conflicts; and Overall coherence.
The coherence subscale is mainly a summary of other subscales, and can
therefore not be considered as a unique indicator of the attachment
construct. It is therefore not included in the analyses for the present study.
An interview is assigned to one of four attachment categories based on a
global judgement of interview, and informed by cut-off points on each of the
subscales (Shmueli-Goetz et al., 2004).
Shmueli-Goetz et al. (2008) demonstrated inter-rater reliability with a
median Intra Class Correlation (ICC) for three coders of 0.88 (ranging from
0.71 to 0.94, except for the subscale ‘‘Idealization of father’’ with an ICC of
0.38), and test–retest stability over three months had a median of r¼.63
(ranging from .55 to .90, except the subscales ‘‘Preoccupied anger with
father’’ with r¼.29, and ‘‘Idealization of father’’ with r¼.42). In spite of the
CAI-classifications being based on the child’s verbal response, neither verbal
FACTOR STRUCTURE OF THE CAI 749
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
IQ score nor expressive language score was related to attachment
classification, and CAI ratings corresponded to other attachment-related
measures (Shmueli-Goetz et al., 2008).
Both coders in this study were trained and certified as reliable at the Anna
Freud Centre, London. The CAI manual was under development by the time
of training, and to increase inter-rater reliability the coders incorporated
general descriptions from the manual into the specific criteria for rating the
subscales. Most importantly, a child speaking only about parents in terms of
what they can provide or do is rated on the dismissing subscale. The
supplements were approved by M. Target and P. Fonagy and have been
added to the CAI coding manual (personal communication, February 2007).
Details about the supplements can be obtained from the first author. Fifty-
two interviews used in the present study were coded by two coders after these
supplements were incorporated. The median ICC for the 11 subscales was
0.88 (range 0.77 to 0.97). There were no significant differences in the coders’
mean scores on any of the subscales. All analyses are based on the first
coder’s ratings of the entire data set. For the four attachment categories
(Secure, Dismissing, Preoccupied, and Disorganized), the inter-rater agree-
ment for the categories was 82% for mother and 78% for father.
Analysis
Missing data comprised less than 1% on all CAI subscales except the item
‘‘Resolution of conflict’’ where 4% were missing. (Missing data during the
rating procedure were due to too little information given in the interview.)
All missing data were input with Expectation-Maximization algorithm using
PASW 17.0 (SPSS, Inc., 2009). All items except the two ratings of
‘‘Preoccupied anger’’ with mother and father, respectively, were normally
distributed. The two ‘‘Preoccupied anger’’ items were highly skewed,
reflecting a rating of three or higher for only nine and eight children, for
mother and father respectively. Scale transformations were not successful.
Because the CAI coding manual (Shmueli-Goetz et al., 2004) is unspecific
with regard to the rating criteria of these subscales (M. Target, personal
communication, February 2007), the items ‘‘Preoccupied anger’’ with
mother and father respectively were dichotomized into one category of
lower than three, and one of three or higher.
CFA was performed with the Mplus version 6.1 (Muthen & Muthen,
2010a). For the first CFA, including only continuous factor indicators, we
used a maximum likelihood estimator, which is robust to non-normality
(MLR; Muthen & Muthen, 2010b). For the second CFA, combining
continuous and binary factor indicators, we used weighted least squares
mean and variance adjusted estimator (WLSMV; Muthen & Muthen,
2010b) estimating probit regressions for the categorical factor indicators,
750 ZACHRISSON ET AL.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
and linear regressions for the continuous factor indicators. WLSMV does
not assume multivariate normality, and has been found to perform well even
in smaller samples (Flora & Curran, 2004).
The fit of the models was assessed with the chi-square value, comparative
fit index (CFI), Tucker–Lewis index (TLI), and root mean square error of
approximation (RMSEA). An insignificant chi-square value, CFI/TFI
values above 0.95, and RMSEA below 0.06 indicate good model fit (Hu
& Bentler, 1999).
The subscales ‘‘Preoccupied anger’’, ‘‘Idealization’’, and ‘‘Dismissal’’ are
according to the CAI manual (Shmueli-Goetz et al., 2004) rated separately
for each parent, giving six subscales altogether. Because of the high
concordance between classification of mother and father, Shmueli-Goetz
and colleagues (2008) recommend focusing on an overall classification of
security when using the CAI. We followed this by including both mother
and father ratings as indicators of the latent subconstruct.
RESULTS
Descriptive statistics
Means and standard deviations for all subscales are presented in Table 1,
and the distribution of attachment classifications can be seen in Table 2.
There was a higher proportion of girls than boys rated in the secure
category, w
2
(1) ¼8.45, p5.01, but no differences with regard to immigrant
status or parental education level.
TABLE 1
Means, standard deviations, and Spearman–Brown inter-rater reliability for all subscales
of the Child Attachment Interview (inter-rater reliability estimates are based on a subset
of 52 interviews)
Subscale Mean SD Inter-rater reliability
Emotional openness 5.59 1.67 0.90
Balance 5.14 1.37 0.80
Use of examples 5.38 1.60 0.92
Preoccupied anger with mother 1.24 0.92 0.97
Preoccupied anger with father 1.19 0.84 0.92
Idealization of mother 2.70 1.63 0.70
Idealization of father 2.71 1.61 0.79
Distancing of mother 3.08 1.59 0.83
Distancing of father 3.15 1.63 0.86
Resolution of conflicts 5.02 1.50 0.88
Overall coherence 5.25 1.59 0.91
FACTOR STRUCTURE OF THE CAI 751
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
CFA models
The one-dimensional model of the CAI consisted of the ‘‘Security–Dismissal’’
factor. This factor was specified as being measured by the subscales ‘‘Emotional
openness’’, ‘‘Balance’’, ‘‘Use of examples’’, ‘‘Resolution of conflicts’’, ‘‘Idealiza-
tion with respect to mother and father’’, and ‘‘Dismissal with respect to mother
and father’’. The residuals of the two items ‘‘Idealization’’ with respect to mother
and father, the two items ‘‘Dismissing’’ with respect to mother and father were
allowed to correlate, because these pairs measure similar psychological processes.
With these two correlated residuals (idealization: r¼.52; dismissal: r¼.73), the
model fitted the data well, w
2
(18) ¼29.57, p¼.042 (TLI ¼.98, CFI ¼.99,
RMSEA ¼.065). All factor loadings were statistically significant (p5.05). Thus,
a single factor accounts for the covariance pattern in all CAI subscales except
those representing preoccupied anger. The model is depicted in Figure 2.
The two-dimensional model of the CAI consisted of the ‘‘Security–
Dismissal’’ factor as described above and the uncorrelated ‘‘Preoccupation–
Dismissal’’ factor. The ‘‘Preoccupation–Dismissal’’ factor was specified to
include the six subscales ‘‘Idealization with respect to mother and father’’,
‘‘Dismissing with respect to mother and father’’, and ‘‘Preoccupied anger
with respect to mother and father’’. This two-dimensional model fitted well,
w
2
(29) ¼39.49, p¼.09 (TLI ¼.96, CFI ¼.95, RMSEA ¼.049).
All factor loadings were significant, except the loadings of the subscales
‘‘Dismissal with respect to mother and father’’ on the ‘‘Preoccupation–
Dismissal’’ factor. Models with and without these paths did not differ
significantly. We therefore preferred the most parsimonious model (with the
loadings of the dismissal indicators on the second factor fixed to zero). The
final model therefore consisted of the factor ‘‘Preoccupation–Idealization’’ in
addition to the factor ‘‘Security–Dismissal’’. The fit indices for this modified
model were w
2
(31) ¼39.10, p¼.15 (TLI ¼.97, CFI ¼.96, RMSEA ¼.042).
The final model with factor loadings can be seen in Figure 3.
In order to examine the adequacy of the two-dimensional model compared
to the traditional attachment classifications, we plotted each individual’s
factor score in a graph indicating each individual’s attachment classification.
Figure 4 illustrates each individual’s placement in the two-dimensional space
TABLE 2
Distribution of attachment categories (missing categories are due to absence of this
specific attachment figure)
Dismissing n (%) Secure n (%)
Preoccupied
n(%)
Disorganized
n (%) Missing n (%)
Mother 47 (31.3) 95 (63.3) 5 (3.3) 1 (0.7) 2 (1.3)
Father 51 (34.0) 92 (61.3) 4 (2.7) 1 (0.7) 1 (0.7)
752 ZACHRISSON ET AL.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
created by the two factors ‘‘Security–Dismissal’’ and ‘‘Preoccupation–
Idealization’’, and indicates each individual’s attachment classification. The
graph illustrates that individual factor scores vary along continua, and at the
same time maintain the basic patterning of the attachment classifications.
Analyses of variance (ANOVA) showed significant differences in factor scores
between the three attachment categories for both factor one, F(2,
139) ¼112.7, p5.001, and factor two, F(2, 139) ¼20.29, p5.001.
We conducted post hoc power analyses utilizing Monte Carlo simula-
tions, as described by Brown (2006). Given the specified factor models and
parameter estimates, the sample size of 150 was sufficient for achieving
unbiased parameter estimates and standard errors, and achieving sufficient
coverage on all parameters.
DISCUSSION
The aim of this study was to specify and test a one-dimensional and a two-
dimensional CFA of the CAI. Applying a CFA to the study of construct
Figure 2. One factor CFA model of the Child Attachment Interview, with the ‘‘Security–Dismissal’’
factor and standardized regression coefficients. w2(18) ¼29.57, p¼.042; Tucker–Lewis index
(TLI) ¼.98; comparative fit index (CFI) ¼.99; root-mean-square error of approximation
(RMSEA) ¼.065. Abbreviations for the indicators are: Em ¼Emotional openness; Bal ¼
Balance of references; Ex ¼Use of examples; Res ¼Resolution of conflict; DsM ¼Distancing of
mother; DsF ¼Distancing of father; IdM ¼Idealization of mother; IdF ¼Idealization of father.
FACTOR STRUCTURE OF THE CAI 753
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
validity of an interview-based attachment measure may be considered a
logical next step in attachment research building on findings that attachment
should be considered a dimensionally rather than categorically structured
construct. Both the one-dimensional ‘‘Security–Dismissal’’ and the two-
dimensional model consisting of the ‘‘Security–Dismissal’’ factor and a
‘‘Preoccupation–Idealization’’ factor fitted the data well. These measure-
ment models provide support for an interpretation of the structure of the
CAI to reflect a dimensional approach to the attachment construct.
Comparisons of the individuals’ factor scores with the traditional
attachment classifications demonstrated a high overlap. As the factor scores
are distributed within each category, they are more sensitive to differences
between individuals within the categories.
Attachment categories: Distribution and demographics
The distribution of attachment patterns in our sample was in accordance
with the distribution found by Shmueli-Goetz et al. (2008) in the normal
population, as was our finding that attachment category was unrelated to
Figure 3. The final two-factor CFA model of the Child Attachment Interview consisting of the
uncorrelated factors ‘‘Security–Dismissal’’ and ‘‘Preoccupation–Idealization’’ with standardized
regression coefficients. w2(31) ¼39.10, p¼.15; Tucker–Lewis index (TLI) ¼.97; comparative fit
index (CFI) ¼.96; root-mean-square error of approximation (RMSEA) ¼.042. Abbreviations
in addition to those used in Figure 2 are: PaMCat* ¼Preoccupied anger with mother, coded
categorically; PaFCat* ¼Preoccupied anger with father, coded categorically.
754 ZACHRISSON ET AL.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
age and immigrant status. In contrast to this previous study, we found boys
to be more likely than girls to be categorized as insecure. Mayseless (2005)
has stated that although gender differences in attachment are uncommon in
infancy and at preschool age, there is evidence that in middle childhood and
adolescence, girls are more likely to be secure, and boys tend to have higher
levels of avoidance. This may reflect differences in attachment security or
gender-specific norms for behavioural expression (Mayseless, 2005). The
gender differences in our data may therefore not, at present, be considered
invalidating, but rather be a venue for further exploration.
The factor models
We tested both a one- and a two-dimensional model. To our knowledge,
more complex models, like three- or four-dimensional models, have not been
proposed, although Sroufe (2003) argues that such models may be more
precise. The structure of the dimensional models we tested was based on
those identified in other age groups. This is consistent with attachment
theory, assuming the same categorical system to be descriptive of individual
differences in attachment regardless of age group (Fraley & Spieker, 2003b).
The one-factor model ‘‘Security–Dismissal’’ resembles the first dimension
described by Roisman et al. (2007). This dimension differs from those
Figure 4. Convergence of the factor scores with the traditional attachment classifications. The
points represent each individual’s factor scores in the two-dimensional space created by the two
factors ‘‘Security–Dismissal’’ and ‘‘Preoccupation–Idealization’’. Each individual’s categorical
attachment classification is indicated by the signs: ¼secure; ~¼insecure dismissing;
¤¼insecure preoccupied. The dotted circle is the approximated border for the secure category.
The full circles are the approximate borders for the dismissing (large circle) and preoccupied
(small circle) categories.
FACTOR STRUCTURE OF THE CAI 755
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
presented by Cummings (1990) and Kobak and colleagues (1993), as these
comprised levels of security/insecurity. In contrast, the dimension we pro-
posed included one specific type of insecurity (dismissal). However, the other
insecure pattern (preoccupation) is very rare, with a prevalence of 3% in our
data and 4% reported by Shmueli-Goetz and colleagues (2008). Conse-
quently, insecurity equals dismissal for practical purposes in both of these
samples.
Idealization and dismissal represent one end of the Security–Dismissal
continuum. In this model, idealization and dismissal are considered to be
related, yet distinct psychological processes. This is reflected in the coding of
the AAI, where individuals who primarily derogate attachment figures are
assigned to a different subcategory of dismissal than those who idealize
(Hesse, 2008).
The other factor, ‘‘Preoccupation–Idealization’’, resembles the dimension
distinguishing between hyperactivating and deactivating strategies (Kobak
et al., 1993). This factor may be considered more provisional than the
‘‘Security–Dismissal’’ factor, because of low prevalence of this phenomenon
in our sample. Although we accounted for this, the model should be
replicated on samples with higher levels of preoccupation. Our analyses
supported the removal of the insignificant paths from the ‘‘Preoccupied–
Dismissal’’ factor to the two indicators of dismissal. Even though this was
ad hoc, it is conceptually meaningful. According to Shmueli-Goetz et al.
(2004), children who are rated as expressing preoccupied anger are likely to
derogate the attachment figure, and are therefore likely to be rated with
some dismissal (derogating both the relationship and the attachment figure).
The conceptual opposite of preoccupation rated in the CAI may therefore
be the idealizing aspect of dismissal rather than ‘‘Dismissal’’ as a whole.
Dimensional approach
Based on these two latent factors, each individual can be identified at one
point in a one-dimensional or a two-dimensional space. We compared each
individual’s factor score with each individual’s categorical attachment
classification, finding a dimensional system based on factor scores maintains
the basic patterning of the attachment categories. Yet, there was variability
within each of these groups as the individuals were spread out along the
continuum. The variability within each category is currently incorporated
into the classificatory procedures of both the ISS (e.g., B1, B2, etc.;
Ainsworth et al., 1978) and the AAI (F1, F2, etc.; see Hesse, 2008, for a
review). These subcategories do, however, carry the limitation of adding to
the number of categories, and thereby decreasing statistical power under
most circumstances. A dimensional approach maintains the added value of
more nuanced classifications, while avoiding such limitations.
756 ZACHRISSON ET AL.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
Limitations
A number of limitations apply to our study. A majority of the children
invited did not participate in the interview. Also, girls and children
with immigrant background were overrepresented among those inter-
viewed. Yet, the distribution within attachment categories in our
sample is similar to the non-clinical sample reported by Shmueli-Goetz
et al. (2008). Our sample was too small to examine measurement
invariance (Brown, 2006) across gender or ethnicity, which would have
been an appropriate way to examine the effect of response bias on our
results.
Other constrains are linked with the CAI subscales. The disorganized
category is not reflected in the CAI subscales. Until relevant subscales are
constructed this category cannot be examined through CFA. In our study,
such subscales would have been of limited value because only one child in
our sample was classified as disorganized. Moreover, the subscales designed
to measure preoccupation are in need of refinement (Shmueli-Goetz et al.,
2004). Currently, only preoccupied anger is rated, whereas both the CAI
coding manual, and the findings by Fraley and colleagues (Fraley &
Spieker, 2003a; Roisman & Fraley, 2007) suggest a more differentiated
subconstruct.
Our study was limited by the fact that all subscales are simultaneously
rated by one rater. Categorically oriented rating of the CAI is implicitly
based on the search for qualities in the interviews that fit the prototypes
(Dismissing, Secure, and Preoccupied). Despite the maintained focus on
subscale scores in our data, there is still a risk that a prototype-orientation
influenced the rating of the subscales. This may lead to an artificially high
coherence between the subscales.
Conclusion
In summary, our findings support the use of CFA as a way to approach
CAI as a continuous measure of attachment in middle childhood.
When comparing each individual’s factor score and attachment
classification, it was evident that the factors we specified maintained
the information inherent in the attachment categories. Together with
the findings of Shmueli-Goetz et al. (2008), these results offer
increased evidence for the CAI as a measure of attachment in middle
childhood.
Manuscript received 17 June 2010
Revised manuscript accepted 3 October 2011
First published online 18 November 2011
FACTOR STRUCTURE OF THE CAI 757
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
REFERENCES
Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of attachment: A
psychological study of the Strange Situation. Hillsdale, NJ: Lawrence Erlbaum Associates,
Inc.
Borelli, J. L., Crowley, M. J., David, D. H., Sbarra, D. A., Anderson, G. M., & Mayes, L. C.
(2010). Attachment and emotion in school-aged children. Emotion,10, 475–485.
Borelli, J. L., David, D. H., Crowley, M. J., & Mayes, L. C. (2010). Links between disorganized
attachment classification and clinical symptoms in school-aged children. Journal of Child
and Family Studies,19, 243–256.
Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York, NY: Guilford
Press.
Crowell, J., Fraley, R. C., & Shaver, P. R. (2008). Measurement of individual differences in
adolescent and adult attachment. In J. Cassidy & P. R. Shaver (Eds.), Handbook of
attachment. Theory, research, and clinical implications (pp. 599–635). New York, NY:
Guilford Press.
Cummings, E. M. (1990). Classification of attachment on a continuum of felt security:
Illustrations from the study of children of depressed parents. In M. Greenberg, D. Cicchetti,
& E. M. Cummings (Eds.), Attachment in the preschool years. Theory, research, and
intervention (pp. 311–338). Chicago, IL: University of Chicago Press.
DeCoster, J., Iselin, A.-M. R., & Gallucci, M. (2009). A conceptual and empirical examination
of justifications for dichotomization. Psychological Methods,14, 349–366.
Dwyer, K. M. (2005). The meaning and measurement of attachment in middle and late
childhood. Human Development,48, 155–182.
Flora, D. B., & Curran, P. J. (2004). An empirical evaluation of alternative methods of
estimation for confirmatory factor analysis with ordinal data. Psychological Methods,9,
466–491.
Fraley, R. C., & Spieker, S. J. (2003a). Are infant attachment patterns continuously or
categorically distributed? A taxometric analysis of strange situation behavior. Developmental
Psychology,39, 387–404.
Fraley, R. C., & Spieker, S. J. (2003b). What are the differences between dimensional and
categorical models of individual differences in attachment? Reply to Cassidy (2003),
Cummings (2003), Sroufe (2003), and Waters and Beauchaine (2003). Developmental
Psychology,39, 423–429.
Hesse, E. (2008). The Adult Attachment Interview. Protocol, method of analysis, and empirical
studies. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attachment. Theory, research, and
clinical applications (2nd ed., pp. 532–598). New York, NY: Guilford Press.
Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indices in covariance structure analysis:
Conventional criteria versus new alternatives. Structural Equation Modeling,6, 1–55.
Humfress, H., O’Connor, T. G., Slaughter, J., Target, M., & Fonagy, P. (2002). General and
relation-specific models of social cognition: explaining the overlap and discrepancies.
Journal of Child Psychology and Psychiatry,43, 873–883.
John, O. P., & Benet-Martinez, V. (2000). Measurement: Reliability, construct validation, and
scale construction. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in
social and personality psychology (pp. 339–369). Cambridge, UK: Cambridge University
Press.
Kerns, K. A., Aspelmeier, J. E., Gentzler, A. L., & Grabill, C. M. (2001). Parent–child
attachment and monitoring in middle childhood. Journal of Family Psychology,15, 69–81.
Kobak, R. R., Cole, H. E., Ferenzgillies, R., Fleming, W. S., & Gamble, W. (1993). Attachment
and emotion regulation during mother–teen problem-solving. A control-theory analysis.
Child Development,64, 231–245.
758 ZACHRISSON ET AL.
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011
Laible, D. (2005). Measuring attachment in middle childhood: Challenges and future directions.
Human Development,48, 183–187.
MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. (2002). On the practice of
dichotomization of quantitative variables. Psychological Methods,7, 19–40.
Mayseless, O. (2005). Ontogeny of attachment in middle childhood—Conceptualization of
normative changes. In K. A. Kerns & R. A. Richardson (Eds.), Attachment in middle
childhood (pp. 1–23). New York, NY: Guilford Press.
Muthen, L. K., & Muthen, B. O. (2010a). Mplus VERSION 6.1. [computer software]. Los
Angeles, CA: Muthen & Muthen.
Muthen, L. K., & Muthen, B. O. (2010b). Mplus user’s guide (6th ed.). Los Angeles, CA:
Muthen & Muthen.
Roisman, G. I., Fraley, R. C., & Belsky, J. (2007). A taxometric study of the Adult Attachment
Interview. Developmental Psychology,43, 675–686.
Shmueli-Goetz, Y., Target, M., Fonagy, P., & Datta, A. (2004). Child Attachment Interview
(CAI): Coding and classification manual Version V. The Sub-Department of Clinical Health
Psychology, University College London, UK. Unpublished manual.
Shmueli-Goetz, Y., Target, M., Fonagy, P., & Datta, A. (2008). The Child Attachment
Interview: A psychometric study of reliability and discriminant validity. Developmental
Psychology,44, 939–956.
SPSS, Inc. (2009). PASW statistics 17.0 Command Syntax Reference. Chicago, IL: SPSS, Inc.
Sroufe, L. A. (2003). Attachment categories as reflections of multiple dimensions: Comment on
Fraley and Spieker (2003). Developmental Psychology,39, 413–416.
Steele, H., & Steele, M. (2005). The construct of coherence as an indicator of attachment
security in middle childhood: The Friends and Family Interview. In K. A. Kerns & R. A.
Richardson (Eds.), Attachment in middle childhood (pp. 137–160). New York, NY: Guilford
Press.
Waller, N. G., & Meehl, P. E. (1998). Multivariate taxometric procedures: Distinguishing types
from continua. Newbury Park, CA: Sage.
FACTOR STRUCTURE OF THE CAI 759
Downloaded by [Universitetsbiblioteket i Oslo] at 05:45 25 November 2011