Testing a Tripartite Model: II. Exploring the Symptom Structure of Anxiety and Depression in Student, Adult, and Patient Samples

Abstract

L. A. Clark and D. Watson (1991) proposed a tripartite model of depression and anxiety that divides symptoms into 3 groups: symptoms of general distress that are largely nonspecific, manifestations of anhedonia and low positive affect that are specific to depression, and symptoms of somatic arousal that are relatively unique to anxiety. This model was tested by conducting separate factor analyses of the 90 items in the Mood and Anxiety Symptom Questionnaire (D. Watson & L. A. Clark, 1991) in 5 samples (3 student, 1 adult, 1 patient). The same 3 factors (General Distress, Anhedonia vs. Positive Affect, Somatic Anxiety) emerged in each data set, suggesting that the symptom structure in this domain is highly convergent across diverse samples. Moreover, these factors broadly corresponded to the symptom groups proposed by the tripartite model. Inspection of the individual item loadings suggested some refinements to the model.

Full text

Journal
of
Abnormal
Psychology
1995.
Vol.
104.
No. I.
15-25
Copyright
1995
by the
American
Psychological
Association,
Inc.
0021-843X/95/S3.00
Testing
a
Tripartite Model:
II.
Exploring
the
Symptom Structure
of
Anxiety
and
Depression
in
Student, Adult,
and
Patient Samples
David
Watson
and Lee
Anna Clark
University
of
Iowa
Kris
Weber
and
Jana
Smith
Assenheimer
Southern
Methodist
University
Milton
E.
Strauss
and
Richard
A.
McCormick
Cleveland
Department
of
Veterans
Affairs
Medical
Center,
Brecksville
Unit
L. A.
Clark
and D.
Watson
(1991)
proposed
a
tripartite model
of
depression
and
anxiety that divides
symptoms into
3
groups: symptoms
of
general distress that
are
largely
nonspecific, manifestations
of
anhedonia
and low
positive
affect
that
are
specific
to
depression,
and
symptoms
of
somatic arousal
that
are
relatively
unique
to
anxiety.
This model
was
tested
by
conducting
separate
factor analyses
of
the 90
items
in the
Mood
and
Anxiety
Symptom Questionnaire
(D.
Watson
& L. A.
Clark,
1991)
in
5
samples
(3
student,
1
adult,
1
patient).
The
same
3
factors (General Distress, Anhedonia
vs.
Positive
Affect,
Somatic
Anxiety)
emerged
in
each data set, suggesting that
the
symptom structure
in
this
domain
is
highly
convergent
across
diverse samples. Moreover, these factors broadly
corresponded
to the
symptom groups proposed
by the
tripartite model. Inspection
of the
individual item loadings
suggested
some
refinements
to the
model.
Recently,
clinicians
and
researchers
have
shown renewed interest
in
the
relation between depression
and
anxiety (see
D. A.
Clark,
Beck,
&
Stewart,
1990;
Kendall
&
Watson,
1989;
Maser
&
Clon-
inger,
1990).
This interest
has
been sparked
by
persistent evidence
that
these
two
constructs
are
difficult
to
differentiate
empirically.
For
example, studies
have
shown consistently that self-report mea-
sures
of
anxiety
and
depression
are
strongly interrelated
in
both
clinical
and
nonclinical
samples, with correlations typically
in the
.45
to .75
range (e.g.,
L. A.
Clark
&
Watson,
1991;
Costa
&
McCrae,
1992;
Gotlib,
1984;
Mendels, Weinstein,
&
Cochrane,
1972).
Sim-
ilarly,
clinicians'
and
teachers' ratings
of
anxiety
and
depression
are
strongly
correlated
with
one
another (e.g., Moras,
DiNardo,
&
Bar-
low,
1992;
Wolfe
et
al.,
1987;
for a
review,
see L. A.
Clark
&
Watson,
1991).
Finally, substantial
comorbidity
has
been observed between
the
mood
and
anxiety disorders
(L. A.
Clark,
1989;
Maser
&
Clon-
inger,
1990;
Sanderson, Beck,
&
Beck,
1990),
leading some investi-
gators
to
suggest
the
need
for a new
diagnostic category
of
mixed
anxiety-depression
(L. A.
Clark
&
Watson,
1991;
Zinbarg
&
Bar-
low,
1991;
Zinbarg
et
al.,
1994).
Tripartite
Model
Three
Symptom
Groups
Why
are
anxiety
and
depression
so
strongly related,
and how
can
they
be
better
differentiated
from
one
another?
L. A.
Clark
David
Watson
and Lee
Anna Clark, Department
of
Psychology, Uni-
versity
of
Iowa; Kris Weber
and
Jana Smith Assenheimer, Department
of
Psychology, Southern Methodist
University;
Milton
E.
Strauss
and
Richard
A.
McCormick, Psychology Service, Cleveland
Department
of
Veterans
Affairs
Medical Center,
Brecksville
Unit.
This
research
is
based
in
part
on the MA
theses
of
Kris Weber
and
Jana
Smith Assenheimer under
the
supervision
of
David Watson.
Correspondence concerning this article should
be
addressed
to
David
Watson,
Department
of
Psychology, University
of
Iowa, Iowa City, Iowa
52242-1407.
Electronic mail
may be
sent
to
david-watson@uiowa.edu
and
Watson
(1991)
reviewed
the
relevant literature
and
pro-
posed
a
tripartite model that
may
provide
a
partial answer
to
these questions.
In
this model, symptoms
of
depression
and
anxiety
are
subdivided
into
three broad groups. First, many
symptoms
of
both constructs
are
strong markers
of a
general
distress
or
negative
affect
factor
and
are, therefore, relatively
nonspecific.
In
other words, these symptoms
are
commonly
ex-
perienced
by
both anxious
and
depressed
individuals. This non-
specific
group includes both anxious
and
depressed
affect,
as
well
as
other symptoms (e.g., insomnia, restlessness, irritability,
poor
concentration) that
are
prevalent
in
both types
of
disorder.
In
the
tripartite model, these nonspecific symptoms
are
primar-
ily
responsible
for the
strong association between measures
of
anxiety
and
depression.
Nevertheless, each construct
is
characterized also
by a
cluster
of
relatively
unique symptoms.
That
is,
symptoms
reflecting
an-
hedonia
and the
absence
of
positive emotional experiences (e.g.,
feeling
disinterested
in
things, lacking energy,
feeling
that noth-
ing is
enjoyable, having
no fun in
life)
are
relatively specific
to
depression.
In
contrast, manifestations
of
somatic tension
and
arousal (e.g., shortness
of
breath,
feeling
dizzy
or
lightheaded,
dry
mouth, trembling
or
shaking)
are
relatively specific
to
anxiety.
L. A.
Clark
and
Watson
(1991)
emphasized that
all
three
types
of
symptoms must
be
included
in a
comprehensive
assess-
ment
of
these
constructs.
However,
a key
implication
of the
tri-
partite
model
is
that depression
and
anxiety
can be
differenti-
ated
better
by
deemphasizing
the
importance
of the
nonspecific
symptoms
and by
focusing more
on the two
unique symptom
clusters.
Evidence
for
the
Tripartite
Model
The
tripartite model
was
derived
from
three types
of
evidence
(L.
A.
Clark
&
Watson,
1991).
First,
content analyses indicated
15

16
WATSON
ET AL.
that anxiety scales with
the
best discriminant validity tended
to
measure
the
somatic symptoms
of
anxiety rather than anxious
mood
per
se;
in
contrast,
the
most differentiating depression
scales tended
to
assess
the
loss
of
interest
or
pleasure,
as
opposed
to
other manifestations
of
depression.
The
second line
of
evi-
dence came
from
studies comparing anxious
and
depressed
pa-
tients.
In
these analyses, only
a
small subset
of
symptoms reli-
ably
differentiated
the
patient groups. Specifically, autonomic
manifestations
of
panic (e.g., dizziness, racing heart)
and
symp-
toms
of
melancholia (e.g., loss
of
pleasure, early morning awak-
ening)
were
the
most
differentiating
markers
of
anxiety
and de-
pression, respectively.
The final
line
of
evidence came
from
fac-
tor
analytic studies that identified symptom dimensions
reflecting
the
three main subgroups
in the
tripartite model.
The
identified
dimensions consisted
of a
general neurotic factor that
included
feelings
of
inferiority
and
rejection,
oversensitivity
to
criticism,
and
anxious
and
depressed
affect;
a
specific depres-
sion
factor that
was
defined
by the
loss
of
interest
or
pleasure,
anorexia,
crying spells,
and
suicidal ideation;
and a
specific anx-
iety
factor
that
was
marked
by
items reflecting tension, shaki-
ness,
and
panic (see
L. A.
Clark
&
Watson,
1991).
In
a
companion article, Watson
et
al.
(1995)
reported
the first
direct test
of the
tripartite model using
the
Mood
and
Anxiety
Symptom
Questionnaire
(MASQ;
Watson
&
Clark,
1991)
and
other symptom
and
cognition measures.
The
MASQ includes
three
scales containing symptoms that, according
to the
tripar-
tite model, should
be
relatively nonspecific.
In
addition,
it
con-
tains
two
specific
scales—Anhedonic
Depression
and
Anxious
Arousal—that
assess
anhedonia/low
positive
affect
and
somatic
arousal, respectively. Consistent with
the
tripartite model, Wat-
son
et al.
(1995)
found
that these specific scales provided
the
best
differentiation
of the
constructs
in
each
of five
samples
(three student,
one
adult,
one
patient). Furthermore, Anxious
Arousal
and
Anhedonic Depression showed excellent con-
vergent
validity.
For
instance, factor analyses indicated that
these scales were clear markers
of the
underlying constructs;
moreover,
hierarchical multiple regression analyses revealed
that
they contained
the
most target-construct variance,
as
well
as
the
least
nontarget
variance. Overall, therefore,
the
data sup-
ported
the
tripartite model
by
demonstrating that scales assess-
ing
anhedonia
and
somatic arousal showed excellent convergent
and
discriminant validity.
Current Study
This study provides
the
second direct test
of L. A.
Clark
and
Watson's
(1991)
tripartite model. Specifically, using
the
same
five
samples
as in
Watson
et al.
(1995),
we
explored
the
factor
structure
of the 90
anxiety
and
depression symptoms that com-
prise
the
MASQ. Although
L. A.
Clark
and
Watson's
(1991)
review
revealed several studies that identified factors that
ap-
peared
to
reflect
the
three basic symptom groups proposed
by
the
tripartite model,
no
study
has
investigated directly
the de-
gree
to
which
the
symptom structure
in
this domain actually
corresponds
to the
model. Accordingly, this
was the
primary
goal
of
this study.
The
MASQ
was
constructed explicitly
to
test
key
aspects
of
the
tripartite model
and
contains items
from
all
three symptom
groups.
On the
basis
of the
model,
we
expected
to find
evidence
of
three broad factors:
(a) a
general distress factor consisting
of
prominent symptoms
of
both anxiety
and
depression, including
items reflecting both anxious
and
depressed mood;
(b) a
specific
depression factor that
is
defined
on one end by
items reflecting
energy,
enthusiasm,
and
high positive
affect,
and on the
other
end by
items
reflecting
anhedonia, loss
of
interest,
and low
pos-
itive
affect;
and (c) a
specific anxiety factor that
is
most strongly
marked
by
symptoms
of
somatic tension
and
arousal.
A
second
and
related goal
of
this study
was to
evaluate
the
composition
of the
MASQ scales.
As
will
be
discussed shortly,
the
MASQ symptoms were rationally grouped into scales
on the
basis
of
their content: Items judged
to be
relatively nonspecific
were
placed into
one of
three "general
distress"
scales, whereas
those viewed
as
relatively specific
to
depression
or
anxiety were
included
in
Anhedonic Depression
and
Anxious Arousal,
re-
spectively. Clearly, however, some
of
these rational judgments
may
have been
faulty;
for
example,
an
anxiety symptom that
was
thought
to be
relatively nonspecific actually might
be a
strong
marker
of the
specific anxiety
factor.
Therefore,
we ex-
amined
the
factor loadings
of the
MASQ items
to
determine
whether
each symptom
was
placed
in the
most appropriate
scale.
The
third goal
of
this study
was not
directly relevant
to the
tripartite model
per se. We
were interested
in
determining
the
extent
to
which
the
symptom structure
in
this domain
is
repli-
cable
across college student, normal adult,
and
psychiatric
pa-
tient samples. This
is an
important
and
timely issue: Although
considerable evidence
in
this area
has
been collected
from
all
three types
of
participants,
the
extent
to
which they yield sim-
ilar
or
dissimilar results remains unclear. This study provides
evidence relevant
to
this issue
by
examining
the
replicability
of
symptom structure across these
different
populations.
Method
Participants
Three
samples
("Student
1,"
"Student
2," and
"Student
3")
were
comprised
of
undergraduates
enrolled
in
psychology
courses
at
South-
ern
Methodist
University:
They
contained
516
(208 men,
304
women,
and 4 for
whom information
is
unavailable),
381
(143
men,
234
women,
and 4
unavailable),
and 522
(206
men and
316
women)
participants,
respectively.
(Because
86% of the
Student
2
participants
also
had
been
included
in the
Student
1
sample,
these
ratings
essentially
represent
a
retest
of the
earlier
assessment.)
The
adult
sample
contained
329
indi-
viduals
(142
men and
187
women) with
a
mean
age of
40.0
years.
Most
of
the
participants
(78%) were
employees
of
various
businesses
in the
Dallas-Fort
Worth
metropolitan
area;
the
others
were
visitors
to a
Dal-
las
area
hospital
(9%)
and
members
of
local
social
and
church
groups
(13%).
Finally,
the
patient
sample
consisted
of 470
consecutive
admis-
sions
(453 men,
5
women,
and 12 for
whom information
was
unavail-
able)
to the
assessment
unit
of a
comprehensive
substance
abuse
treat-
ment
program
at the
Cleveland
Department
of
Veterans
Affairs
Medical
Center.
Their
mean
age was
39.3
years.
(For
more
information
regarding
these
samples,
see
Watson
et
al.,
1995.)
Measures
All
participants
completed
the
MASQ
(Watson
&
Clark,
1991),
which
consists
of 90
items
culled
from
the
symptom
criteria
for the
anxiety
and
mood
disorders
(see
Watson
et
al.,
1995).
Participants
indi-

ANXIETY
AND
DEPRESSION
SYMPTOM
STRUCTURE
17
cated
to
what extent they
had
experienced each symptom
(1
= not at
all,
5 =
extremely)
"during
the
past week, including today."
Using
the
tripartite model
as a
conceptual guide, Watson
and
Clark
(1991)
initially
grouped
the
MASQ items into
six
scales
on the
basis
of
their
content. Paraphrased versions
of the
items—grouped
according
to
their
initial
placement
in
these
six
scales—are
presented
in
Table
6.
Three MASQ scales contain symptoms
that—according
to the
tripar-
tite
model—should
be
relatively nonspecific.
The
criteria
of the
revised
third
edition
of the
Diagnostic
and
Statistical Manual
of
Mental
Disor-
ders
(DSM-IH-R;
American Psychiatric Association,
1987)
guided
the
placement
of
these
general
distress
symptoms
into
the
three
scales;
that
is, the
items were subdivided
on the
basis
of
whether they
are
included
in
the
DSM-IH-R
criteria
of (a) one or
more
anxiety
disorders,
(b) one
or
more mood
disorders,
or (c)
both types
of
disorder. Thus,
the
General
Distress:
Mixed Symptoms (GD: Mixed) scale contains
15
items that
appear
in the
symptom criteria
of
both
the
anxiety
and
mood
disorders
(e.g., insomnia). Conversely,
the
General
Distress:
Anxious Symptoms
scale (GD: Anxiety;
11
items) includes several items
reflecting
anxious
mood,
as
well
as
other symptoms
of
anxiety
disorder that were expected
to be
relatively
nondifferentiating.
Finally,
the
General Distress: Depres-
sive
Symptoms scale (GD: Depression;
12
items) contains several indi-
cators
of
depressed mood along with other
relatively
nonspecific symp-
toms
of
mood
disorder.
The
other three original MASQ scales contain symptoms that were
hypothesized
to be
relatively
specific
to
either anxiety
or
depression.
First, Anxious Arousal
(17
items) includes symptoms
of
somatic ten-
sion
and
hyperarousal (e.g.,
feeling
dizzy
or
lightheaded, shortness
of
breath,
dry
mouth). This scale originally contained
19
items. However,
a
preliminary factor analysis
in the
Student
1
sample indicated that
two
of the
items ("was afraid
I was
losing
control,"
"felt
like
I was
going
crazy") actually loaded more strongly
on the
general distress factor than
on
the
specific
anxiety factor. Consequently, these items were eliminated
from
the
scale.
The
final two
scales both contained items that were expected
to be
relatively
specific
to
depression;
initially,
they were assessed separately
to
examine empirically whether they should
be
combined into
a
single
scale. Loss
of
Interest originally contained
9
items that
reflect
anhedo-
nia,
disinterest,
and low
energy (e.g.,
"felt
nothing
was
enjoyable").
One
item
("felt
like
being
alone")
was
dropped, however, because
a
reliability
analysis
in the
Student
1
sample indicated that
it was
uncorrelated with
the
others.
The
other
scale—High
Positive
Affect—included
24
items that
di-
rectly
assessed
positive emotional experiences (e.g.,
felt
cheerful,
opti-
mistic;
had a lot of
energy; looked forward
to
things with enjoyment).
These
items were included
in the
MASQ
on the
basis
of
previous
re-
search indicating
that
it is
desirable
to
assess
high Positive
Affect
directly
because these high-end items tend
to be
stronger, purer markers
of the
underlying
factor than
are
items
reflecting
anhedonia
and low
Positive
Affect
(see Watson, Clark,
&
Carey,
1988;
Watson
&
Kendall,
1989).
As
noted earlier,
the
Loss
of
Interest
and
High Positive
Affect
items
both were expected
to be
relatively specific
to
depression. Furthermore,
these
two
scales were substantially interrelated, with
a
weighted mean
correlation
of
-.53
across
the five
data
sets (see Watson
et
al.,
1995).
Therefore,
Watson
and
Clark
(1991)
created
a new
22-item
scale—An-
hedonic
Depression—that
contained
the 8
Loss
of
Interest items
to-
gether
with
14 of the
(reverse-keyed) High Positive
Affect
items. This
Anhedonic
Depression scale
was
used
as the
specific
depression
mea-
sure
in the
analyses
reported
in
Watson
et al.
(1995).
Results
Initial
Factor Analyses
Exploring
one-
through
eight-factor
solutions.
The 90
MASQ
items
were
subjected
to
separate principal
factor
analy-
Table
1
Eigenvalues
of
the
First
15
Unrelated
Factors
in
Each Sample
Factor
number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Overall
common
variance
Student
1
(n
=
516)
20.91
7.43
2.73
2.59
1.47
1.34
1.26
1.08
0.99
0.98
0.90
0.78
0.73
0.70
0.67
47.94
Student
2
(n
=
381)
20.52
8.12
3.34
2.60
1.82
1.47
1.36
1.15
1.06
1.03
0.94
0.88
0.86
0.80
0.75
52.34
Student
3
(n
=
522)
21.28
7.58
2.70
2.12
.92
.44
.32
.13
.02
0.91
0.90
0.79
0.74
0.69
0.65
48.71
Adult
(n
=
329)
25.01
8.23
3.68
1.91
1.79
.59
.53
.42
.16
.08
0.96
0.92
0.82
0.80
0.72
57.95
Patient
(n
=
470)
26.85
6.44
3.38
1.81
1.36
1.26
1.09
1.03
0.88
0.84
0.78
0.74
0.67
0.62
0.62
51.94
ses
(squared
multiple
correlations
in the
diagonal;
communality
estimates
were
not
iterated)
in
each sample. Table
1
lists
the
eigenvalues
for the first
15
unrotated
factors
in
each solution.
The
most noteworthy aspect
of
these data
is
that
the five
solu-
tions
all
showed
a
very
similar pattern. Thus,
we
already
see
suggestive
evidence
of
structural convergence across these
samples.
We
initially explored
a
broad range
of
solutions.
Specifically,
we
examined
the
full
range
of
solutions
up to and
including
eight
factors,
by
which
point
it
became clear that
too
many
fac-
tors
were
being
extracted
(as we
describe shortly). Starting
with
the
two-factor
solutions,
all
factors
were rotated
using
varimax.
Our
initial
inspection
of the 1
-factor
solutions indicated that
a
very
large
general
factor
emerged
in
each data set;
it was
defined
by
the
depression,
anxiety,
and
general distress symptoms
on
one
pole
and by the
positive emotionality items
on the
other.
Virtually
all of the
items
were
salient markers
of
this
dimension.
The
highly
general
nature
of
this
factor
is
depicted
in
Table
2,
which
presents
the
mean number
of
markers (out
of 90
items,
averaged
across
the five
samples)
for
each
factor
in
each
solu-
tion;
in
these
and all
subsequent analyses,
a
marker
was
defined
as
a
variable that loaded
|
.301
or
greater
on a
factor
and had its
highest
loading
on
that
factor.
Table
2
indicates that,
on
average,
82.4
of the 90
items
(92%)
were
significant
markers
of
this gen-
eral
factor.
It
also should
be
noted,
however,
that
the
magnitude
of
the
loadings varied
widely
across items.
Averaged
across
the
five
solutions,
four
items
had
mean loadings less than
|.30|,
44
had
loadings between
|
.301
and |
.50),
and 42 had
loadings
greater
than
|
.501;
overall,
the
median loading
on
this
first
fac-
tor
was
|
.481.
Each
of the
samples also
yielded
a
highly
similar
two-factor
solution.
In
each case,
one
factor
was a
broad distress
dimension
that
was
defined
most strongly
by the
anxiety
and GD:
Mixed
symptoms,
but
also included many symptoms
of
depression.
In
contrast,
the
other
factor
was
relatively specific
to
depression:
It

18
WATSON
ET AL.
Table
2
Mean
Number
of
Markers
(Averaged
Across
the
Five
Samples)
for
One-
Through
Eight-Factor
Solutions
No. of
factors
in
solution
1
2
3
4
5
6
7
8
Mean
no. of
markers
for
factor
no.
1
82.4
55.0
30.8
29.2
26.4
26.2
27.6
27.4
2
31.4
29.6
26.4
25.2
24.8
24.6
24.6
3
24.8
23.6
20.0
20.2
19.8
19.2
4
5.8
10.2
10.4
7.2
7.8
5
6
3.8
3.6
0.8
5.0 1.6
3.4 3.4
7
8
1.0
1.0
0.4
Note.
A
marker
was
denned
as a
variable
that
loaded
|
.301
or
greater
on
a
factor
and had its
highest loading
on
that factor.
All
factors (other
than those
in the
one-factor solutions) were
rotated
using varimax.
was
defined
most strongly
by the
positive emotionality items
on
one
end,
and by
symptoms
of
depression
on the
other. Clearly,
these
two
factors resemble closely
the
negative
affect
and
posi-
tive
affect
dimensions that have been
identified
by
Tellegen
(1985;
Watson
&
Tellegen, 1985)
and
others. Table
2
indicates
that
both
of
these factors were quite large, averaging 55.0
and
31.4
markers, respectively.
For
our
purposes,
the
three-factor solutions were
the
most
crucial.
In
each data set, these solutions yielded factors that
ap-
peared
to
correspond closely
to the
symptom groups compris-
ing
the
tripartite model.
In
each sample,
the
factors consisted
of:
(a) a
broad,
nonspecific
distress
factor
that included symp-
toms
of
both anxiety
and
depression;
(b) a
specific
depression
factor
that
was
defined
on one
pole
by the
positive emotionality
items
and on the
other
by
anhedonia
and
other symptoms
of
depression;
and (c) a
specific
anxiety factor that
was
marked
by
items
reflecting
somatic arousal.
As
shown
in
Table
2,
these
factors
were
all
large
and
roughly similar
in
size: Across
the five
samples,
they averaged 30.8, 29.6,
and
24.8 markers,
respectively.
After
three factors,
the
solutions diverged appreciably;
in
fact,
no
later factor could
be
identified consistently
in all five
sam-
ples.
For
instance,
the
fourth
factor
in the
four-factor
solutions
was
defined
variously
by
items
reflecting
fatigue
and
poor con-
centration
(Student
1 and
Student
3
samples), laughing
and
talkativeness
(Student
2
sample),
and
insomnia (adult
and pa-
tient
samples). Similarly,
the fifth
factor
in the five-factor
solu-
tions
was
narrowly
defined
by
insomnia
and
sleep items
in two
solutions
(Student
2 and
patient)
and
more broadly character-
ized
by
general distress symptoms
in a
third (adult);
in the two
remaining
solutions (Student
1 and
Student
3),
however,
it had
no
markers
at
all.
Note
also
that
succeeding factors
were
substantially smaller
than
the first
three, with
few
significant
markers.
For
example,
in
the
four-factor
solutions
the
fourth
factor
had a
mean
of
only
5.8
markers,
and in the five-factor
solutions
the fifth
factor
aver-
aged
only
3.8
markers (see Table
2).
Beyond
five
factors,
all of
the
extracted dimensions were small
and
poorly
defined.
In
this
context,
it is
noteworthy that
the first
three factors remained
large
and
well-defined
even
in
later solutions. Thus,
in the
eight-
factor
solutions,
the first
three factors still averaged 27.4, 24.6,
and
19.2
markers, respectively;
in
other words,
the
large major-
ity
of the
anxiety
and
depression
symptoms continued
to
define
the
first
three factors, even
as
more
and
more factors were
extracted.
Quantitative
assessment
of
factor
convergence.
In
summary,
this initial evaluation suggested
that
the
solutions were highly
convergent
up to and
including three factors,
but
then diverged
sharply
from
one
another. Because factor replicability
across
different
samples
is a
crucial consideration
in
determining
the
best solution (Everett, 1983), this suggests that
no
more than
three factors
be
retained. Nevertheless,
it is
important
that
this
conclusion
be
corroborated
using
more
formal
quantitative
analyses.
Two
basic
approaches
for
assessing
factor
similarity
are
computing congruence
coefficients
that
are
based
on the
fac-
tor
loadings
and
correlating
the
factor
scores that
are
generated
by
each solution (see Gorsuch,
1983;
Harman,
1976).
Because
the
issue
of
factor replicability
is
central
to
this article,
we
pres-
ent findings
using both approaches.
First,
we
considered evidence
on the
basis
of
factor scores.
A
factor
solution generates
a set of
factor scoring weights
(in
this
case, regression-based weights)
for
each
of the
extracted factors.
A
set
contains
a
separate
weight
for
each
of the
factored vari-
ables; these weights
can
then
be
multiplied against
the
partici-
pants' actual item responses
to
yield
an
overall score
on
that
factor
for
each
participant.
For
example,
a
two-factor
solution
generates
two
sets
of
weights that
can be
multiplied
by the
item
responses
to
yield
two
factor scores
for
each participant; sim-
ilarly,
a
three-factor solution yields three sets
of
weights
that
can
be
used
to
compute three factor scores,
a
four-factor
solution
yields
four
sets
of
weights (and thus
four
scores),
and so on.
In
these analyses,
we had a
series
of
solutions
for
each
of five
data
sets.
Thus,
across
the five
samples,
the
one-factor solutions
generated
a
total
of five
sets
of
factor scoring weights (one
from
each
data
set),
the
two-factor solutions yielded
a
total
of
10
sets
of
factor
scoring weights
(2
from
each data set),
and so on.
These
weights
can be
used
not
only
to
compute
factor
scores
in the
data
set
from
which they were derived,
but
also
to
create
scores
in
any
data
set
that contains
all of the
originally factored vari-
ables.
In our
analyses,
we
used them
to
compute factor
scores
in
our
largest
data
set,
the
Student
3
sample
(N =
522).
If the
solutions
are
truly convergent across
the
different
samples, then
the
factor scoring weights
from
each
of the five
data sets should
produce corresponding factor scores that
are
highly correlated
with
each other.
For
instance,
the
weights
from
the five
one-
factor
solutions should generate
five
scores that
are
very
highly
intercorrelated. Similarly,
the
weights
from
the
two-factor solu-
tions should produce
two
groups (one
for
each factor)
of five
scores (one
from
each
data
set); within each group,
the five
scores should
be
very
highly
interrelated.
Table
3
presents mean convergent correlations (i.e., those
among scores within
the
same group that presumably
reflect
the
same factor)
for
each
factor
in
each solution.
As was
noted ear-
lier,
beyond three factors
it was
impossible
to
identify
any
factor
consistently
on the
basis
of
content;
we
therefore matched later
factors
in
such
a way as to
maximize
the
overall
level
of
con-
vergence
in
that
solution.1
1
It is
frequently
the
case
that
factors emerge
in
different
orders
in
different
solutions,
particularly
as
larger numbers
of
factors
are ex-

ANXIETY
AND
DEPRESSION SYMPTOM STRUCTURE
19
Table
3
Assessing
the
Cross-Sample
Convergence
of
One-
Through
Eight-Factor
Solutions: Mean Convergent Correlations
of
Factor
Scores From
the
Five Samples Computed
in
the
Student
3
Data
Number
of
factors
in
solution
1
2
3
4
5
6
7
8
Factor number
1
.99
.99
.99
.98
.92
.92
.93
.92
2
_
.99
.93
.93
.86
.84
.90
.91
3
.93
.92
.95
.94
.95
.94
4
—
.45
.59
.56
.49
.53
5
—
.57
.51
.74
.50
6
—
.18
.61
.76
7
8
.42 —
.61
.50
Note.
N =
522. Mean correlations
of .90 or
greater
are
shown
in
bold-
face.
Everett
(1983)
suggested that
a
correlation
of .90 or
greater
indicates that
the
factors truly converge with
one
another.
Ac-
cording
to
this criterion,
the
one-
and
two-factor solutions were
both
highly
convergent.
The five
scores generated
by the
one-
factor
solutions
had a
mean convergent correlation
of
.99; sim-
ilarly,
the
two-factor solutions yielded
two
groups
of
factors (one
from
each sample) that each
had an
average
coefficient
of
.99.
The
three-factor solution
is the
most crucial
for the
tripartite
model.
It is
noteworthy, therefore, that
the
mean convergent
correlations
for
this
solution—.99,
.93,
and
.93,
respectively—
easily
meet Everett's
(1983)
criterion.
In
contrast,
no
succeed-
ing
factor even
approached
an
acceptable level
of
convergence.
In
the
four-factor
solutions,
the
fourth factor
had an
average
convergent
correlation
of
only .45;
in
subsequent solutions,
no
factor
beyond
the
third
had a
mean
coefficient
above .80.
Another
interesting aspect
of
these
data
is
that
the first
three
factors
remained
highly
convergent even
as
more
and
more fac-
tors were extracted.
For
instance,
in the
eight-factor solution,
these
factors still
had
mean
coefficients
of
.92,
.91,
and
.94,
re-
spectively.
In
other words, extracting additional factors
did not
substantially
diminish
the
replicability
of the first
three. This
pattern
probably
reflects
the
earlier
finding
that
the first
three
factors
remained large
and
well-defined even
as
more factors
were
extracted (see Table
2).
Solely
on the
basis
of the
factor similarity
data,
one can
justify
retaining
one, two,
or
three factors.
All
three solutions yielded
structures that were
highly
convergent
across
the five
samples;
beyond
that,
the
structures diverged sharply. However, because
the
three-factor structure
was
predicted
theoretically—and
be-
tracted (for
a
discussion,
see
Everett, 1983). This
was
also true
in our
analyses.
For
instance,
in
some solutions
the
General
Distress
dimen-
sion
emerged
first,
followed
by the
Positive Emotionality versus
Depres-
sion
factor;
in
other solutions,
the
order
of
these
two
factors
was re-
versed.
Accordingly,
we
matched
the
factors
by the
content
of
their
marker
items, rather than simply using
the
order
in
which they emerged.
The
factor numbers shown
in
Table
3
reflect
the
order
in
which
the
factors
emerged
in the
Student
3
data.
cause
the
most differentiated structure
is
also
likely
to be the
most clinically
informative—we
selected this solution
for
fur-
ther examination.
Further Analyses
of
Convergence
Among
the
Three-
Factor
Solutions
As
predicted
by the
tripartite model,
the
dimensions com-
prising
the
three-factor structure appeared
to
consist
of a
non-
specific
distress factor that included many symptoms
of
both
constructs,
a
specific depression factor,
and a
specific anxiety
factor.
We
therefore labeled these factors General
Distress,
An-
hedonia Versus Positive
Affect,
and
Somatic Anxiety, respec-
tively.
Before
examining
the
content
of
these factors,
we
investi-
gated
the
structural convergence among
the five
samples
in
more detail.
Factor
score convergence between individual samples.
We
have
seen already that
the
three-factor solutions showed
an im-
pressive
level
of
convergence overall. However,
the
Table
3
data
do not
show
how
individual samples converged with
one an-
other.
In
this regard,
one
might wonder whether
the
three stu-
dent samples produced extremely similar three-factor solutions
but
were somewhat less convergent with
the
adult
and
patient
samples. Accordingly, Table
4
presents
the
convergent
corre-
lations
for
each
of the
individual factor scores
in the
Student
3
data.
Two
aspects
of the
results
are
particularly noteworthy. First,
virtually
all of the
individual factors showed strong con-
vergence. Overall,
26 of the 30
convergent correlations (87%)
were
.90 or
greater,
and
none
was
lower than .85. Second, con-
Table
4
Assessing
the
Cross-Sample
Convergence
of
the
Three-Factor
Structure: Convergent Correlations
of
Factor
Scores From
the
Five
Samples Computed
in the
Student
3
Data
Factor score
1
Factor
1
(Anhedonia
vs.
Positive
Affect)
1.
Student
1
2.
Student
2
3.
Student3
4.
Adult
5.
Patient
—
.99
.99
.99
.98
—
.99
.99
.99
—
.99 —
.98 .99 —
Factor
2
(General Distress)
1.
Student
1
2.
Student
2
3.
Student3
4.
Adult
5.
Patient
—
.94
.96
.88
.85
—
.97
.96
.94
—
.94 —
.90 .96 —
Factor
3
(Somatic Anxiety)
1
.
Student
1
2.
Student
2
3.
Students
4.
Adult
5.
Patient
—
.93
.94
.85
.86
—
.97
.96
.95
—
.93 —
.92 .97 —
Note.
N =
522. Correlations
of .90 or
greater
are
shown
in
boldface.

20
WATSON
ET AL.
Table
5
Assessing
the
Cross-Sample
Convergence
of
the
Three-Factor
Structure:
Congruence
Coefficients
Based
on the
Factor
Loadings
From
the
Five Solutions
Solution
1234!
Anhedonia
vs.
Positive
Affect
1
.
Student
1
2.
Student
2
3.
Students
4.
Adult
5.
Patient
—
.98
.99
.97
.94
—
.98
.97
.95
—
.98 —
.95
.95 —
General
Distress
1.
Student
1
2.
Student
2
3.
Students
4.
Adult
5.
Patient
—
.97
.97
.94
.95
—
.98
.96
.96
—
.97 —
.96 .97 —
Somatic
Anxiety
1.
Student
1
2.
Student
2
3.
Students
4.
Adult
5.
Patient
—
.93
.93
.87
.91
—
.97
.95
.95
—
.94 —
.94 .96 —
Note.
Congruence
coefficients
of .90 or
greater
are
shown
in
boldface.
vergence
between
the
student
and
nonstudent samples
was
only
slightly
lower than that among
the
various
student
groups.
The
mean
convergent correlations among
the
three student samples
were
.99
(Anhedonia
vs.
Positive
Affect),
.96
(General Distress),
and
.95
(Somatic Anxiety).
The
corresponding
coefficients
be-
tween
the
adult
and
student samples were .99, .93,
and
.91,
re-
spectively;
those between
the
patient
and
student samples were
.98, .90,
and
.91,
respectively. Finally,
the
adult
and
patient sam-
ples
were strongly convergent, yielding
correlations
of
.99, .96,
and
.97, respectively. Thus,
the
Table
4
data
indicate that stu-
dents, adults,
and
patients
all
generate extremely similar three-
factor
structures.2
Factor
loading
convergence.
As
mentioned earlier,
a
second
approach
to
factor similarity
is to
compute
congruence
coeffi-
cients
(Tucker,
1951)
on the
basis
of the
factor loadings
in
each
solution.
Congruence
coefficients
have
the
same range
as
corre-
lations (i.e.,
from
—1
to 1).
Moreover, similar
to
correlations,
factors
that
are
presumed
to be
convergent should have highly
positive
coefficients
with
one
another (i.e.,
.90 and
above).
It
should
be
noted, however, that unlike correlations, congruence
coefficients
reflect
not
only
the
rank order
and
scatter
of the
factor
loadings,
but
also their magnitude. Thus,
for a
congru-
ence
coefficient
to
approach
unity,
the
loadings
on two
factors
not
only must show
a
very
similar
pattern,
they must also
be
generally
similar
in
size (see also
Gorsuch,
1983; Harman,
1976).
Table
5
presents congruence
coefficients
among
the
factors
that were judged
to be
convergent. These
data
essentially con-
firmed
the
earlier
findings
that
were
based
on
factor
scores;
if
anything,
they
demonstrated
a
slightly higher level
of
replicabil-
ity.
Overall,
29 of the 30
congruence
coefficients
(97%) were
above
.90,
and
none
was
lower than .87. Furthermore, there
was
strong convergence across
the
student, adult,
and
patient sam-
ples, that
is, the
three student solutions
produced
mean congru-
ence
coefficients
of .96
(General Distress),
.97
(Anhedonia
vs.
Positive
Affect),
and .92
(Somatic
Anxiety)
with
the
adult fac-
tors,
and
corresponding values
of
.96, .95,
and
.93, respectively,
with
the
patient
factors. Similarly,
the
congruence coefficients
between
the
adult
and
patient factors were .97, .95,
and
.96,
respectively.
Clearly,
the
three-factor structure
was
highly
repli-
cable
across
the
different
types
of
participants.
Three
Replicated
Factors
Orthogonal
varimax
rotation.
Our
analyses demonstrated
an
impressive level
of
convergence
in the
three-factor structure
across
the five
samples.
Next,
we
considered
the
nature
of the
predicted structure
in
more detail
and
examined
the
extent
to
which
these
three
robust factors conformed
to the
symptom
groups
hypothesized
in the
tripartite model.
As
stated earlier,
we
expected
the
three-factor structure
to
consist
of (a) a
general distress factor reflecting symptoms
of
both anxiety
and
depression,
(b) a
specific
anxiety factor that
is
most strongly marked
by
symptoms
of
somatic tension
and
arousal,
and (c) a
specific depression factor
that
is
defined
on
one
end by
items reflecting energy, enthusiasm,
and
high
posi-
tive
affect,
and on the
other
end by
items
reflecting
anhedonia,
loss
of
interest,
and low
positive
affect.
In
terms
of
specific
scales
and
symptoms,
we
therefore predicted that
all 38
items com-
prising
the
three
GD
scales (GD: Mixed,
GD:
Anxiety,
GD:
Depression) would load primarily
on a
common general
distress
factor.
Note, however, that many
of
these items also might have
significant
secondary loadings (i.e.,
|
.301
or
greater)
on one of
the
specific
factors;
for
instance, some
of the GD:
Anxiety
symptoms
might load secondarily
on the
somatic anxiety
factor,
whereas
some
GD:
Depression items might load
significantly
on
the
specific depression factor.
In
addition,
we
predicted that
the 17
retained Anxious
Arousal
symptoms
all
would load primarily
on the
specific anx-
iety
factor; again, however, some
of
these items also might have
significant
secondary loadings
on
another factor.
No
predictions
were
made regarding
the two
items that were dropped
from
Anxious
Arousal.
Finally,
we
expected
the 24
High Positive
Affect
items
to de-
fine one end of the
specific depression
factor.
The
expected pat-
tern
for the
eight retained Loss
of
Interest items
was
less clear,
however.
As
noted earlier,
the
high-end items tend
to be
stronger,
purer markers
of the
underlying factor than
are
items reflecting
anhedonia
and low
Positive
Affect
(see Watson
et
al.,
1988;
Wat-
son &
Kendall,
1989).
Accordingly,
it is
uncertain whether
the
Loss
of
Interest items should
be
expected
to
load primarily
on
2
As
noted
earlier,
these
factor
scores
can be
computed
in any of our
data
sets.
Accordingly,
we
repeated these
analyses
in the
four
remaining
samples
and
obtained
virtually
identical
results. That
is, in the
other
four
samples
the
three
factors
produced
mean
convergent
correlations
ranging
from
.98 to .99
(Anhedonia
vs.
Positive
Affect),
from
.92 to .95
(General
Distress),
and
from
.92 to .95
(Somatic
Anxiety).
It is
interest-
ing
to
note
that
the
best
overall
convergence
was
obtained
using
the
patient
data
(mean
rs
=
.98, .95,
and
.95,
respectively).

ANXIETY
AND
DEPRESSION SYMPTOM STRUCTURE
21
the
specific
depression
factor
or,
alternatively,
on the
general dis-
tress
factor.
Clearly, however, these items should load
signifi-
cantly
on the
specific
depression factor; moreover, they should
have
relatively
stronger loadings
on
this factor than
the GD: De-
pression
symptoms.
With
these predicted patterns
in
mind, Table
6
presents
the
mean
varimax-rotated loading
for
each item (computed across
all
five
solutions)
on
each
of the
three replicated factors.
The
most noteworthy aspect
of
these
data
is
that although there
are
several
unpredicted
findings, the
overall structure
is
broadly
consistent
with
the
tripartite model. That
is, we see
clear evi-
dence
of (a) a
General Distress factor that
is
denned
by
many
symptoms
of
both depression
and
anxiety,
(b) a
specific
anxiety
factor
that
is
most strongly marked
by
numerous somatic items,
and (c) a
specific
depression factor that
is
characterized
by the
High
Positive
Affect
items
on one
pole
and by
various depressive
symptoms
on the
other.
We
now
consider each
of the
factors
in
more detail. First,
as
expected,
the
large majority
of the GD
items loaded strongly
on
the
General Distress factor. Overall,
29 of the 38 GD
symptoms
(76%)
loaded
significantly
on
this factor; moreover,
27 of
these
items
had
their highest loading
on it.
Support
for the
tripartite
model
was
particularly strong among
the GD:
Depression
symptoms,
all of
which were markers
of
this
factor.
Note, how-
ever,
that over half
of the
symptoms
on the
other
two GD
scales
(i.e.,
9 of 15 GD:
Mixed items,
6 of
11
GD:
Anxiety
items)
also
loaded most
highly
on
General Distress. Finally,
six
Loss
of
Interest
items
and the two
discarded Anxious Arousal symp-
toms also marked this
factor.
On
the
other hand,
several
of the GD
items
did not
behave
as
predicted.
One
reverse-keyed
GD:
Mixed item
("slept
very
well")
did not
load
significantly
on any
factor. Five additional
GD:
Mixed symptoms
had low to
moderate loadings (i.e.,
in the
.20
to .45
range)
on
both General Distress
and
Somatic
Anxiety.
The
most striking pattern,
however,
was
exhibited
by five so-
matic symptoms (e.g., "lump
in
throat,"
"tense
or
sore mus-
cles")
from
the GD:
Anxiety scale. Although clearly somatic,
these items were
not
placed
in
Anxious Arousal because they
did
not
appear
to
reflect
autonomic
hyperarousal
as
strongly
as
many
other anxiety symptoms. Contrary
to our
expectations,
however,
these items were markers
of the
specific anxiety factor
(with
loadings ranging
from
.37 to
.54),
and did not
load sig-
nificantly
on
General Distress (loadings ranged
from
only.
11
to
.24).
Turning
to
Somatic Anxiety, Table
6
indicates that
16
of the
17
retained Anxious Arousal items (94%) were clear markers
of
this
factor,
with loadings ranging
from
.39 to
.66;
the
only item
that
did not
show
the
expected pattern ("easily
startled")
split
evenly
between this factor
and
General Distress.
In
addition,
as
described earlier,
five
somatic
GD:
Anxiety symptoms loaded
primarily
on
this factor. Finally, seven items
from
other scales
(five
from
GD:
Mixed,
two
from
Loss
of
Interest) also were
markers
of
this dimension. Thus,
the
factor that emerged
was
somewhat broader than expected; most notably,
it
included sev-
eral
somatic items that
do not
appear
to
reflect
a
strong
state
of
perceived arousal. Having said this, however,
we
must also
emphasize
that
the
Anxious Arousal scale
contributed
14
of the
16
items that loaded
.50 or
higher
on
this factor.
In
other words,
the
strongest,
clearest
markers
of
this
factor were,
in
fact,
the
symptoms
predicted
by the
model.
Finally,
as
expected,
the
specific
depression
factor
was the
only
one
that
was
strongly bipolar. Consistent with
our
predic-
tion,
23 of the 24
High Positive
Affect
items (96%) clearly
de-
fined the
high
end of
this
factor,
with loadings ranging
from
.47
to .76
(the
one
deviant item, "felt
I
didn't
need much
sleep,"
failed
to
load
significantly
on any
factor).
In
addition,
10
symp-
toms
had
significant
secondary loadings
on the low end of
this
factor:
Six
were
from
GD:
Depression, three were
from
Loss
of
Interest,
and one was
from
GD:
Mixed.
Put
another way, nine
of
the 20
depression symptoms (45%; this
figure
excludes
the
one
dropped Loss
of
Interest item)
had
significant
secondary
loadings
on
this dimension.
In
contrast,
no
anxiety symptoms
loaded significantly
on
this factor;
in
fact,
the
mean loading
across
the 30
items that were originally included
in
either
GD:
Anxiety
or
Anxious Arousal
was
only
—.05.
These
findings
strongly
support
the
identification
of
this dimension
as a
specific
depression factor that
is
unrelated
to
anxiety.
It
is
also noteworthy that
the GD:
Depression
and
Loss
of
Interest items tended
to
load quite similarly
on
this factor.
In
fact,
the 12 GD:
Depression symptoms
had
loadings ranging
from
—.
19
to
-.35,
with
a
mean value
of—.28,
whereas
the
eight
retained Loss
of
Interest symptoms
had
loadings ranging
from
—.17
to
—.40,
with
an
average value
of
-.27. Thus,
we see no
evidence
that
the
Loss
of
Interest items were more strongly
re-
lated
to the
specific
depression
factor.
However, consistent with
our
model, these items tended
to be
less strongly saturated with
general
distress variance. That
is, the GD:
Depression items
had
loadings ranging
from
.41
to .64 on the
General Distress
factor,
with
an
average value
of
.55;
in
contrast,
the
corresponding
loadings
for the
eight retained Loss
of
Interest items ranged
from
. 14 to
.49, with
a
mean
of
.40. Hence, consistent with
our
prediction,
the
Loss
of
Interest items have
a
higher proportion
of
specific
factor variance.
Obliquepromax
rotation.
One
could argue that oblique
ro-
tation
(in
which
the
factors
are
allowed
to be
correlated) might
provide
a
more realistic representation
of the
symptom struc-
ture
in
this domain. Accordingly,
we
also subjected
the
three-
factor
solutions
to
oblique promax rotations
in
which
the
vari-
max
loadings were raised
to a
power
of 3
(see Gorsuch,
1983;
Hendrickson
&
White,
1964).
The
resulting factors correlated
.49
(General Distress
vs.
Anhedonia/Positive
Affect),
.58
(Gen-
eral Distress
vs.
Somatic
Anxiety),
and .23
(Anhedonia/Positive
Affect
vs.
Somatic
Anxiety).
Nevertheless,
these
oblique
rota-
tions produced factors that
are
highly similar
to
those displayed
in
Table
6. The
only notable
difference
was
that
the
Anhedonia/
Positive
Affect
factor
was
less strongly bipolar
in the
oblique
solutions:
Specifically,
although this factor continued
to be
strongly
defined
by the
High Positive
Affect
items
on one
end,
the
depression symptoms
had
weaker loadings
on the
other.
Discussion
Evidence Regarding
the
Tripartite
Model
The
results
of
this study
offer
broad support
for the
tripartite
model proposed
by L. A.
Clark
and
Watson
(1991).
In
this
model, symptoms
of
depression
and
anxiety
are
divided into

22
WATSON
ET AL.
Table
6
Mean
Varimax-Rotated
Factor
Loadings
of
the
MASQ
Items
Averaged
Across
the
Five
Solutions
MASQ Scale/item
General
Distress:
Mixed Symptoms
Worried
a lot
about
things
Trouble concentrating
Felt dissatisfied with things
Felt confused
Felt irritable
Trouble making decisions
Trouble paying attention
Felt restless
Felt something
awful
would happen
Got
fatigued easily
Trouble remembering things
Trouble
falling
asleep
Trouble staying asleep
Loss
of
appetite
Slept very
well"
General Distress: Depressive Symptoms
Felt
depressed
Felt discouraged
Felt
sad
Felt hopeless
Disappointed
in
myself
Felt like crying
Felt like
a
failure
Felt worthless
Blamed
myself
for
things
Felt
inferior
to
others
Pessimistic
about
the
future
Felt tired
or
sluggish
General Distress: Anxious Symptoms
Felt tense, "high-strung"
Felt uneasy
Felt nervous
Felt
afraid
Felt
"on
edge,"
keyed
up
Unable
to
relax
Lump
in my
throat
Upset stomach
Tense
or
sore muscles
Felt nauseous
Had
diarrhea
Loss
of
Interest
Felt unattractive
Felt nothing
was
enjoyable
Felt withdrawn from
others
Took extra
effort
to get
started
Felt slowed down
Nothing
was
interesting
or fun
Felt bored
Thought about death, suicide
Felt like being
aloneb
Anxious
Arousal
Felt dizzy, lightheaded
Was
trembling, shaking
Shaky
hands
Trouble swallowing
Short
of
breath
Dry
mouth
Twitching
or
trembling muscles
Hot or
cold spells
Cold
or
sweaty hands
Felt
like
I was
choking
General
Distress
.63*
.60*
.59*
.55*
.53*
.52*
.49*
.45*
.44*
.40
31
.29
.25
.22
-.16
.64*
.61*
.60*
.59*
.58*
.57*
.57*
.55*
.54*
.54*
.44*
.41*
.57*
.55*
.54*
.51*
.51*
.50*
.24
.23
.22
.20
.11
.49*
.48*
.47*
.43*
.39
.35*
.32*
.28
.14
.19
.25
.23
.04
.15
.18
.19
.22
.13
.02
Mean loading
on
Anhedonia-
Positive
Affect
-.22
-.08
-.33
-.14
-.20
-.09
-.05
.05
-.21
-.20
-.06
-.05
-.11
-.02
.26*
-.35
-.31
-.27
-.34
-.28
-.23
-.32
-.32
-.20
-.19
-.30
-.19
.01
-.19
-.04
-.08
.04
-.09
-.09
-.05
.01
-.06
.00
-.24
-.40
-.33
-.19
-.24
-.32
-.17
-.25
.19*
-.04
-.07
-.09
-.08
.00
-.03
.01
-.05
-.05
-.09
Somatic
Anxiety
.17
.33
.24
.23
.28
.29
.38
.29
.36
.42*
.39*
.35*
.40*
.31*
-.21
.18
.16
.10
.25
.18
.17
.22
.20
.21
.21
.17
.36
.32
.31
.22
.18
.38
.31
.54*
.53*
.42*
.47*
.37*
.19
.30
.27
.27
.41*
.28
.19
.34*
.03
.66*
.63*
.58*
.57*
.56*
.55*
.55*
.52*
.52*
.51*

ANXIETY
AND
DEPRESSION SYMPTOM STRUCTURE
23
Table
6
(continued)
Mean
loading
on
MASQ Scale/item
Felt
faint
Pain
in
chest
Racing
or
pounding heart
Felt numbness
or
tingling
Afraid
I was
going
to die
Had to
urinate
frequently
Was
afraid
I was
losing
control1"
Felt
like
I was
going
crazyb
Easily
startled
High
Positive
Affect
Felt
really
lively,
"up"c
Felt
really
happyc
Felt
I had a lot of
energy0
Was
having
a lot of
func
Felt
I had
much
to
look
forward
toc
Felt good about myself
I
had
many interesting things
to
doc
Felt
confident
Looked
forward
to
things0
Felt
I had
accomplished
a
lot0
Was
proud
of
myself0
Felt
cheerful0
Felt
successful
Felt
optimistic0
Felt
really
talkative
Moved
quickly
and
easily0
Felt
hopeful
about
future0
Able
to
laugh easily
Felt
like
being
with
others
Felt
very
clearheaded
Thoughts
came
to me
very
easily
Felt
very
alert
Could
do
everything
I
needed
to
Felt
I
didn't need much sleep
General
Distress
.17
.08
.34
.13
.14
.22
.46*
.55*
.31*
-.08
-.16
-.08
-.09
-.15
-.32
-.13
-.34
-.10
-.19
-.23
-.13
-.24
-.14
.09
-.10
-.19
-.07
.00
-.24
-.11
-.15
-.29
.07
Anhedonia-
Positive
Affect
-.05
-.07
.09
-.03
-.09
.06
-.17
-.17
.03
.76*
.72*
.71*
.69*
.68*
.68*
.66*
.65*
.64*
.63*
.63*
.62*
.62*
.59*
.58*
.57*
.56*
.53*
.52*
.52*
.51*
.49*
.47*
.15
Somatic
Anxiety
.51*
.51*
.51*
.50*
.39*
.39*
.38
.32
.31*
-.06
-.08
-.07
.05
-.07
-.04
-.02
-.04
-.03
.00
.03
-.11
.03
.00
-.02
-.12
-.03
-.11
-.13
-.16
-.09
-.18
-.07
.20*
Note.
Loadings
of |
.301
or
greater
are
shown
in
boldface.
An
asterisk indicates
the
highest loading
for
that
item. MASQ
=
Mood
and
Anxiety Symptom Questionnaire.
*
Reverse-keyed item.
b
Item
was
originally included
in
scale
but
later eliminated;
see
text
for
more details.
0
Selected
as a
reverse-keyed item
for the
Anhedonic Depression scale.
three groups: nonspecific symptoms
of
general distress,
symp-
toms
of
anhedonia
and low
positive
affect
that
are
relatively
unique
to
depression,
and
manifestations
of
somatic tension
and
arousal that
are
relatively specific
to
anxiety. Consistent
with
this model,
our
analyses
of the
MASQ items demonstrated
that
the
same three symptom factors emerged
in
each
of five
samples.
Moreover, these factors converged
well
with
the
symptom
groups hypothesized
in the
model.
As
predicted,
one of the
fac-
tors
(General Distress)
was
nonspecific
to
depression
and
anxi-
ety.
It was
defined
by a
broad
range
of
symptomatology, includ-
ing
several items
from
each
of the
general
distress
scales.
It is
especially noteworthy
that—consistent
with
prediction—items
reflectiflg
both anxious (e.g., "felt afraid," "felt
nervous,"
"felt
uneasy")
and
depressed
(e.g., "felt
depressed,"
"felt sad")
affect
were
strong markers
of
this factor. This factor clearly
taps
vari-
ance that
is
common
to
depression
and
anxiety.
As
predicted, each
of the
other symptom factors
was
more
specifically
related
to one of the
constructs. That
is, the
Somatic
Anxiety
factor
was
defined largely
by
somatic manifestations
of
anxiety.
Note
that
all
16
of the
items loading
.50 or
greater
on
this factor were somatic symptoms
of
anxiety
(14
from
Anxious
Arousal,
2
from
GD:
Anxiety);
in
contrast,
only
two
depression
items ("felt slowed down,"
"thought
about
death,
suicide")
were
markers
of
this dimension. Conversely,
the
specific depression
factor
was
defined
by
positive emotionality items
on its
high
end,
and by
various symptoms
of
depression (e.g., "felt nothing
was
enjoyable," "felt
hopeless,"
"nothing
was
interesting
or
fun,"
"felt
depressed")
on the
other.
The
specificity
of
this
di-
mension
was
clearly demonstrated: none
of the
anxiety
symp-
toms
loaded
significantly
on it.
However,
although
the
factor analytic
data
strongly supported
the
broad
outlines
of the
tripartite
model, many items showed
factor
loading patterns that
differed
significantly
from
our
the-
oretical
predictions.
In
this regard,
the
most
striking
finding was
that several somatic symptoms that were
predicted
to be
mark-
ers
of
General
Distress actually were
clear
markers
of the
spe-
cific
anxiety factor. These
results
strongly suggest
that
our
con-

24
WATSON
ET AL.
ceptualization
of the
specific anxiety symptom group overem-
phasized
the
importance
of
perceived autonomic hyperarousal
as the
centrally
denning
feature;
in
actuality,
the
specific factor
that
emerged
was
defined
by a
broader range
of
somatic symp-
toms, including several that
do not
clearly
reflect
autonomic
hyperarousal
(e.g., nausea, diarrhea).
Thus,
our
data
simultaneously demonstrate both
(a)
broad
support
for the
tripartite model
and (b) the
need
for
further
refinements
and
modifications
to it.
Moreover, they indicate
that
although most
of the
MASQ items were
put in the
most
appropriate scales, some were placed incorrectly. This,
in
turn,
suggests
the
need
for
further refinements
of the
MASQ
scales.
We
have,
in
fact,
already conducted some exploratory revisions.
For
instance,
we
created
an
expanded Anxious Arousal scale
by
adding
the five
somatic
GD:
Anxiety
symptoms that were
markers
of the
Somatic Anxiety
factor,
and an
expanded Anhe-
donic
Depression scale
by
adding
the six GD:
Depression symp-
toms
with
significant
secondary loadings
on the
specific
depres-
sion
factor (see Table
6).
Preliminary analyses, however, indi-
cated that these augmented scales
did not
show
significantly
better convergent
and
discriminant validity than
the
originals.
Nevertheless,
further
examination
of
this
issue—together
with
further
conceptual refinements
in the
tripartite structure
it-
self—is
an
important task
for
future
research.
Replicability
of
Symptom
Structure
Our
findings
also have important implications that
are
unre-
lated
to the
tripartite model. Most notably,
we
have demon-
strated that
the
basic symptom structure
in
this domain
(at
least
as it is
operationalized
in the 90
MASQ items)
is
highly
con-
vergent
across college student, normal adult,
and
psychiatric
pa-
tient
samples.
Specifically,
our
data show that extremely similar
one-,
two-,
and
three-factor structures
can be
identified
in di-
verse
samples.
After
three factors
the
individual solutions
di-
verged
sharply,
so
that
no
additional factors could
be
consis-
tently
identified
in
every data set. Thus,
the
crucial
finding of
substantial
replicability
was
obtained
at the
basic factor level.
This
replicability obviously increases one's confidence
in the
tripartite model. More
fundamentally,
however,
it
suggests
that
the
basic symptom structure
in
this domain
is
itself robust
across
different
populations. Much
of the
research
in
this area
has
been based
on
patient data,
but
countless studies
of
depres-
sion
and
anxiety
have used college student samples.
It is
reas-
suring,
therefore,
to
have clear evidence that these
different
pop-
ulations
may
yield
substantially similar results,
at
least
in
terms
of
structural analyses.
In
other words,
on the
basis
of our
results,
it
appears that basic structural analyses conducted with college
students
will
generalize reasonably
well
to
adult
and
patient
samples. Conversely, structural analyses involving clinical
pa-
tients
can be
expected
to
replicate
in
nonclinical samples.
Clearly,
our
results themselves require replication using other
measures
and
different
samples; nevertheless, they provide pre-
liminary
evidence
of an
underlying coherence
in
symptom
structure
across
different
populations.
Limitations
of
the
Study
We
must note
two
limitations
of our
study.
First,
our
struc-
tural
analyses demonstrated
an
impressive level
of
convergence
across
five
samples,
but
they were
confined
to a
single
set of
self-
rated symptoms. Although these
90
items appear
to
assess this
domain more comprehensively than many existing instru-
ments, they
may not
cover
it
completely
or
optimally.
It is
cer-
tainly
possible,
for
instance, that
certain
types
of
symptoms
are
underrepresented relative
to
others. Thus,
it is
important that
the
current results
be
replicated using other symptom
measures.
Second,
our
analyses included only
one
clinical sample.
Moreover,
this
sample—composed
primarily
of
male patients
with
substance
use
disorders (Watson
et
al.,
1995)—is
less than
optimal
for a
study involving structural analyses
of
anxious
and
depressive symptomatology.
It is
possible
that other patient
groups would show somewhat
different
results,
and
that they
might
not
converge
as
well with
the
student
and
adult samples.
Accordingly,
our
results require replication using other patient
groups.
Conclusion
We
hope that
our findings
stimulate
further
investigation
of
the
issues addressed
in
this
article.
Specifically,
we
hope
to
have
encouraged
further
research into
(a) the
tripartite model
of de-
pression
and
anxiety
and (b) the
replicability
of
symptom struc-
ture across
different
populations. Despite
the
limitations
we
have
noted,
the
clarity
and
consistency
of our
data
suggest
that
these topics warrant
further
study.
References
American
Psychiatric Association.
(1987).
Diagnostic
and
statistical
manual
of
mental
disorders
(3rd ed.,
rev.).
Washington,
DC:
Author.
Clark,
D.
A.,
Beck,
A. X, &
Stewart,
B.
(1990).
Cognitive specificity
and
positive-negative
affectivity:
Complementary
or
contradictory views
on
anxiety
and
depression!
Journal
of
Abnormal
Psychology,
99,
148-
155.
Clark,
L. A.
(1989).
The
anxiety
and
depressive
disorders:
Descriptive
psychopathology
and
differential
diagnosis.
In P. C.
Kendall
& D.
Watson
(Eds.),
Anxiety
and
depression:
Distinctive
and
overlapping
features
(pp.
83-129).
San
Diego,
CA:
Academic
Press.
Clark,
L.
A.,
&
Watson,
D.
(1991). Tripartite model
of
anxiety
and de-
pression: Psychometric evidence
and
taxonomic implications.
Jour-
nal
of
Abnormal
Psychology,
100,
316-336.
Costa,
P.
T,
Jr.,
&
McCrae,
R. R.
(1992).
Revised
NEO
Personality
Inventory
(NEO-PI-R)
and NEO
Five-Factor
Inventory
(NEO-FFI)
professional
manual. Odessa,
FL:
Psychological Assessment
Resources.
Everett,
J. E.
(1983).
Factor comparability
as a
means
of
determining
the
number
of
factors
and
their rotation.
Multivariate
Behavioral
Re-
search,
18,
197-218.
Gorsuch,
R. L.
(1983).
Factor
analysis (2nd
ed.).
Hillsdale,
NJ:
Erlbaum.
Gotlib,
I. H.
(1984).
Depression
and
general psychopathology
in
univer-
sity
students. Journal
of
Abnormal
Psychology,
93,
19-30.
Harman,
H. H.
(1976).
Modern
factor
analysis
(3rd
ed.).
Chicago: Uni-
versity
of
Chicago
Press.
Hendrickson,
A.
E.,
&
White,
P.
Q
(1964).
Promax:
A
quick method
for
rotation
to
oblique simple structure. Journal
of
Statistical Psy-
chology,
17,65-10.
Kendall,
P.
C.,
&
Watson,
D.
(Eds.).
(1989).
Anxiety
and
depression:
Distinctive
and
overlapping
features.
San
Diego,
CA:
Academic Press.

ANXIETY
AND
DEPRESSION SYMPTOM STRUCTURE
25
Maser,
J., &
Cloninger,
C. R.
(Eds.).
(1990).
Comorbidity
in
anxiety
and
mood
disorders.
Washington,
DC:
American Psychiatric Press.
Mendels,
J.,
Weinstein,
N., &
Cochrane,
C.
(1972).
The
relationship
between depression
and
anxiety. Archives
of
General Psychiatry,
27,
649-653.
Moras,
K.,
DiNardo,
P.
A.,
&
Barlow,
D. H.
(1992).
Distinguishing anx-
iety
and
depression: Reexamination
of the
reconstructed Hamilton
scales.
Psychological
Assessment,
4,
224-227.
Sanderson,
W.
C.,
Beck,
A.
T,
&
Beck,
J.
(1990).
Syndrome comorbidity
in
patients with major depression
or
dysthymia: Prevalence
and
tem-
poral
relationships.
American Journal
of
Psychiatry,
147,
1025-1028.
Tellegen,
A.
(1985).
Structures
of
mood
and
personality
and
their rele-
vance
to
assessing anxiety, with
an
emphasis
on
self-report.
In A. H.
Tuma
& J. D.
Maser
(Eds.),
Anxiety
and the
anxiety disorders
(pp.
681-706).
Hillsdale,
NJ:
Erlbaum.
Tucker,
L. R.
(1951).
A
method
for
synthesis
of
factor
analysis studies
(Personnel
Research Section Report
No.
984). Washington,
DC: De-
partment
of the
Army.
Watson,
D.,
&
Clark,
L. A.
(1991).
The
Mood
and
Anxiety
Symptom
Questionnaire.
Unpublished manuscript, University
of
Iowa,
Depart-
ment
of
Psychology,
Iowa City.
Watson,
D.,
Clark,
L.
A.,
&
Carey,
G.
(1988).
Positive
and
negative
affectivity
and
their relation
to
anxiety
and
depressive
disorders.
Jour-
nal
of
Abnormal
Psychology,
97,
346-353.
Watson,
D.,
&
Kendall,
P. C.
(1989).
Understanding anxiety
and
depres-
sion: Their relation
to
negative
and
positive
affective
states.
In P. C.
Kendall
& D.
Watson
(Eds.),
Anxiety
and
depression:
Distinctive
and
overlapping
features (pp.
3-26).
San
Diego,
CA:
Academic Press.
Watson,
D.,
&
Tellegen,
A.
(1985).
Toward
a
consensual structure
of
mood.
Psychological
Bulletin,
98,
219-235.
Watson,
D.,
Weber,
K.,
Assenheimer,
J. S.,
Clark,
L.
A.,
Strauss,
M. E.,
&
McCormick,
R. A.
(1995).
Testing
a
tripartite model:
I.
Evaluating
the
convergent
and
discriminant validity
of
anxiety
and
depression
symptom scales. Journal
of
Abnormal
Psychology,
104,
3-14.
Wolfe,
V.
V.,
Finch,
A.
J.,
Jr., Saylor,
C. E,
Blount,
R.
L.,
Pallmeyer,
T. P.,
&
Carek,
D. J.
(1987).
Negative
affectivity
in
children:
A
multitrait/
multimethod investigation. Journal
of
Consulting
and
Clinical Psy-
chology,
55,
245-250.
Zinbarg,
R.
E.,
&
Barlow,
D. H.
(1991). Mixed anxiety-depression:
A
new
diagnostic category?
In R. M.
Rapee
& D. H.
Barlow
(Eds.),
Chronic
anxiety:
Generalized
anxiety disorder
and
mixed anxiety-
depression
(pp.
136-152).
New
York:
Guilford Press.
Zinbarg,
R.
E.,
Barlow,
D.
H.,
Liebowitz,
M.,
Street,
L.,
Broadhead,
E.,
Katon,
W,
Roy-Byrne,
P.,
Lepine,
J.-P., Teherani,
M.,
Richards,
J.,
Brantley,
P.
J.,
&
Kraemer,
H.
(1994).
The
DSM-IV
field
trial
for
mixed
anxiety-depression.
American
Journal
of
Psychiatry,
151,
1153-1162.
Received
January
25,
1993
Revision
received June
6,
1994
Accepted
June
6,
1994
•
New
Editors
Appointed,
1996-2001
The
Publications
and
Communications Board
of the
American Psychological Association
announces
the
appointment
of
three
new
editors
for
6-year terms beginning
in
1996.
As of
January
1,
1995, manuscripts should
be
directed
as
follows:
• For
Behavioral
Neuroscience,
submit manuscripts
to
Michela Gallagher, PhD,
Department
of
Psychology, Davie Hall,
CB#
3270,
University
of
North Carolina,
Chapel Hill,
NC
27599.
• For the
Journal
of
Experimental
Psychology:
General,
submit manuscripts
to
Nora
S.
Newcombe,
PhD, Department
of
Psychology, Temple University,
565
Weiss Hall,
Philadelphia,
PA
19122.
• For the
Journal
of
Experimental Psychology: Learning, Memory,
and
Cognition,
submit
manuscripts
to
James
H.
Neely,
PhD, Editor, Department
of
Psychology, State
University
of New
York
at
Albany, 1400 Washington Avenue, Albany,
NY
12222.
Manuscript submission patterns make
the
precise
date
of
completion
of
1995 volumes
uncertain.
The
current editors, Larry
R.
Squire, PhD, Earl Hunt, PhD,
and
Keith Rayner,
PhD, respectively, will
receive
and
consider
manuscripts until December
31,1994.
Should
any
of the
volumes
be
completed before that
date,
manuscripts will
be
redirected
to the new
editors
for
consideration
in
1996
volumes.