ERIC ED335406: The Beck Depression Inventory: Testing for Invariant Measurement and Structure across Gender for Nonclinical Adolescents.

31 Pages·1991·0.44 MB·English

by  ERIC

DOCUMENT RESUME
TM 017 089
ED 335 406
Byrne, Barbara M.; And Others
AUTHOR
The Beck Depression Inventory: Testing for Invariant
TITLE
Measurement and Structure across Gender for
Nonclinical Adolescents.
PUB DATE
91
31p.; Paper presented at the Annual Meeting of the
NOTE
American Psychological Association (San Francisco,
CA, August 16-20, 1991).
Reports - Research/Technical (143) --
PUB TYPE
Speeches/Conference Papers (150)
MF01/PCO2 Plus Postage.
EDRS PRiCE
*Adolescents; *Affective Measures; Analysis of
DESCRIPTORS
Covariance; Comparative Testing; Factor Structure;
Females; Foreign Countries; Item Analysis; Males;
Secondary Education; *Secondary School Students; *Sex
Differences; *Test Items
*Beck Depression Inventory; Canada; Covariance
IDENTIFIERS
Structural Analysis; *Invariance Principle
ABSTRACT
Extending the aarlier work of B. M. Byrne and F.
Baron (1990), the factorial invariance of the 21-item Beck Depression
Inventory (BDI) was tested using 351 non-clinical adolescent males
and 334 non-clinical adolescent females. All subjects were in grades
9 through 12 and attended the same secondary school in a large
metropolitan area in central Canada. Participants, who ranged in age
from 12 to 18 years, completed the BDI and other assessment measures
during one regular class period. All analyses were based on
confirmatory factor analytic procedures within the framework of
covariance structure modeling. Despite the differential loading
pattern of Item 20 (hypochondria) for females, a three-factor
structure that comprised one second-order general factor of
depression and three first-order factors representing negativta
attitudes, performance difficulty, and somatic elements was found to
be invariant across gender. Furthermore, except for two additional
inventory items (items 8 and 10, self-accnsation and crying,
respectively), the remaining BDI measurements were group-invariant.
The results are expected to be of substantial interest to both
researchers and zlinicians whose concerns focus on depression as it
affects adolescents. A 40-item list of references is included. Two
data tables and two figures are provided. Items comprising the BDI
are appended. (RLC)
***********************************************************************
Reproductions supplied by EDRS are the best that can be made
from the original document.
**********************************************************************
U.S. DEPARTMENT Of EDUCATION
"PERMISSION TO REPRODUCE THIS
OflCe of Educaloonal Ramearch end improvement
MATERIAL HAS BEEN GRANTED BY
EDUCATIONAL RESOURCES INFORMATION
CENTER (ERIC)
ae(is
document has been reproduced as
received from the person or oispnintion
origmOnd
C Minor changes have been made to improve
reproduction Quality
Points ot view Or opinions stated in tole docu-
TO THE EDUCATIONAL RESOURCES
ment do not neceSaanly represent official
OERI position or pOliCy
INFORMATION CENTER (ERIC)."
tt.
The Beck Depression Inventory: Testing for Invariant
f /
ftook
Measurement ani Structure Across Gender for Nonclinical
,A=N
Adolescents
*am
po:T4
Barbara M. Byrne
T. Leanne Campbell
Pierre Baron
University of Ottata
Paper presented at the knerican Psychological Association
Annual Meeting, San Francisco 1991
2
REST COPY IMAM F
BDI Gender Invariance
9
Abstract
Extending the earlier work of Byrne and Baron (1990), the
factorial invariance of the Beck Depression Inventory (BDI) was
(n.351) and females
tested across nonclinical adolescent males
(n.334) using confirmatory factor analytic procedures. Despite
the differereial loading pattern of Item 20 for females, a
3-factor structure that comprised one 2nd-order general factor
of depression and three lst-order factors representing negative
attitudes, performance difficulty, and somatic* elements was
found to be invariant across gender. Furthermore, except for
BDI measurements
two additional items (#8, #10), the remaining
Results are expectecito be of substantial
were group-invariant.
interest to both researchers and clinicians whose concerns
focus on depression as it bears on this population.
BDI Gender Iuvariance
3
7actorial Validity and
Measuring Adolescent Depression:
Inventory Across Gender
Invarance of the Beck Depression
Inventory (BDI; :ieck, Ward,
Although the Beck Depression
1961) was originally developed for
Mendelson, Mock, & Erbaugh,
populations, a review of the literature
use with clinical
measuring instrument for
reveals its popularity as a
well. In particular, the BDI has
nonclinical populations as
in the assessment of depression
been used almost exclusively
Cross-validated findings from a
for nonclinical adolescents.
study of the BDI for this population
recent factor analytic
support for an
(byrne & Baron, 1990) demonstrated strong
of one higher-order general
hierarchical structure consisting
lower-order factors
factor of depression, and three
Performance Difficulty, and
representing Negative Attitudes,
study extends the work of Byrne
Somatic Elements. The present
the equivalence of BDI item
and Baron by testing for
second-order factorial structure across
measurements and a
females.
adolescent males and
epidemiological studies of depression
A frequent finding in
significantly higher rate of
is that adult females report a
has typically been
of 2 to 1
depression than males; a ratio
(Hirschfield &
and nonclinical samples
found for both clinical
1987; Weissman & Klerman, 1977).
Cross, 1982; Nolen-Hoeksema,
have recently been reported for
Indeed, similar findings
BDI Gender Invariance
4
Lewinsohn, & Hops, 1990; Baron &
adolescents (Allgood-Merten,
& Baron, 1990; Reynolds, 1986;
Perron, 1986; Campbell, 3yrne,
P,utter, Izard, a Read, 1986).
differences in
Important assumptions in testing for gender
BDI, however, are that (a) all
depression scores based on the
gender (i.e., all
item scaling units are equivalent across
exactly the same way by males and
items are interpreted in
is equivalent across
females), and (b) the factoiial structure
the facets of
.1ales and females (i.e., relation.s among
gender). Despite the fact
depression remain consistent across
clearly invilidates
that violation of these assumptions
study to
interpretations based on mean score differences, no
of these equivalencies.
date has tested the validity
be, as demonstrated in recent
in this regard it may
research related to adolescent self-concept,
construct validity
1987), that interrelations
(Byrne, 1990; Byrne & Shavelson,
depression vary somewhat for adolescent
among the facets of
FurthermOre, it'is possible that a different
males and females.
of
the underlying factorial structure
pattern of items defines
findings would suggest either a
the 13DI for each sex. Such
differential
differential perception of item content, or a
.
gender.
nomological network of depression structure across
validity of adolescent
These questions bear on the construct
hypotheses that can be
depression and can be formulated into
the task of the present study,
tested statistically. This was
BDI Gender Invariauce
5
factorial structure of the BDI proposed by
as it related to the
Byrne and Baron (1990).
(1994),
Building upon the earlier work of Tanaka and Huba
factor analyses (CFAs)
Byrne and Baron (1990) used confirmatory
modeling to test
within the framework of covariance structure
In contrast to Tanaka
the validity of a 2nd-order structure.
based on clinical adult data,
and Huba, whose validity work was
samples of nonclinical
Byrne and Baron used three independent
for, and cross-validate
adolescents (mean age=15 yrs) to test
BDI. Consistent with the Tanaka
the factorial validity of the
found an hierarchical
and Huba findings, Byrne and Baron
the data,
factorial structure to most adequately represent
the Clinical and
albeit certain item loadings differed for
basis of their cross-validated
nonclinical populations. On the
anderlying
findings, Byrne and Baron concluded that the
comprises one 2nd-order v.neral
structure of BDI responses
1st-order factors which they
factor of depression, and three
Performance Difficulty, and
labelled Negative Attitudes,
schematically in
Somatic Elements. This model is presented
Figure 1.
Insert Figure 1 about here
Baron (1990), all previous
Except for the work of Byrne and
nonclinical
Cactor analyses of the BDI, as they relate to the
BDI Gender Invariance
6
(EFA) procedures
adolescent population, have used exploratory
involving principal components analyses with varimax rotation.
7indings from this research, however, have been inconsistent;
2-factor (Shek, 1990), 3-factor (Baron & LaPlante, 1984), and
4-factor (Teri, 1932) solutions have been reported. This
discordance can be linked to the widely known limitations
(e.g., 3ollen, 1989;
associated with EFA procedures in general
Long, 1983), and prir:ipal components analyses in particular
1990; Hubbard
(e.g., Borgatta, Kercher, & Stull, 1986; Gorsuch,
Allen, 1987; Snook & Gorsuch, 1989).
The present study addressed the methodological weaknesses
of previous factor analytic and gender difference research by
assessing the equivalency of BDI measurement and structure
females based on a rigorously
across adolescent males and
analyses were
tested and cross-validated factorial model. All
based on CFA procedures within the framework of covariance
structure modeling.
Method
Sam le and Procedures
The data comprised BDI responses from 685 (males, n=351;
the same
females, n=334) adolescents (grades 9-12) attending
secondary school in a large metropolitan area in central
18 years (I=15
Canada. Participants ranged in age from 12 to
included in
years), Only questionnaires with complete data were
the analyses.
7
BDI Gender Invariance
7
Subjects completed the BDI, along with other assessment
all testing
measures, during one regular class period;
materials were completed anonymously. Test instructions were
paraphrased by the test administrator, and procedural ques
were solicited and answered. All participation, in keeping with
school and Ethics Committee policies, was voluntary and no
incentives were offered.
Instrumentation
is a 21-item scale that measures symptoms relateC
The
to cognitive, behavioral, affective, and somatic components
of
depression. Although originally designed for use by trained
interviewers, it is now most typically used as a self-report
Crrickshaw,
measure (Beck, Steer, & Garbin, 1988; Kearns,
McGuigan, Riley, Shaw, & Snaith, 1982; Vredenburg, Krames, &
Flett, 1985). For each item, respondents are presented
four statements rated from 0 to 3 in terms of intensity, and
asked to select the one which most accurately des-Lribes their
level of
own feelings; higher scores represent a more severe
reported depression. Total scores range from 0 to 63 and are
used to categorize four levels of depression: none to minimal
(0-9), mild to moderate (10-18), moderate to severe (19-29),
and severe (30-63) (Beck et al., 1988). Zero-rated statements
for BDI items are presented in the Appendix.
Previous research has demonstrated the BDI to be
psychometrically sound in measuring depression for nonclinical
BDI Gender Invariance
relir'ility of
adolescents. Tests for the internal consistency
coefficients
the 11)I for this population have revealed alpha
Laplante, 1934;
ranging from S0 to .90 (mean a =.36; Baron &
7,arrera & Garrison-Jones, 1988; Shek, 1990; Teri, 1982). A
.74 has also been
test-retest reliability coefficient of
reported (Baron & LaPlante, 1984), Additionally, Barrera and
Garrison-Jones have demonstrated evidence of convergent
validity with the Child Assessment Schedule (CAS; Hodges,
Kline, Stern, Cytryn, & NcKnew, 1982) for items measuring
depression symptoms (r=.73) and with the General Self-worth
Children
subscale of the Perceived Competence Scale for
(Harter, 1982; L.-.64); they also reported evidence of
discriminant validity with CAS items measuring conduct disorder
(r..29) and anxiety (r..29) symptoms.
Data Analyses
Data were analyzed in two stages. First, the 3-factor
2nd-order structure proposed by Byrne and Baron (1990) was
tested separately for goodness-of-fit to male and female data.
Because it is important that the model for each sex be
well-fitting before testing for equivalencies of measurement
respecification was
and structure across the two groups, model
conducted if there was statistical and theoretical
1
justification for doing so. Once the baseline model was
determined for each sex, the equivalence of item measurements
females.
and factorial structure was tested across males and
BDI Gender Invariance
9
Cognizant of caveats related to tests for partial measurement
luthen, 1989), equality
procedures (see Byrne, Shavelson,
constraints were imposed on only those parameters similarly
specified for both groups.
Assessment of model fit was based on multiple criteria Ciat
practical
reflected statistical, theoretical, and
2
likelihood ratio, (h)
considerations; these included (a) the
X
& Lewis, 1973), (c) the
the Tucker-Lewis index (ILI; Tucker
Marsh, 1990), (d)
relative noncentrality index (RNI; McDonald &
(AGFI), t-values (parameter
the adjusted goodness-of-fit index
of estimate), and
estimates relative to their standard errors
2
the LISREL VI
modification indices (Ms), all provided by
Sorbom, 1985), and (e) the
computer program (Joreskog &
(see MacCallum, 1986;
substantive meaningfulness of the model
of
Suyapa, Silvia, & MacCallum, 1988). Criteria indicative
for the TLI and RNI, and
adequate model fit wete values >.90
>2.00 for the t-values.
Results
Determination of Baseline Models
For adolescent males, as indicated in Table 1,
the
model cf depression
hypothesized 3-factor hierarchical
represented a reasonably good fit to the data; all parameters
significant. (Model 0 argues that each item
were statistically
the null model required for
represents a factor, and provides
review of the MIs, however,
computation of tie ILI and RNI.) A
10