Acta Psychiatr Scand 2007: 116: 195–200 Copyright  2006 The Authors

All rights reserved Journal Compilation  2006 Blackwell Munksgaard

DOI: 10.1111/j.1600-0447.2006.00934.x ACTA PSYCHIATRICA
SCANDINAVICA

A short matrix-version of the Edinburgh

Depression Scale
Eberhard-Gran M, Eskild A, Samuelsen SO, Tambs K. A short matrix-
version of the Edinburgh Depression Scale.
Objective: Given the importance of depression as a world health
problem, depression assessment should be standard practice in large-
scale health surveys. The aim of the study was to construct a short
matrix-version of the Edinburgh Depression Scale (EDS) that can be
used in questionnaire studies.
Method: On the basis of the complete EDS scale of ten items, answered
by 2730 women, stepwise multiple regression analysis was used to find
the combination of items that explains the maximum proportion of the
variance of the full scale sum score. The selected EDS items were
thereafter correlated with the Hopkins Symptom Check List (SCL-25)
for external validation.
Results: The sum of five selected items from the EDS correlated at r ¼
0.96 with the full version. The EDS-5 scores correlated strongly with
the SCL-25 (r ¼ 0.75).
Conclusion: The EDS-5 version shows good psychometric properties
and may, for some scientific purposes, substitute the full EDS scale.
M. Eberhard-Gran1, A. Eskild1,2,
S. O. Samuelsen1,3, K. Tambs1
1
Division of Mental Health, Norwegian Institute of
Public Health, Oslo, 2Department of Gynaecology and
Obstetrics, Akershus University Hospital, Lorenskog and
3
Department of Mathematics, University of Oslo, Oslo,
Norway
Key words: Edinburgh Depression Scale; depression;
questionnaire
Malin Eberhard-Gran, Division of Mental Health, Nor-
wegian Institute of Public Health, Post Box 4404 Nyd-
alen, N-0403 Oslo, Norway.
E-mail: malin.eberhard-gran@fhi.no
Accepted for publication October 4, 2006

Significant Outcomes
• A five item version of the EDS shows good psychometric properties and may, for some scientific
purposes, replace the full version in questionnaire studies.
• The EDS-5 matrix version requires little space and can therefore be included in large-scale health
surveys, as depression assessment should be standard practice in such studies.
• The EDS-5 includes no items on somatic complaints and only one on anxiety. This is an advantage,
not only in studies of postnatal women, but also in studies of subjects with serious somatic diseases.

Limitations
• The study sample included women of reproductive age only. The validity of the EDS-5 needs to be
tested in other samples and compared with other measures of depression.
• In developing the matrix version of the EDS, we changed the lay-out and the order of response
alternatives in two of the items. To what extent this influences the answering trends is uncertain.
• The EDS-5 is meant primarily for use in research. In most clinical settings the original EDS version is
recommended.

fourth leading contributor to the global health

Introduction
Mental health is an important aspect of public health
that has long been segregated and neglected. Depres-
sion is, according to figures from the World Health
Organisation, the leading cause of disability and the
burden of disease (http://www.who.int/mental_
health/management/depression/definition/en/).
Given the importance of depression as a world
health problem, and emerging evidence of its
comorbidity with and negative effects on other

195
Eberhard-Gran et al.

health problems, depression assessment should be Measures of mental health

a standard practice in large-scale health surveys.
Two different mental health-screening instruments
Hence, there is a need to develop scales that
were included in the questionnaire, the EDS and
conserve space and time. Some established psy-
the Hopkins Symptom Check List 25-items version
chometric scales have been shortened without
(SCL-25) (16, 17).
severely affecting their reliability or validity (1, 2).
Edinburgh Depression Scale is a 10-item self-
There has been increasing research interest in
rating scale, with a maximum score of 30. The
mental health in the reproductive period (3–5). In
scoring of each item ranges from 0 to 3, with zero
studies of diseases that may have several risk
for absence of symptoms and three for maximum
factors such as mood disorders in the perinatal
severity and duration of symptoms (18). The EDS
period, it is particularly important to include both
is constructed from the ÔIrritability, Depression
biological and psychosocial variables (6, 7).
and Anxiety ScaleÕ (IDA) (19) and the ÔHospital
The Edinburgh Postnatal Depression Scale
Anxiety and Depression ScaleÕ (HADS) (20) in
(EPDS), a 10-item self-rating scale, was developed
addition to items formulated by Cox et al. (18), the
to detect postnatal depression (8). This scale was
constructors of the scale. Five items are about
developed because the established depression
dysphoric mood itself, two about anxiety, and one
screening instruments (9) were assumed to be
each about guilt, suicidal ideas and Ônot copingÕ
suboptimal when applied to postnatal women
(Table 1). The main feature of the EDS when
(10), as ÔnormalÕ postnatal symptoms, may be
compared with other depression scales is the
misconstrued as depressive symptomatology.
exclusion of items which may reflect physical
Although the EPDS was developed to screen for
discomfort, and could confuse depression with
depression in postnatal women, it is also useful in
somatic effects of childbirth. The EDS has been
the assessment of depression outside the postnatal
translated into Norwegian and validated against
period and the scale has also been validated in men
the DSM-IV criteria of major depression (15). A
(11, 12). When used in non-postnatal women or
score of ‡10 on the EDS scale identified all women
men the scale is referred to as the Edinburgh
with a major depression, with a sensitivity of 100%
Depression Scale (EDS) to reflect the broadening
(95% CI: 72–100%) and a specificity of 87% (95%
of its use (11, 13). The scale is henceforth referred
CI: 77–95%).
to as the EDS in our text.
Symptom Check List-25 is a self-rating scale
with 25 items on depression and anxiety. Each item
Aims of the study is scored from 1 (Ônot at allÕ) to 4 (ÔextremelyÕ), and
the sum score ranges from 25 to 100.
In order to reduce the request for space in a
questionnaire, we decided to construct a short
version of the EDS with a matrix format. Hence, Methods of analyses
the aim of this study was to select a few items from
For construction of a short EDS version that could
the EDS with the highest correlation to the full
be used in questionnaires, it was decided to choose
version on the basis of data from a population-
five items, half of the items in the full EDS version
based study.
(Table 1), and to design a matrix version with a
common set of response alternatives for all items
Material and methods (see Table 2). This could easily be done for all but
item six (ÔThings have been getting on top of meÕ,
Data source
Table 1, item 6) without notably changing the
In 1998–1999 we conducted a population-based meaning of the response alternatives. Also the
questionnaire study of mental health in the repro- original understanding of this item and its answer-
ductive period, including 2730 women, of whom ing alternative are not easily understood in the
416 were in the postpartum period. The sample is Norwegian language. This item was therefore
thoroughly described elsewhere (14). A total of excluded from the analyses.
2688 subjects had completed the EDS in the Stepwise linear regression analysis, with entering
questionnaire. A subsample of these women, all one item at a time, was used to identify the
postnatal women with EDS scores of ‡10 (n ¼ 26) combination of items that best explained the
and a control group with EDS scores of <10 maximum proportion of variance of the full scale
(n ¼ 31), were interviewed in a validation study of sum score (1, 2). The full scale EDS was used as the
the Norwegian version of the EDS (15). These dependent variable. At each step, the item
women were included in a subanalysis. that maximally increased the explained variance,

196
 Short-EDS

Table 1. The 10-item version of the Edinburgh Depression Scale (EDS) by Cox between the unweighted sum of the included items
et al. (8)
and the full EDS scale (rsum) and the correlations
In the past 7 days: between each item and the original score (ritem).
Often rsum is approximately equal to the square
1. I have been able to laugh and see the funny side of things
root of adjusted R2.
As much as I always could
Not quite so much now The data analyses were first performed in the
Definitely not so much now total population sample (n ¼ 2 688) and thereafter
Not at all in the postpartum women sample (n ¼ 416).
2. I have looked forward with enjoyment to things
As much as I ever did
Cronbach alpha was estimated as a measure of
Rather less than I used to internal consistency reliability of the short form
Definitely less than I used to scales.
Hardly at all The short EDS version should reflect the factor
3. I have blamed myself unnecessarily when things went wrong
Yes, most of the time
structure inherent in the EDS. Therefore a factor
Yes, some of the time analysis of the full scale EDS (principal compo-
Not very often nents analysis with oblimin rotation) was used to
No, never explore the distribution of items for each factor.
4. I have been anxious or worried for no good reason
No, not at all
Unlike an orthogonal rotation, this procedure
Hardly ever allows the rotated factors to be correlated. This
Yes, sometimes was a priori judged to be more realistic than forcing
Yes, very often the dimensions to be uncorrelated.
5. I have felt scared or panicky for no very good reason
Yes, quite a lot Also the correlation between the SCL-25 and the
Yes, sometimes EDS-5 and with the full scale EDS was estimated.
No, not much The association between the EDS-5 and clinical
No, not at all
depression, derived from the clinical interview
6. Things have been getting on top of me
Yes, most of the time I haven't been able to cope at all PRIME-MD (21), was studied in the subsample
Yes, sometimes I haven't been coping as well as usual of women (n ¼ 56) who had taken part in the EDS
No, most of the time I have coped quite well validation study.
No, I have been coping as well as ever
7. I have been so unhappy that I have had difficulty sleeping
Yes, most of the time Results
Yes, sometimes
Not very often When entering the EDS items in a stepwise
No, not at all
8. I have felt sad or miserable
regression analysis, the first item entered (item 8,
Yes, most of the time Table 1) explained 0.64 of the variance (adjusted
Yes, quite often R2) of the full scale EDS score. Item 4 was entered
Not very often in the next step, thereafter item 7, 3, 2, 5, 9, 1 and
No, not at all
9. I have been so unhappy that I have been crying
10. The proportions of explained variance in the
Yes, most of the time second and later steps were 0.78, 0.85, 0.89, 0.93,
Yes, quite often 0.95, 0.97, 0.98 and 0.98. Only the results from the
Only occasionally first five steps are tabulated (Table 3). The
No, never
10. The thought of harming myself has occurred to me explained variance of 0.93 after the inclusion of
Yes, quite often five items implies a correlation of 0.96 between the
Sometimes weighted sum (adjusted R) of these items and the
Hardly ever
full EDS. We conducted an additional analysis
Never
including women in the postpartum period only
(n ¼ 416), which resulted in exactly the same
compared with the explained variance from the selection of items. Table 3 shows values of
previous step, was automatically selected and explained variance (adjusted R2), correlations
entered in the analyses by the computer program. between unweighted sums of items and the original
Additional analyses running through all possible scores (rsum) in addition to the correlations between
combinations of five items were run to check that each item and the original score (ritem).
the stepwise linear regression in fact arrived at the The association between the EDS-5 scores and
optimal model for the short EDS version. The the EDS full scale scores showed a small trend of
adjusted R2 shows how well a weighted sum of the deviation from linearity, with somewhat decreasing
included items, with weights given by the unstand- slope at the extreme part of the distribution.
ardized regression coefficients, compares with the However, 92.3% of the variance for the EDS-5
full EDS scale. We also report the correlation could be explained by a linear association with the

197
Eberhard-Gran et al.

Table 2. A short matrix-version of the Edinburgh Depression Scale

In the past 7 days: Yes, most of the time Yes, some-times Not very often No, never

I have felt sad or miserable h h h h

I have been anxious or worried for no good reason h h h h
I have been so unhappy that I have had difficulty sleeping h h h h
I have blamed myself unnecessarily when things went wrong h h h h
I have looked forward with enjoyment to things h h h h

Table 3. Statistics for various short versions of the EDS (minus item number 6 in the original version)

Step Adjusted R2 Adjusted R rsum ritem a

1 I have felt sad or miserable* 0.64 0.80 0.80 0.80 –

2 I have been anxious or worried for no good reason 0.78 0.88 0.86 0.65 0.56
3 I have been so unhappy that I have had difficulty sleeping* 0.85 0.92 0.92 0.72 0.68
4 I have blamed myself unnecessarily when things went wrong 0.89 0.94 0.94 0.66 0.74
5 I have looked forward with enjoyment to things* 0.93 0.96 0.96 0.61 0.76

*Items loading at the dysphoric mood factor in the factor analysis of the original EDS
rsumis the correlation between the sum of the included items and the full EDS scale.

Table 4. Sensitivity and specificity of the EDS five item scale and the EDS full
scale for different cut-off scores. Based on an interview study among a subsample 15.00
of 56 women

EDS-5 EDS full scale

Sensitivity Specificity Sensitivity Specificity 10.00

Cut-off (%) (%) Cut-off (%) (%)
EDS-5

‡4.5 100 70 ‡9.5 100 87

‡5.5 89 74 ‡10.5 67 92
‡6.5 56 92 ‡11.5 56 94
‡7.5 44 94 ‡12.5 44 96 5.00
‡8.5 22 98 ‡13.5 22 98

full scale EDS, and only an additional 0.7% could 0.00

be explained by a non-linear association, so this
non-linear trend is negligible (Fig. 1).
The internal consistency reliability estimated as
Cronbach’s alpha increased from 0.56 in a two-
item version to 0.76 in a five-item version
(Table 3). The Cronbach’s alpha for the full scale
EDS was 0.81.
A factor analysis of the full scale EDS was
conducted. Specifying two factors was judged to
give the most comprehensive solution. The items in
the first group included six items focusing on
dysphoric mood and suicidal ideas (items number
1, 2 and 7–10). The second group included items
focusing on anxiety and guilt (items number 3–6)
(Table 1). The correlation was 0.46 between the
two groups. A factor analysis of the EDS-5
(Table 3) did reflect the factor structure of the
full scale. The items clustered into two groups.
Three items reflected dysphoric mood (item
number 2, 8 and 9) and two reflected anxiety and
guilt (items number 3 and 4) (Table 3).
0.00 10.00 20.00 30.00
EDS full scale
Fig. 1. The association between the 5-item EDS scores and the
full scale EDS scores in 2688 Norwegian women
The SCL-25 scores correlated strongly with the
EDS-5 scores (r ¼ 0.75) as well as with the EDS
full scale (r ¼ 0.80) in the questionnaire study
(n ¼ 2688).
Among the subsample of women who had
participated in the full scale EDS validation study
(n ¼ 56) a score of ‡5 on the five-item version
identified all women with major depression accord-
ing to DSM-IV, giving a sensitivity of 100% and a
specificity of 70% (Table 4). The sensitivity
decreased to 89%, and the specificity increased to
74% when a cut-off value of ‡6 was used (Table 4).

198

When a cut-off of ‡7 was used, the specificity was
increased to 92%.
Discussion
In this study we have constructed a five item matrix
version of the EDS. This EDS-5 version demon-
strates high correlation to the full scale EDS and to
the SCL-25, suggesting that a five item EDS matrix
version can replace the full version in questionnaire
studies. The EDS-5 version had a correlation
rsum ¼ 0.96 between the unweighted sum of the
items and the full scale EDS and thus explains
0.962 ¼ 0.92 of the variance. This corresponds well
to the adjusted R2 for the weighted sum of the
items equal to 0.93. The stepwise procedure thus
led to a highly predictive model also on the
unweighted scores. The exclusion of half of the
items in the EDS resulted in only minimal decrease
in Cronbach’s alpha. However, additional testing
of the internal consistency of the EDS-5 would be
useful. The sensitivity and specificity estimates for
the EDS-5 version are quit high. A loss of precision
is demonstrated when the cut-off levels are low.
However, there seems to be almost no deterior-
ation in prediction of severe cases.
In the full EDS scale a cut-off value of ‡13 is
considered to indicate a high level of depression
symptoms, whereas ‡10 indicates a moderate level,
which gives prevalences of 6.2% and 12.9%
respectively. The EDS-5 cut-off values, ‡8 for
high symptom level and ‡7 for moderate symptom
level, resulted in approximately comparable prev-
alences (7.4% and 10.4%). Due to few items, it was
not possible to arrive at short version cut-off values
that accurately corresponded to the conventional
cut-off values used with the original instrument.
For clinical use, the sensitivity and specificity of the
EDS-5 has to be estimated through further valid-
ation.
In developing the matrix version of the EDS we
changed the lay-out and the order of response
alternatives in two of the items (number 2 and 4).
Also some of the response sets were changed. Most
importantly item 1 and 2 may have been slightly
moved towards trait-like responses as opposed to
state-like. For example, item 2 in the original EDS
ÔI have looked forward with enjoyment to thingsÕ
ranges in answer alternatives from Ôas much as I
ever didÕ to Ôhardly at allÕ. These answer alterna-
tives highlight the state-nature of that item as
opposed to the short matrix version with answer
alternatives ranging from ÔYes, most of the timeÕ to
ÔNo, neverÕ. Persons with a depressive personality
trait may tend to score higher on item number two
in the EDS-5 version than in the original version.
Short-EDS
To what extent possible individual differences in
answering trends between the full and the EDS-5
version will influence the individual sum score or
study sample mean score cannot be estimated in
our study.
Suicidal ideation is often considered an especi-
ally important symptom in depression. Item 10,
about self-harm, was not selected in the statistical
analyses to be among the items best explaining the
maximum proportion of variance of the full scale.
If screening for suicidal ideation is needed, the full
EDS scale is recommended. The rationale for
construction of a matrix version of the EDS was
for use in research, primarily in questionnaire
studies. In clinical settings it is probably preferable
to use the original EDS version, which produces
slightly more precise results.
The main reason for proposing a short version of
the EDS was to save space in questionnaires in
which the competition of space often is strong. In
this study we have shown that it is possible to
exclude half of the items with only a small loss in
psychometric quality. The five item matrix version
of the EDS is a rather crude indicator of depres-
sion, but for some research purposes it appears to
be sufficiently precise and may thus encourage
depression assessment in large-scale health surveys.
The EDS-5 includes no items on somatic com-
plaints and only one item on anxiety. This is an
advantage not only in assessing depression in
postnatal women, but also in subjects with serious
somatic diseases. However, the validity of the
EDS-5 needs to be tested in other samples and
compared with other measures of depression. A
very brief depression measure that has adequate
reliability and validity for use in general health
questionnaires has been called for. Our study
suggests that EDS-5 may have these qualities and
may therefore be a useful tool in large-scale health
surveys in the future.
Acknowledgements
This research was supported by the Norwegian Research
Council.
Reference
1. Bukstein D, McGrath M, Buchner D, Landgraf J, Goss T.
Evaluation of a short form for measuring health-related
quality of life among pediatric asthma patients. J Allergy
Clin Immunol 2000;105:245–251.
2. Tambs K, Moum T. How well can a few questionnaire items
indicate anxiety and depression? Acta Psychiatr Scand
1993;87:364–367.
3. Austin M-P, Priest SR. Clinical issues in perinatal mental
health: new developments in the detection and treatment of
199
Eberhard-Gran et al.
perinatal mood and anxiety disorders. Acta Psychiatr
Scand 2005;112:97–104.
4. Evans J, Heron J, Francomb H, Oke S, Golding J. Cohort
study of depressed mood during pregnancy and after
childbirth. BMJ 2001;323:257–260.
5. Patel RR, Murphy DJ, Peters TJ. Operative delivery and
postnatal depression: a cohort study. BMJ 2005;
doi:10.1136.
6. Dennis C-LE, Janssen PA, Singer L. Identifying women
at-risk for postpartum depression in the immediate post-
partum period. Acta Psychiatr Scand 2004;110:338–346.
7. Ross LE, Sellers EM, Gilbert Evans SE, Romach MK. Mood
changes during pregnancy and the postpartum period:
development of a biopsychosocial model. Acta Psychiatr
Scand 2003;109:457–466.
8. Cox JL, Holden JM, Sagovsky R. Detection of postnatal
depression. Development of the 10-item Edinburgh Post-
natal Depression Scale. Br J Psychiatry 1987;150: 782–786.
9. Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J. An
inventory for measuring depression. Arch Gen Psychiatry
1961;4:561–571.
10. Harris B, Huckle P, Thomas R, Johns S, Fung H. The use of
rating scales to identify post-natal depression. Br J Psy-
chiatry 1989;154:813–817.
11. Cox J, Chapman G, Murray D, Jones P. Validation of the
Edinburgh postnatal depresion scale (EPDS) in non-post-
natal women. J Affect Disord 1996;39:185–189.
12. Alder E, Cox J. Breast feeding and postnatal depression. J
Psychosom Res 1983;27:139–144.
13. Austin M-P, Hadzi-Pavlovic D, Saint K, Parker G. Ante-
natal screening for the prediction of postnatal depression:
200
validation of a psychosocial pregnancy risk questionnaire.
Acta Psychiatr Scand 2005;112:310–317.
14. Eberhard-Gran M, Eskild A, Tambs K, Samuelsen S, Opj-
ordsmoen S. Depression in postpartum and non-postpar-
tum women: prevalence and risk factors. Acta Psychiatr
Scand 2002;105:426–433.
15. Eberhard-Gran M, Eskild A, Tambs K, Schei B, Opjordsmoen
S. The Edinburgh Postnatal Depression Scale: validation
in a Norwegian community sample. Nord J Psychiatry
2001;55:113–117.
16. Hesbacher P, Rickels R, Morris R, Newman H, Rosenfeld M.
Psychiatric illness in family practice. J Clin Psychiatry
1980;41:6–10.
17. Winikur A, Winikur D, Rickels K, Cox D. Symptoms of
emotional distress in a family planning service: stability
over a four-week period. Br J Psychiatry 1984;144:395–
399.
18. Cox J. Origin and development of the 10-item Edinburgh
Postnatal Depression Scale. In: Cox J, Holden J, eds.
Perinatal psychiatry: use and misuse of the Edinburgh
Postnatal Depression Scale. Glasgow: Gaskell, 1996:115–
124.
19. Snaith RP, Constantopoulos AA, Jardine MY, McGuffin P.
A clinical scale for the self-assessment of irritability. Br J
Psychiatry 1978;132:164–171.
20. Zigmond AS, Snaith RP. The hospital anxiety and Depres-
sion Scale. Acta Psychiatr Scand 1983;67:361–370.
21. Spitzer RL, Williams JB, Kroenke K, Linzer M, deGruy FV,
Hahn SR et al. Utility of a new procedure for diagnosing
mental disorders in primary care. The PRIME-MD 1000
study. JAMA 1994;272:1749–1756.