Essay Grade

Journal of Counseling Psychology
2004, Vol. 51, No. 2, 201–212
Copyright 2004 by the American Psychological Association
0022-0167/04/$12.00 DOI: 10.1037/0022-0167.51.2.201
Maladaptive Perfectionism as a Mediator and Moderator Between Adult
Attachment and Depressive Mood
Meifen Wei
Brent Mallinckrodt
Iowa State University
University of Missouri–Columbia
Daniel W. Russell and W. Todd Abraham
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Iowa State University
This study examined maladaptive perfectionism (concern over mistakes, doubts about one’s ability to
accomplish tasks, and failure to meet high standards) as both a mediator and a moderator between adult
attachment (anxiety and avoidance) and depressive mood (depression and hopelessness). Survey data
were collected from 310 undergraduates and analyzed using structural equation modeling (SEM)
methods. Results indicated that maladaptive perfectionism partially mediated the relationship between
attachment anxiety and depressive mood and fully mediated the relationship between attachment
avoidance and depressive mood. Bootstrap methods were used to assess the magnitude of the indirect
effects. Significant moderator effects were also found with SEM methods. The association between
attachment anxiety and depressive mood was stronger as perfectionism increased. Perfectionism was not
a significant moderator for attachment avoidance and depressive mood.
an insecure adult attachment orientation. By contrast, people with
low levels of attachment anxiety and avoidance have the capacity
for secure adult attachment, a positive sense of personal competence, and the ability to maintain supportive attachments (Brennan
et al., 1998; Lopez & Brennan, 2000; Mallinckrodt, 2000).
Previous empirical research has provided strong evidence for a
link between insecure attachment and various forms of psychological distress (for reviews, see Lopez & Brennan, 2000; Mikulincer
& Shaver, 2003). For example, relative to their secure counterparts, people with insecure attachment reported greater distress and
hostility during a laboratory problem-centered discussion (Simpson, Rholes, & Phillips, 1996), greater affective intensity and
emotionality in their daily life (Pietromonaco & Barrett, 1997),
more depressive symptoms (Roberts, Gotlib, & Kassel, 1996),
greater interpersonal problems (Mallinckrodt & Wei, 2003), and
more emotional distress (Collins, 1996). Thus, the link between
various forms of insecure attachment and indices of psychological
distress (e.g., depressive mood) has been fairly well established.
More recently, research linking attachment insecurity and distress
(e.g., depressive mood) has been shifting from an examination of
simple bivariate linear relationships to multivariate interactional
models that examine the roles of mediators and moderators of
these relationships (Collins, 1996; Lopez, Mitchell, & Gormley,
2002; Roberts et al., 1996; Wei, Heppner, & Mallinckrodt, 2003).
One example of this new emphasis on multivariate models is
recent research that has examined the relationships among attachment, perfectionism, and adjustment (Rice & Mirzadeh, 2000).
Perfectionism has been conceptualized as a multidimensional construct, with both adaptive and maladaptive aspects (Flett & Hewitt,
2002). Adaptive perfectionism involves setting high (but achievable) personal standards, a preference for order and organization,
a sense of self-satisfaction, a desire to excel, and a motivation to
Throughout the past decade, there has been a growing interest
among counseling psychologists in applying Bowlby’s (1973,
1980, 1988) attachment theory to understanding adult development
and the counseling process (Lopez, 1995; Lopez & Brennan, 2000;
Mallinckrodt, 2000). The initial formulations of adult attachment
posited four qualitative categories of attachment based on combinations of positive and negative working models of self and others
(e.g., Bartholomew & Horowitz, 1991). However, research has
failed to confirm the existence of qualitative cutoff points, and
instead supports two continuous dimensions as the best way to
model adult attachment (Fraley & Waller, 1998). In a factor
analysis of data gathered from over 1,000 undergraduates, Brennan, Clark, and Shaver (1998) included all of the extant self-report
measures of adult attachment (14 measures, 60 subscales, 323
items) and identified two relatively orthogonal dimensions of
Anxiety and Avoidance. Adult attachment anxiety is characterized
as an excessive need for approval from others and fear of interpersonal rejection or abandonment. Adult attachment avoidance
involves an excessive need for self-reliance and fear of interpersonal closeness or dependence. People with high levels of either
dimension or both dimensions in combination are assumed to have
Meifen Wei and W. Todd Abraham, Department of Psychology, Iowa
State University; Brent Mallinckrodt, Department of Educational, School,
and Counseling Psychology, University of Missouri–Columbia; Daniel W.
Russell, Department of Human Development and Family Studies, Iowa
State University.
We thank Robyn Zakalik, Shanna Behrendsen, Anne Giusto, and Mike
McGregor for their assistance with data collection.
Correspondence concerning this article should be addressed to Meifen
Wei, Department of Psychology, W112 Lagomarcino Hall, Iowa State
University, Ames, IA 50011-3180. E-mail: wei@iastate.edu
201
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
202
WEI, MALLINCKRODT, RUSSELL, AND ABRAHAM
achieve positive rewards. Maladaptive perfectionism involves unrealistically high standards, intense ruminative concern over mistakes, perceived pressure from others to be perfect, a perceived
large discrepancy between one’s performance and personal standards, compulsive doubting of one’s actions, and motivation to
avoid negative consequences (Enns & Cox, 2002).
Theorists suggest that maladaptive perfectionism results when a
child’s need for acceptance and love from parents is accompanied
by a parent’s failure to provide the needed acceptance and positive
regard (Hamachek, 1978). Observational research has shown that
if caregivers are inconsistent and unreliable in responding to the
emotional or physical needs of young children, anxious attachment
is frequently the result (Ainsworth, Blehar, Waters, & Wall, 1978).
Serious interpersonal problems may develop in adults whose parents used a love withdrawal style of discipline involving threats to
withhold affection as a means of control (Mallinckrodt & Wei,
2003). Children with attachment anxiety may quickly learn that if
they are “perfect” boys or girls, they may be more likely to gain
their parents’ love and acceptance. This pattern of striving for
perfection as a way to earn acceptance that was only intermittently
available in childhood may persist as a maladaptive pattern in
adults.
A different dynamic may underlie the connection between perfectionism and attachment avoidance. Attachment avoidance is
believed to involve a negative working model of others along with
a positive working model of self (Bartholomew & Horowitz,
1991). However, striving to be “perfect” in the view of others may
be an outward defense that masks a deeply wounded inner sense of
self resulting from the inadequate emotional responsiveness of
caregivers early in development (Lapan & Patton, 1986; Robbins
& Patton, 1985). Children with avoidant attachment tend to describe themselves as perfect (Cassidy & Kobak, 1988), but they
may drive themselves to attain perfection to avoid others’ rejection
and to manage their own hidden sense of imperfections. For
example, a child may think, “If I am perfect, no one will hurt me”
(Flett, Hewitt, Oliver, & Macdonald, 2002). Thus, initially striving
to be perfect may be a positive coping mechanism for children
whose caregivers are unresponsive or inconsistent in their responsiveness to the child’s needs. However, if striving to be perfect is
overused as a coping strategy, it may lead to depressive mood in
adulthood. Therefore, the specific form that the maladaptive striving for perfection may take might depend on the particular mixture
of attachment avoidance or attachment anxiety experienced in
adulthood.
Although several theorists have suggested that the origins of
perfectionism are related to problematic attachment in the parent–
child relationship, until recently there were very few empirical
studies of perfectionism and attachment. Among the small number
of available studies, Rice and Mirzadeh (2000) reported that maladaptive perfectionism was related to insecure attachment,
whereas adaptive perfectionism was related to secure attachment
in college students. Similarly, Andersson and Perris (2000) found
that perfectionism was positively associated with insecure attachment. Additionally, Flett et al. (2001) found that persons with high
attachment anxiety and avoidance reported higher perceived pressure from others to be perfect. Thus, previous studies have provided tentative evidence that attachment avoidance and attachment
anxiety are positively associated with maladaptive perfectionism.
Several studies have shown that perfectionism is positively
associated with depression or hopelessness. For example, perfectionism in college students was associated with greater depressive
symptoms (e.g., Chang, 2002; Chang & Sanna, 2001; Cheng,
2001; Hewitt & Flett, 1991) and suicidal preoccupation (Adkins &
Parker, 1996; Chang, 1998). In longitudinal studies, perfectionism
has been linked to both depression and hopelessness over time
(Chang & Rand, 2000; Flett, Hewitt, Blankstein, & Mosher, 1995).
Also, Hewitt and Flett (2002) reported that perceived pressure
from others to be perfect was associated with hopelessness across
different studies and populations (e.g., Chang & Rand, 2000;
Dean, Range, & Goggin, 1996). On the basis of these previous
studies, in the present study we chose to represent the latent
variable of depressive mood with indicators of depression and
hopelessness.
It is possible that adults with high attachment anxiety or avoidance are likely to develop maladaptive perfectionism and, in turn,
experience significant depressive mood. Some studies have examined how maladaptive perfectionism might serve as a mediator
between parent– child interactions and depressive mood. Randolph
and Dykman (1998) found that perfectionism fully mediated the
relationship between critical parenting and depression-proneness
and partially mediated the relationship between perfectionistic
parenting and depression-proneness in undergraduate students.
Enns, Cox, and Clara (2002) reported that maladaptive perfectionism mediated the relationship between harsh parenting (e.g., critical parenting, parental overprotection, and parental lack of care)
and depression. However, our search of the literature could not
locate any previous study that examined perfectionism as a mediator between attachment and depressive mood. If maladaptive
perfectionism does serve as a mediator, interventions could be
targeted at adults with attachment anxiety or avoidance to help
decrease their maladaptive perfectionism and in turn decrease their
depressive mood.
Hewitt and Flett (2002) argued that perfectionism could serve as
a moderator (as well as a mediator) between insecure attachment
and depressive mood. Several studies have found that specific
dimensions of perfectionism (e.g., pressure from others to be
perfect) interacted with general stress (e.g., major life stress or
self-appraisal stress) to predict increased depression symptoms or
negative affect (e.g., Chang & Rand, 2000; Cheng, 2001; Dunkley,
Zuroff, & Blankstein, 2003; Flett et al., 1995). That is, greater
depression or negative affect was reported by participants with
higher combined levels of perfectionism and perceived stress. In
addition, other studies reported that specific dimensions of perfectionism interacted with specific stressors to predict higher levels of
depression. Hewitt and Flett (1993) found that perfectionism,
particularly in the form of perceived pressure from others to be
perfect, interacted with interpersonal stressors (e.g., relationship
problems or lack of intimacy) to predict depression. It appears that
maladaptive perfectionism could serve as a potential moderator of
the relationship between general or specific stressors and psychological distress.
Attachment anxiety or attachment avoidance could be viewed as
a source of chronic interpersonal stress. Perfectionism may lead to
depressive mood because it generates core interpersonal needs that
are difficult to satisfy (i.e., the need for others’ approval, or the
need to be perfect to avoid others’ rejection). Maladaptive perfectionism might interact with attachment anxiety or attachment
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
PERFECTIONISM AND ATTACHMENT
avoidance to worsen depressive mood (Hewitt & Flett, 2002).
From the standpoint of putative causal links, in a mediating scenario attachment insecurity (x1) is believed to cause higher levels
of maladaptive perfectionism (x2), which in turn causes higher
levels of depressive mood (y). If the mediation is partial rather than
complete, there would also be a significant direct link between (x1)
attachment insecurity and (y) depressive mood (Baron & Kenny,
1986; Holmbeck, 1997). By contrast, in a moderating scenario
there is no requirement that x1 causes x2 and, in fact, the two
variables may be uncorrelated. However, the strength of association between x1 (in this case, attachment insecurity) and y (depressive mood) is believed to vary for differing levels of x2 (maladaptive perfectionism). Unfortunately, there has been no empirical
research studying how maladaptive perfectionism might interact
with attachment to predict depressive mood.
Because it is possible for maladaptive perfectionism to serve as
both an intermediate link in the causal chain leading from attachment insecurity to depressive mood (i.e., as a mediator) and as a
203
variable that alters the strength of association between attachment
insecurity and depressive mood (i.e., as a moderator), both types of
relationships were explored in this study. Specifically, the purpose
of the present study was to examine whether the maladaptive
aspects of perfectionism (e.g., concern over mistakes, doubts about
actions, and perceived discrepancy between one’s standards and
performance) serve as a mediator, as a moderator, or as both in the
context of the relationship between adult attachment insecurity
(anxiety and avoidance) and depressive mood (depression and
hopelessness). Figures 1A and 1B depict both of these hypothesized relationships. Structural equation modeling (SEM) was used
to test the models depicted in this figure. Slaney, Rice, Mobley,
Trippi, and Ashby (2001) argued that the discrepancy between
high standards and perceptions of performance was a defining
feature of maladaptive perfectionism, whereas high standards
without perceived discrepancy could indicate adaptive perfectionism. Therefore, measures of discrepancy between standards and
performance, concern over mistakes, and doubts about one’s ac-
Figure 1. Hypothesized mediating effects (A) and moderating effects (B) of maladaptive perfectionism on the
links between attachment anxiety and attachment avoidance with depressive mood. The moderating effects (B)
of maladaptive perfectionism on the links between attachment anxiety and attachment avoidance with depressive
mood were examined separately.
WEI, MALLINCKRODT, RUSSELL, AND ABRAHAM
204
tions served as the indicators for the construct of maladaptive
perfectionism, in addition to measures of depression and hopelessness, which served as indicators of the latent variable depressive
mood.1
Method
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Participants
Participants were 310 undergraduate students enrolled in introductory
psychology classes at a large midwestern university. The participants were
told that the purpose of the research was “to learn about factors affecting
college students’ adjustment.” The sample included 225 (73%) women and
85 (27%) men. Their mean age was 19.27 years (SD ⫽ 1.88, range ⫽
18 –30 yrs.). Approximately 53% of the participants were freshmen. Ethnic
identification was predominantly White/Caucasian (84%), followed by
international students of various ethnicities (4.8%), Asian American
(4.2%), African American (2.3%), Hispanic American (2.3%), multiracial
American (1.0%), and others (1.3%). Most participants (98.0%) indicated
they were single or never married. Students received partial credit toward
their course grade for participating in this study. The amount of credit
varied depending on their particular section of the course.
Instruments
Experiences in Close Relationships Scale (ECRS; Brennan et al., 1998).
The ECRS is a 36-item self-report measure of adult attachment containing
two 18-item subscales derived from the factor analysis by Brennan et al.
(1998) described previously. The subscales assess dimensions of adult
attachment, Anxiety and Avoidance. Participants use a 7-point Likert-type
scale (1 ⫽ disagree strongly, 7 ⫽ agree strongly) to rate how well each
statement describes their typical feelings in romantic relationships. The
Anxiety subscale taps fears of abandonment and rejection. The Avoidance
subscale assesses discomfort with dependence and intimate self-disclosure.
Brennan et al.’s reported coefficient alpha was .91 and .94 for the Anxiety
and Avoidance subscales, respectively. In the present study, coefficient
alpha was .90 for the Anxiety subscale and .91 for the Avoidance subscale.
Brennan et al. also reported that scale scores were correlated in expected
directions with scores on self-report measures of touch aversion and
postcoital emotions. Measured indicators for the two latent variables of
attachment anxiety and attachment avoidance were created from three
6-item parcels for each subscale. Following the recommendation of Russell, Kahn, Spoth, and Altmaier (1998), exploratory factor analyses were
conducted using maximum-likelihood extraction for the two factors (Anxiety and Avoidance) separately. The items were then rank-ordered on the
basis of the magnitude of the factor loadings and successively assigned
pairs of the highest and lowest items to each parcel to equalize the average
loadings of each parcel on its respective factor.
Almost Perfect Scale–Revised (APS-R; Slaney et al., 2001). The
APS-R is a 23-item self-report measure designed to assess levels of
perfectionism. Respondents use a 7-point Likert-type scale (1 ⫽ strongly
disagree, 7 ⫽ strongly agree) in responding to the items. The APS-R is
made up of three subscales: High Standards, Order, and Discrepancy. In
this study only the 12-item Discrepancy subscale was used. This subscale
measures the degree to which respondents perceive themselves as failing to
meet personal standards for performance. Slaney et al. reported a coefficient alpha of .92 for the Discrepancy subscale, whereas coefficient alpha
was .94 in the present sample. Slaney et al. reported evidence of construct
validity in the form of significant correlations between the Discrepancy
subscale and other perfectionism measures such as Concern Over Mistakes
(r ⫽ .55) and Doubts About Actions (r ⫽ .62).
Multidimensional Perfectionism Scale (FMPS; Frost, Marten, Lahart, &
Rosenblate, 1990). The FMPS is a 35-item instrument designed to measure perfectionism. Each item uses a 5-point Likert-type scale (1 ⫽
disagree strongly, 5 ⫽ agree strongly). Consistent with Dunkley, Blankstein, Halsall, Williams, and Winkworth (2000), only two of the six FMPS
subscales were used as indicators of perfectionism in this study: (a)
Concern Over Mistakes (9 items) taps a tendency to interpret mistakes as
failures and to believe that one will lose the respect of others when one
fails; and (b) the Doubts About Actions (4 items) subscale, which measures
the tendency to doubt one’s ability to accomplish tasks or the quality of
one’s performance. In the present study, coefficient alphas were .89 and .74
for Concern Over Mistakes and Doubts About Actions, respectively. Frost,
Heimberg, Holt, Mattia, and Neubauer (1993) found that Concern Over
Mistakes and Doubts About Actions not only reflected maladaptive evaluative concerns of perfectionism, but were also the subscales most strongly
related to depression. Criterion-related validity is evidenced by correlations
between FMPS subscales and measures of psychological symptoms (e.g.,
Brief Symptom Inventory) and adjustment such as compulsiveness, selfesteem, procrastination, and depression (Frost et al., 1993, 1990).
Beck Depression Inventory (BDI; Beck, Ward, Mendelson, Mock, &
Erbaugh, 1961). The BDI is a widely used 21-item self-report measure of
depressive symptoms. Each item consists of a depression symptom cluster
scored on a 0 –3 response scale based on the severity of the symptom.
Scores across the items are summed to obtain a total BDI score, with higher
scores indicating more severe depression. Internal consistency for the BDI
for undergraduates ranges from .78 to .92, with a mean coefficient alpha of
.85. In the present study, coefficient alpha was .86. Test–retest reliabilities
for nonpsychiatric participants ranged from .60 (7 days) to .83 (1– 6 hr),
with reports of .78 for a 2-week and a 3-week period. Considerable
evidence of validity has been demonstrated for the BDI as a measure of
depressive symptoms (Beck, 1967; Bumberry, Oliver, & McClure, 1978).
Beck Hopelessness Scale (BHS; Beck, Weissman, Lester, & Trexler,
1974). The BHS is a 20-item inventory that assesses the degree to which
an individual’s cognitive schemata are characterized by pessimistic expectations. The scale uses a true–false response format. Scores can range from
0 to 20, with higher scores indicating a greater degree of hopelessness.
Internal consistency of .93 has been reported, along with concurrent
validity of .74 with clinical ratings of hopelessness and .60 with other
scales of hopelessness (Beck et al., 1974). In the present study, coefficient
alpha for the BHS was .78.
Procedure
The questionnaires were administrated to small groups of 3–25 students
who signed up for one of several data collection times. Participants were
guaranteed anonymity of their responses and confidentiality of the data,
given that no personal identifying information was solicited on the questionnaires. Completing the entire packet of instruments typically required
25– 40 min.
Results
Descriptive Statistics
Means, standard deviations, and zero-order correlations for the
13 measured variables are shown in Table 1. Data were checked
for normality, which is a critical assumption underlying the
1
One issue raised by reviewers concerned the fact that we only used one
measure, the ECRS, to operationalize the attachment variable. Because the
ECRS was developed on the basis of a factor analysis of existing measures
of attachment (see Brennan et al., 1998), we felt that this measure adequately represented the nature of the construct. Indeed, it is very likely that
items on any other measure of adult attachment would be redundant with
items on this measure. Therefore, we did not feel it was necessary to use
other measures of adult attachment in this investigation.
PERFECTIONISM AND ATTACHMENT
205
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 1
Means, Standard Deviations, and Zero-Order Correlations Among 13 Observed Variables
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
Anxiety 1
Anxiety 2
Anxiety 3
Avoid 1
Avoid 2
Avoid 3
DIS
CM
DA
BDI
BHS
Anxiety 1 ⫻ DIS
Avoid 1 ⫻ DIS
M
1
2
3
4
5
6
7
8
9
10
11
12
13
21.88
23.66
20.17
16.78
17.05
15.62
43.58
22.62
10.76
8.36
1.71
34.85
31.45
6.17
.76
6.59
.77
.73
6.52
.19
.10
.22
6.60
.12
.08
.18
.87
6.70
.08
⫺.02
.10
.85
.85
6.55
.38
.39
.41
.32
.24
.21
14.83
.37
.34
.33
.27
.21
.20
.58
7.33
.32
.33
.32
.34
.27
.23
.61
.54
3.30
.38
.39
.38
.18
.12
.07
.53
.43
.43
6.71
.34
.28
.33
.24
.17
.19
.36
.33
.29
.60
0.81
⫺.05
⫺.03
⫺.06
⫺.13
⫺.12
⫺.15
⫺.10
⫺.10
⫺.14
.06
.07
98.25
⫺.13
⫺.13
⫺.21
.07
.04
.03
.06
.04
.17
.05
⫺.03
.14
104.78
Note. N ⫽ 310. Standard deviations are shown on the diagonal. Anxiety 1, Anxiety 2, Anxiety 3 ⫽ Anxiety Parcel 1, Anxiety Parcel 2, Anxiety Parcel
3 from the Anxiety subscale of the Experiences in Close Relationship Scale; Avoid 1, Avoid 2, Avoid 3 ⫽ Avoid Parcel 1, Avoid Parcel 2, Avoid Parcel
3 from the Avoidance subscale of the Experiences in Close Relationship Scale; DIS ⫽ Discrepancy subscale of the revised Almost Perfect Scale; CM ⫽
Concern Over Mistakes subscale from the Frost Multidimensional Perfectionism Scale; DA ⫽ Doubts About Actions subscale from the Frost
Multidimensional Perfectionism Scale; BDI ⫽ Beck Depression Inventory; BHS ⫽ Beck Hopelessness Scale (after square root transformation); Anxiety
1 ⫻ DIS ⫽ the interaction between Anxiety Parcel 1 and Discrepancy (after centering Anxiety 1 and DIS); Avoid 1 ⫻ DIS ⫽ the interaction between Avoid
Parcel 1 and Discrepancy (after centering Avoid 1 and DIS). Absolute values of correlations greater than .17 were significant at p ⬍ .01.
maximum-likelihood procedure used in this study. Results indicated univariate normality for all measured variables except the
BHS (Beck et al., 1974; skew Z ⫽ 1.87, and kurtosis Z ⫽ 4.47).
We therefore conducted a square root transformation for the BHS
variable. The skew and kurtosis for the transformed BHS were Z ⫽
.16 and .75, respectively, indicating a normal distribution. The
BHS and the transformed BHS were highly correlated (r ⫽ .94).
Therefore, we used the transformed BHS variable in subsequent
analyses.2
Measurement Model for Testing Mediation Effects
The analysis of the proposed mediation model followed the
two-step procedure recommended by Anderson and Gerbing
(1988). First, we used a confirmatory factor analysis to develop a
measurement model with an acceptable fit to the data. Once an
acceptable measurement model was developed, the structural
model was tested. The measurement model was estimated using
the maximum-likelihood method in the LISREL 8.50 program. As
suggested by Hu and Bentler (1999) and Quintana and Maxwell
(1999), three indices were used to assess goodness of fit for the
models: the comparative fit index (CFI; values of .95 or greater are
desirable), the standardized root-mean-square residual (SRMR;
values of .08 or less are desirable), and the root-mean-square error
of approximation (RMSEA; values of .06 or less are desirable).
Finally, we used the chi-square difference test to compare nested
models. An initial test of the measurement model resulted in a
relatively good fit to the data, ␹2(38, N ⫽ 310) ⫽ 72.60, p ⬍ .001,
CFI ⫽ .98, SRMR ⫽ .04, and RMSEA ⫽ .05 (95% confidence
interval [CI]: .03, .07).3 All of the loadings of the measured
variables on the latent variables were statistically significant ( p ⬍
.001; see Table 2). Therefore, all of the latent variables appear to
have been adequately measured by their respective indicators. In
addition, the correlations among the independent (exogenous) latent variables, the mediator latent variable, and dependent latent
variable were statistically significant ( p ⬍ .05; see Table 3)
Structural Model for Testing Mediation Effects
A number of methods have been suggested in the literature for
testing mediation effects. Recently, MacKinnon, Lockwood, Hoffman, West, and Sheets (2002) evaluated 14 methods with regard to
2
We also tested the multivariate normality of the observed variables as
a set, including the transformed BHS (Beck et al., 1974) variable, based on
the test developed by Mardia (see Bollen, 1989). The significant result,
␹2(2, N ⫽ 310) ⫽ 114.05, p ⬍ .001, indicated that the data were not
multivariate normal. Therefore, we used the procedure developed by Satorra and Bentler (1988) to adjust the chi-square statistics and standard
errors of the parameter estimates for the impact of nonnormality. In the
mediation model, the results after adjusting for the impact of nonnormality
did not differ from the results when we did not adjust for nonnormality. In
the moderation model, the results for the path coefficients were identical
whether or not we adjusted for the impact of nonnormality. However, the
standard error of the latent interaction term became very large following
the adjustment for nonnormality. This problem associated with interaction
terms and the Satorra-Bentler adjustment for nonnormality has been noted by
others (e.g., Yang-Wallentin & Joreskog, 2001). Therefore, we report results
for the moderation model without adjusting for the impact of nonnormality.
3
We examined whether the results would be equivalent for men and
women in the measurement model, structural model, and the models with
interaction effect. A series of multiple-group analyses were conducted
using LISREL 8.50 to examine whether female and male groups differed
from one another in terms of the measurement model, the structural model,
and the models with interaction effects (Byrne, 1998). Results suggested
that the measurement model and structural model were equivalent for the
male and female groups. However, the models comparing men and women
that included the interaction effect did not converge. This was likely
because of the relatively small number of men (n ⫽ 85) included in the
sample. Therefore, we conducted a hierarchical regression analysis to
examine whether the interaction effect varied for men and women. Results
of the regression analysis indicated that the three-way interaction (Attachment
Anxiety ⫻ Maladaptive Perfectionism ⫻ Gender) predicting depressive mood
was not significant (␤ ⫽ .002), t(302)⫽ 0.55, p ⬎ .05. Thus, it appears that the
interaction effect was equivalent for male and female participants.
WEI, MALLINCKRODT, RUSSELL, AND ABRAHAM
206
Table 2
Factor Loadings for the Measurement Model
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Measure and variable
Unstandardized
factor loading
SE
Z
Standardized
factor loading
5.52
5.59
5.61
0.26
0.30
0.29
21.08
18.87
19.06
.89***
.85***
.86***
6.15
6.25
5.97
0.29
0.30
0.33
21.14
21.17
18.35
.93***
.93***
.91***
12.27
5.23
2.42
0.71
0.38
0.18
17.33
13.67
13.60
.83***
.71***
.73***
6.08
0.54
0.43
0.05
14.00
10.93
.90***
.66***
Attachment Anxiety
Anxiety Parcel 1
Anxiety Parcel 2
Anxiety Parcel 3
Attachment Avoidance
Avoidance Parcel 1
Avoidance Parcel 2
Avoidance Parcel 3
Maladaptive Perfectionism
Discrepancy
Concern Over Mistakes
Doubts About Actions
Depressive Mood
BDI
BHS
Note. N ⫽ 310. Discrepancy ⫽ one subscale from the revised Almost Perfect Scale; Concern Over Mistakes ⫽
one subscale from the Frost Multidimensional Perfectionism Scale; Doubts About Actions ⫽ one subscale from
the Frost Multidimensional Perfectionism Scale; BDI ⫽ Beck Depression Inventory; BHS ⫽ Beck Hopelessness
Scale (after square root transformation).
*** p ⬍ .001.
Type I error and statistical power. They found that the commonly
used method recommended by Baron and Kenny (1986) for testing
mediation had the lowest statistical power of the 14 methods
examined. Instead, MacKinnon et al. (2002) recommend testing
for mediation using the test of the indirect effect of the causal
variable through the hypothesized mediator reported by the LISREL program. However, MacKinnon et al. (2002) have shown that
the method used by LISREL to calculate the standard error of the
indirect effect tends to yield incorrect estimates. To develop more
accurate estimates of standard errors of the indirect effects