identifying vulnerability in grief: psychometric properties of the adult attitude to grief scale
TRANSCRIPT
Identifying vulnerability in grief: psychometric propertiesof the Adult Attitude to Grief Scale
Julius Sim • Linda Machin • Bernadette Bartlam
Accepted: 1 October 2013 / Published online: 16 October 2013
� Springer Science+Business Media Dordrecht 2013
Abstract Purpose Grief is a reaction to a significant loss
that can profoundly affect all aspects of life and capacity to
function well. The consequences can vary from severe psy-
chological distress through to physical disturbances and sig-
nificant social problems. This study sought to identify a
measure of vulnerability in grief, by examining the psycho-
metric properties of the Adult Attitude to Grief (AAG) scale in
a sample of 168 people seeking help in their bereavement.
Methods The factor structure of the scale, its internal
consistency, its construct validity and optimum classifica-
tion cutoffs were tested.
Results Confirmatory factor analysis broadly supported the
factor structure of the AAG, but identified one item that
could profitably be reworded. Internal consistency of the
three subscales was acceptable. Construct validity and
discriminative validity were supported by correlations with
allied constructs (depression and anxiety) and a significant
difference between scores for clients with Prolonged Grief
Disorder and those without. A correlation with counsellors’
own clinical ratings of vulnerability demonstrated crite-
rion-related validity of the AAG. Using receiver operating
characteristic methods, optimum cutoff scores on the scale
were identified for the classification of different levels of
vulnerability.
Conclusion The AAG was found to be a psychometrically
promising tool for identifying vulnerability in grief.
Keywords Adult Attitude to Grief scale �Vulnerability � Psychometrics � Validity � Reliability �Factor analysis
List of symbols
AAG Adult Attitude to Grief scale
CFI Comparative fit index
GAD-7 Generalized Anxiety Disorder Assessment 7
PG-13 Prolonged Grief Disorder Scale
PGD Prolonged grief disorder
PHQ-9 Patient Health Questionnaire 9
RMSEA Root means square error of approximation
ROC Receiver operating characteristic
TLI Tucker–Lewis Index
WLSMV Weighed least squares mean and variance
adjusted
Introduction
There were 484,367 deaths registered for England and
Wales in 2011 [1]. While some people demonstrate resil-
ience in response to traumatic or disturbing life events,
quality of life can be seriously impaired for many bereaved
people, and it is estimated that 10–15 % of the general
bereaved population are susceptible to complicated grief [2].
Certain individuals are therefore vulnerable in their grief,
and identifying this group is a practice and research
imperative, both because intervening inappropriately can
cause harm [3, 4] and because of the need to use increas-
ingly scarce health and social care resources effectively [5].
Identifying the components of complicated grief remains
a challenge for both researchers and practitioners. Foremost
in this field of research have been Prigerson and Macie-
jewski [6], who have produced a definition of prolonged
grief disorder (PGD) that is now proposed as an officially
recognized psychiatric disorder [7]. It is worth noting that,
while other psychiatric conditions such as depression and/or
anxiety have been regarded as components of complicated
J. Sim (&) � L. Machin � B. Bartlam
Research Institute for Social Sciences, Keele University,
Staffordshire ST5 5BG, UK
e-mail: [email protected]
123
Qual Life Res (2014) 23:1211–1220
DOI 10.1007/s11136-013-0551-1
grief, they do not provide a complete definition of the
complex spectrum of grief responses that may make a per-
son vulnerable in their bereavement.
The Range of Response to Loss (RRL) model has been
developed to explain individuals’ responses to grief [8].
The characteristics of grief articulated within the RRL
model became evident in a study of 97 bereaved people [9],
and the pattern was confirmed in clinical practice. Initially,
three categories of response were identified: overwhelmed,
controlled and balanced [8, 10]. Reflecting the language
used by clients, the RRL model was found to resonate with
practitioners and provides clear conceptual links into their
work, while at the same time having parallels with other
key theories (Table 1).
The Adult Attitude to Grief scale (AAG) scale was
devised to reflect the categories in the RRL. In the scale,
three items reflect qualities associated with being over-
whelmed—stressfulness, irreversibility and loss of mean-
ing in life [15]; three items represent control—restricted
acknowledgement of distress, exaggerated need for self-
reliance and avoidance of grief through overly engaging in
day-to-day life [15]; and three items reflect balance (later
defined as resilience)—courage, resourcefulness and opti-
mism [16, 17]. It should be noted that while the concept of
‘control’ might usually be seen as an appropriately adap-
tive response to loss in the context of clinical practice, here
it is conceptualized as a potentially problematic feature in
which loss disrupts the usual controlling coping mecha-
nisms. The nine items are each scored on a five-point
Likert scale from ‘strongly agree’ to ‘strongly disagree’.
On the basis of a previous exploratory factor analysis [8],
these items provided support for the three dimensions:
overwhelmed (items 2, 5, and 7), controlled (items 4, 6 and
8) and resilient (items 1, 3 and 9). The research also sug-
gested that the scale could be used to provide an individual
profile of grief based on the relative presence or absence of
the overwhelmed, controlled and balanced/resilient
components reported by respondents [8]. Subsequent
studies [18, 19] have supported the practical usefulness of
the AAG scale to achieve a picture of individual grief and
its changes overtime. The findings from those studies also
indicated the value of practitioners using the self-report
statements as prompts in exploring clients’ experiences of
loss in greater depth and in determining the direction of
therapeutic intervention.
Further theoretical debate about the interactive dynamic
between stressors and coping as the basis for understanding
individual adjustment to bereavement [20] and practice-
based reflection have helped to clarify the functioning of
the elements within the RRL model. As a result of this
process, it was clear that the model is describing two dis-
tinct dimensions in response to loss. The first of these
consists of the core grief impact (stressor), producing
variable overwhelmed and controlled reactions. These
states are not by definition evidence of vulnerability but are
dependent on the second dimension, that of coping, in
which the capacity to respond to grief with equilibrium is
characteristic of resilience, while a limited capacity to
manage grief is characteristic of vulnerability. These
reflections have resulted in vulnerability being included as
a fourth component in the RRL model.
Using this theoretical rationale, it was proposed that the
fourth component of the RRL may be calculated using the
AAG scale. Specifically, it was hypothesized that vulner-
ability is a higher-order component directly related to
scores on the overwhelmed and controlled subscales and
inversely related to the score on the resilient subscale. The
rationale for this proposition is that the overwhelmed and
the controlled items on the scale represent a spectrum of
core grief reactions, while the resilient items indicate a
capacity to cope with the consequences of grief. By off-
setting the stressful elements of grief against the positive
mechanisms of coping with grief, an emergent picture of
relative vulnerability can potentially be drawn from the
AAG scale. Hence, an indication of vulnerability can be
derived by summating those items on the scale that are
indicative of vulnerability, i.e. the overwhelmed and con-
trolled items, and discounting the mediating resilient items.
Scores are recorded on a vulnerability rating grid on which
the resilient scores are reversed to allow for simple addition
of all the nine items in the AAG scale (Table 2). The
resulting range of possible scores is from 0 to 36, with
higher scores indicating greater vulnerability.
The aim of this study was to evaluate the AAG scale as a
measure to identify vulnerability, by examining its psy-
chometric properties and determining its potential utility as
a tool to calculate the degree of vulnerability in people
presenting for support in their loss, as an indicative guide
to intervention. Specifically, the research sought to test four
key psychometric properties of the AAG scale:
Table 1 Conceptual comparisons between the range of response to
loss model and other related theories
Other
theories
Range of response to loss model
Overwhelmed Resilient Controlled
Attachment
theory [11]
Anxious/
ambivalent
Secure Avoidant
Stress theory
[12]
Intrusion Avoidance
Dual process
model [13]
Loss
orientation
Oscillation Restoration;
orientation
Personality-
related
theory [14]
Intuitive
grief:
emotional
coping
Blended grief:
emotional and
cognitive coping
Instrumental
grief:
cognitive
coping
1212 Qual Life Res (2014) 23:1211–1220
123
• its factor structure
• its internal consistency
• its construct and criterion-related validity and
• optimum cutoffs for classification.
Methods
Sampling
Following ethical approval, participants were recruited
from three hospices and one community-based bereave-
ment service. Those clients accessing the collaborating
services—in which the AAG scale was routinely used as a
standard method of exploring their grief—were included in
the study unless they chose to opt out. Sample size was
determined in relation to the proposed CFA analysis. It is
recommended that there should be a minimum of between
5 and 10 cases per estimated parameter [21]. With 20
parameters to be estimated, a sample size of at least 160
was sought, so as to provide an intermediate figure of 8
cases per parameter.
Analysis
To examine the dimensionality of the scale, a factor
structure previously identified through exploratory factor
analysis [8], with the addition of the vulnerability indicator
(Fig. 1), was tested on the data by means of confirmatory
factor analysis (CFA), using Mplus 7 (www.statmodel.
com). A WLSMV estimator was employed, and variables
were designated as ordered categorical. The fit of the
hypothesized factor structure to the data was quantified
using a range of fit indices: the Tucker–Lewis index (TLI),
the comparative fit index (CFI) and the root mean square
error of approximation (RMSEA) [22]. Hu and Bentler [23]
recommend using a combination of such fit indices, as no
single index has been found to be optimal. The three
indices have in common that they are not influenced by
sample size and—in the case of the TLI and the RMSEA—
reward parsimony in the hypothesized factor structure [24].
Each fit index normally produces a value between 0 and 1
(though the RMSEA can produce values above, and the
TLI values both above and below, this range). In the case
of the TLI and the CFI, values close to 1 indicate good fit,
whereas for the RMSEA values close to 0 indicate good fit.
Benchmarks for the fit indices have been proposed by a
number of authors: values on the TLI and the CFI should
ideally exceed .95, and those on the RMSEA should ideally
lie below .05 [25, 26]. Marsh et al. [27] warn, however,
against a mechanistic interpretation of these fit indices. The
v2 statistic for each model was also calculated; while the
associated p value should be non-significant for a well-
fitting model, its magnitude is sensitive to sample size and
is not therefore a reliable indicator of fit [28]. Interpretation
of the statistic is also dependent on the degrees of freedom
of the model, and the normed v2 was therefore also cal-
culated (v2/degrees of freedom); lower values of this sta-
tistic indicate better fit. The v2 statistic also served as a
means of comparing the difference in fit of nested models.
Modification indices and item characteristic curves [29]
were generated to help investigate items in the scale that
appeared to be responsible for any lack of fit.
The dimensionality of the AAG was tested in two
models. In the first model, the nine items were posited to
measure just the three dimensions represented by the sub-
scales: overwhelmed (items 2, 5, and 7), controlled (items
4, 6 and 8) and resilient (items 1, 3 and 9). The second
model stated additionally that these three dimensions
Table 2 The Adult Attitude to Grief Scale shown in its domains and with a scoring system to obtain an indication of vulnerability
Domain AAG items Response options
Strongly
agree
Agree Neither agree
nor disagree
Disagree Strongly
disagree
Overwhelmed 2. For me, it is difficult to switch off thoughts about the person
I have lost
4 3 2 1 0
5. I feel that I will always carry the pain of grief with me 4 3 2 1 0
7. Life has less meaning for me after this loss 4 3 2 1 0
Controlled 4. I believe that I must be brave in the face of loss 4 3 2 1 0
6. For me, it is important to keep my grief under control 4 3 2 1 0
8. I think its best just to get on with life after a loss 4 3 2 1 0
Resilient 1. I feel able to face the pain which comes with loss 0 1 2 3 4
3. I feel very aware of my inner strength when faced with grief 0 1 2 3 4
9. It may not always feel like it but I do believe that I will
come through this experience of grief
0 1 2 3 4
Numbering of the items indicates the order in which they appear in the scale
Qual Life Res (2014) 23:1211–1220 1213
123
together constituted a vulnerability indicator, in the way
described in the description of the AAG scale above and
illustrated in Fig. 1. In addition to an evaluation at the level
of the scale as a whole, dimensionality was examined at the
level of individual items, by calculating polychoric corre-
lations between each item and the three domain scores
(with the item in question excluded from the calculation of
its own domain score). Kline [30] and Nunnally and
Bernstein [31] propose that the correlation of each item
with its own domain should exceed .3. Additionally, one
would expect an item to correlate more strongly with its
own domain than with other domains.
Internal consistency was quantified by calculating an
index of composite reliability for each subscale [32]. This
method generates coefficients that are analogous to Cron-
bach’s alpha and are interpreted in a similar way. A value
of .7 or greater is generally regarded as acceptable [33–35].
The construct validity of the AAG scale was assessed in
two ways. First, Pearson correlations were calculated with
two measures to which, on theoretical grounds, the AAG
should be related: a measure of anxiety, the Generalized
Anxiety Disorder Assessment 7 (GAD-7) [36], and a
measure of depression, the Patient Health Questionnaire 9
(PHQ-9) [37]. These measures have been shown to be
psychometrically sound [38, 39]. Clients with high vul-
nerability scores might be expected to score high on anx-
iety and/or depression; therefore, a correlation between the
AAG scale and each of these measures would support the
construct validity of the AAG scale. Second, the discrim-
inative validity of the AAG was tested using a ‘known
groups’ approach [40] in relation to a classification of PGD
based on the 13-item Prolonged Grief Scale (PG-13) [7]. A
classification of PGD is based on a criterion of more than
6 months since the bereavement and the fulfilment of ten
other criteria; this classification has shown good diagnostic
validity [7]. On theoretical grounds, we would expect cli-
ents with PGD to be more vulnerable than clients without
PGD, and thereby, a significant difference to be found
between the AAG scores for these two groups. This ana-
lysis was restricted to those clients (n = 76) who were
potentially eligible for classification as having prolonged
grief (more than 6 months since bereavement).
Criterion-related validity of the AAG scale was tested in
relation to its correlation (Spearman’s rho) with counsel-
lors’ clinical assessments of vulnerability, which were
available for 162 clients. Counsellors were asked to place
Fig. 1 Hypothesized factor
structure of the Adult Attitude
to Grief Scale (see Table 2 for
item descriptions). The ellipses
denoted by e represent residual
error terms
1214 Qual Life Res (2014) 23:1211–1220
123
the client in one of four categories in response to the
question ‘What level of vulnerability best describes your
client?’ The classifications produced were as follows: low
vulnerability (n = 19, 12 %); moderate vulnerability
(n = 59, 36 %); high vulnerability (n = 76, 47 %); and
severe vulnerability (n = 8, 5 %). These assessments were
made by counsellors without knowledge of a client’s AAG
score.
The counsellors’ clinical assessments of vulnerability
were also used to identify threshold scores on the AAG
scale, so as to establish whether the scale can be used to
generate a classification of vulnerability. Using receiver
operating characteristic (ROC) analysis, a C statistic was
calculated for three dichotomous classifications based on
the boundaries of adjacent categories of vulnerability, i.e.
those between (1) severe and high, (2) high and moderate,
and (3) moderate and low. This statistic represents the area
under the ROC curve and can range from 0 to 1 [40]. A C
statistic of 1 indicates that the scale has the greatest pos-
sible sensitivity and specificity with regard to the dichot-
omous classification concerned (all categorizations of
individuals with respect to this classification are perfectly
accurate), and one of 0 denotes the worst possible sensi-
tivity and specificity (all such categorizations are totally
inaccurate); an intermediate value of .5 indicates that the
scale categorizes no more accurately than guessing. It
follows that the closer the C statistic is to 1, the more
accurately the scale categorizes individuals.
Additionally, specific threshold scores were identified
on the AAG that separate adjacent classifications in the
counsellors’ assessments of vulnerability with the greatest
sensitivity and specificity, using the Youden index (J). This
statistic expresses the maximum difference between sen-
sitivity and 1–specificity (i.e. between the true positive rate
and the false positive rate) for a given cutoff score, on a 0–
1 scale [41]:
J ¼ sensitivity þ specificity�1ð Þ:
For example, a cutoff with a sensitivity .61 of and a
specificity of .72 would yield a Youden index of .61 ? (.72
- 1) = .33. A cutoff score with the largest Youden index
is therefore the one for which sensitivity and specificity for
a particular classification are simultaneously maximized.
Findings
The AAG scale was completed in full by 168 clients: 40
male (23.8 %) and 128 female (76.2 %). Four other clients
had missing data on one or more items in the scale and
were excluded from analysis. The median age category was
46–55 (age was recorded in intervals rather than continu-
ously for reasons of anonymity).
Mean (standard deviation [SD]) scores on the subscales
of the AAG were as follows: overwhelmed 8.92 (2.53);
controlled 7.98 (2.31); and resilient 5.25 (2.46). The mean
(SD) for the total score was 22.15 (4.38). Figure 2 shows
the distribution of values for the domain scores, and Fig. 3
shows the distribution of the total score. The negative skew
of scores on the overwhelmed domain suggests that there
may be a ceiling effect for this subscale. The other domain
scores and the total score exhibit a reasonably symmetrical
distribution.
Factor structure
The results of the CFA are shown in Table 3. The fit
indices do not show a very good fit. From the X2 test for
difference, the model with just the domain scores seems to
fit slightly better than the domain and total score model—
the difference in fit is statistically significantly
(X2 = 14.09. df = 1, p \ .001).
Table 4 shows polychoric correlations of each item with
its own domain and with other domains. Across all items,
with the exception of item 8, correlations with the item’s
own domain are C.442 (exceeding the benchmark value of
.3 [30, 31]), and correlations with other domains are B.280.
Item 8 had a stronger absolute correlation with the resilient
domain (-.357) than with its own controlled domain
(.264). Thus, the pattern of correlations, together with
examination of the modification indices from the CFA
output, suggested that item 8 does not just load on the
controlled domain, as hypothesized in the original model.
To explore this further, item 8 was temporarily re-assigned
to the resilient domain, producing a slight improvement in
fit (Table 5). However, when item 8 was allowed to load on
both the controlled and the resilient domains, fit was
noticeably improved over the original model, and the val-
ues for TLI, CFI and RMSEA were close to the desirable
cutoffs; the improvement in fit over the original model was
significant (X2 = 41.62. df = 1, p \ .001). The perfor-
mance of the item was investigated in more detail through
an item characteristics curve; this plots the probability of
endorsement of an item (in this case, the endorsement of
the highest category on the scale) against the underlying
latent construct (h) on a scale standardized to SD units
(Fig. 4). Such a curve should show a steep incline at a point
corresponding to .5 probability (indicating good discrimi-
nation for the item) and should indicate a probability of
endorsement close to 1.0 at the upper extreme of h (indi-
cating an appropriate threshold on h at which respondents
endorse the highest point on the scale; analogous to
appropriate ‘difficulty’ for an item testing knowledge) [29].
Figure 4 shows that when included in the controlled
domain (solid curve), item 8 has poor discrimination, and
even when h is high (?4 SDs), the probability of
Qual Life Res (2014) 23:1211–1220 1215
123
endorsement of the highest category on the item is only just
over .5. The item’s performance when included in the
resilient domain (dashed curve) is markedly better in these
respects, though not ideal.
Internal consistency
The coefficients of composite reliability, with 95 % con-
fidence intervals (CIs), for the three subscales (as per the
Fig. 2 Distribution of the three subscale scores on the Adult Attitude
to Grief Scale. The dashed vertical line indicates the mean score
Fig. 3 Distribution of total scores on the Adult Attitude to Grief
Scale. The dashed vertical line indicates the mean score
Table 3 Results of the confirmatory factor analysis for the original
hypothesized model
Statistic Domain scores and
total score
Domain scores only
TLI .652 .714
CFI .758 .809
RMSEA .153 .139
X2 123.53 (df = 25) p \ .001 101.84 (df = 24) p \ .001
Normed X2 4.941 4.243
TLI Tucker–Lewis index, CFI comparative fit index, RMSEA root
mean square error of approximation
Table 4 Item polychoric correlations with domain scores
Item Domain
Overwhelmed Controlled Resilient
2 .465 .012 .103
5 .526 .252 .241
7 .475 .017 .256
4 .159 .446 -.167
6 .283 .514 -.045
8 -.105 .264 -.357
1 .149 -.127 .482
3 .184 -.269 .479
9 .231 -.172 .442
An item’s correlation with its own domain is shown in bold
1216 Qual Life Res (2014) 23:1211–1220
123
original model) were as follows: overwhelmed .730 (95 %
CI .689, .771); controlled .683 (95 % CI .638, .728); and
resilient .692 (95 % CI .649, .736). These values either
closely approached or exceeded the benchmark of .7 for
acceptable internal consistency.
Construct and criterion-related validity
As indications of construct validity, the AAG scale had a
correlation of .470 (95 % CI .338, .584; p \ .001;
n = 156) with the PHQ-9 and .442 (95 % CI .308, .559;
p \ .001; n = 160) with the GAD-7. The mean (SD) score
on the AAG for clients who were negative for prolonged
grief (n = 47) on the PG-13 was 21.13 (4.03), and that for
clients who were prolonged grief positive (n = 29) was
24.38 (3.76); the latter group therefore had a score 3.25
scale points higher (95 % CI 1.40, 5.10; p = .001), rep-
resenting 9.03 % of the scale and a standardized difference
of .83 [42]. As regards criterion-related validity, the cor-
relation of the AAG scale with the counsellors’ assess-
ments of vulnerability was .521 (95 % CI .399, .625;
p \ .001; n = 162).
Classification of vulnerability
The ROC analysis generated C statistics of .837 (95 % CI
.743, .932) for a classification of severe vulnerability, .760
(95 % CI .688, .833) for a classification of high or severe
vulnerability and .846 (95 % CI .755, .937) for a classifi-
cation of low vulnerability.
Table 6 shows the results of the ROC analysis in rela-
tion to classification thresholds (cutoff scores). The opti-
mum cutoff between adjacent classifications of
vulnerability is the point on the AAG scale at which the
Youden index is largest. If sensitivity and specificity are
regarded as equally important, the optimum thresholds for
classifying a client are 24 for severe vulnerability, 23 for
high or severe vulnerability and 21 for vulnerability that is
moderate or greater. However, a false negative may be
more serious than a false positive in the diagnosis of vul-
nerability; it is a potentially worse error to withhold pro-
fessional intervention from a client who needs it than to
provide such help to a client who does not need it.
Accordingly, Table 6 also shows the optimum cutoffs
when sensitivity is required to be at least .85; these are
lower for classifications of high or severe vulnerability (21)
and of moderate or greater vulnerability (20).
Table 6 also shows that the Youden index is lower when
classifying clients at the ‘midpoint’ of vulnerability (severe
or high vs. moderate or low), indicating that the scale
produces less clinically reliable classifications at this point;
this reflects the lower C statistic for this classification
threshold.
Discussion
This study supports the view that the AAG provides a
measure of vulnerability capable of wide practice appli-
cation that is distinct from existing measures such as the
PG-13 [7]. The PG-13 asks questions about symptoms of
grief relating to feelings, thoughts and actions, which if
persisting for 6 months or more are seen to be associated
with significant functional impairment. In contrast, the
AAG aims to provide a broader grief profile, in which both
core grief reactions and the coping response to them are
seen to function interactively, providing a way of estab-
lishing to what degree the symptoms of grief are being
Table 5 Results of the confirmatory factor analysis for the domain
and total score model with adjustments to the loading of item 8
Statistic Item 8 loading on
resilient domain
Item 8 loading on both controlled
and resilient domains
TLI .778 .873
CFI .846 .915
RMSEA .122 .093
X2 87.75 (df = 25)
p \ .001
58.55 (df = 24) p \ .001
Normed
X23.51 2.44
TLI Tucker–Lewis index, CFI comparative fit index, RMSEA root
mean square error of approximation
Fig. 4 Item characteristic curve for item 8 in relation to the
controlled domain (solid curve) and the resilient domain (dashed
curve). The horizontal axis represents the latent variable (h) for these
domains. The vertical axis represents probability of endorsement of
the highest category on the five-point scale for the item. Item 8 is
hypothesized to have an inverse relationship with the resilient
domain, but for the purpose of clarity has been reverse-scored so that
the curve lies in the same direction as that for the controlled domain
Qual Life Res (2014) 23:1211–1220 1217
123
managed effectively. The AAG does not seek to identify a
time point at which vulnerability becomes more problem-
atic. Rather, the use of the AAG scale in practice makes
more variable assumptions about individual vulnerability
dependent on wider circumstantial factors, e.g. the death of
a child is known to generate a more extended period of
vulnerability, which is accepted as normal under such
circumstances.
The CFA results suggest that the hypothesized factor
structure of the AAG, with the vulnerability indicator
added, does not provide an optimum fit to the sample data
in this study. Although the pattern of correlations for item
8—positive with the controlled domain and negative with
the resilient domain—is in keeping with the calculation of
the vulnerability score, where the controlled and resilient
domain are hypothesized to be inversely related, its item
characteristic curve suggests that it does not perform well.
It is likely, therefore, that this item would benefit from a re-
evaluation of its wording, so that it loads highly on just one
domain and displays greater discrimination. Other items’
correlations with their own and with other domains support
the dimensionality of the scale.
The internal consistency of the three subscales is
acceptable and is highest for the overwhelmed subscale.
The overwhelmed subscale appears to be subject to a
ceiling effect, such that it may be hard to differentiate
between clients who lie at the upper extreme of this
domain, and some modification of one or more of the items
in this domain may be required—for example, making the
descriptors somewhat stronger so that they would receive
lower average levels of endorsement, thereby shifting the
distribution of the subscale scores downwards.
As regards construct validity, the AGG exhibited sta-
tistically significant ‘medium’ correlations (i.e. C.3; [42])
with measures of depression (PHQ-9) and anxiety (GAD-
7), which reflect the hypothesized theoretical relationships
between vulnerability and these other constructs. The dis-
criminative validity of the scale was also supported by the
findings of a statistically significant difference of ‘large’
magnitude (i.e. C.8 [42]) between the scores of those
clients with prolonged grief disorder and those without.
The ‘large’ correlation (i.e. C.5 [42]) between the AAG
scale and counsellor’s own assessments of vulnerability
provides evidence for the criterion-related validity of the
scale.
Overall, as judged by the C statistic, the AAG scale
performed well when classifying clients’ vulnerability at
the extremes (severe vulnerability vs. lower levels of vul-
nerability or low vulnerability vs. higher levels of vulner-
ability). Optimum cutoff scores on the scale were also
identified for the classification of different levels of vul-
nerability. If the sensitivity of the classification is required
to be at least .85, so as to reduce the risk of false negatives,
a score of 24 or greater best differentiates clients with
severe vulnerability from those with lower levels of vul-
nerability, a score of 21 or greater best differentiates those
with at least high vulnerability from those with lower levels
and a score of 20 or lower best differentiates those with
moderate or greater levels of vulnerability from those with
low levels. The cutoff separating those with at least high
vulnerability from those with lower levels appears to be
clinically less reliable than the other two cutoffs.
Conclusion
Overall, the AAG has sound psychometric properties,
though modification of one item in particular would be
fruitful; item 8 should ideally load on just one domain, and
more highly than it does currently, and should show greater
discrimination. The imperative for providers of bereave-
ment services is to target care appropriately to those most
in need, based on the complexity and persistence of grief
symptoms. This demands a measure that can give a con-
fident indication of vulnerability. The AAG is already used
to profile and explore the dynamics of individual grief and
as a guide for therapy, and its extension to provide an
indication of vulnerability would be an important devel-
opment for individual practitioners and for service provi-
sion more broadly. Further development of the
Table 6 Evaluation of optimum cutoff scores on the AAG scale; these are provided for the situation where sensitivity and specificity are equally
weighted and for the situation where sensitivity is prioritized (C.850)
Sensitivity and specificity equally weighted Sensitivity required to be at least .850
Optimum
cutoff
Youden
index
Sensitivity Specificity Optimum
cutoff
Youden
index
Sensitivity Specificity
Severe (n = 8) versus high/moderate/
low (n = 154)
24 .630 1.000 .630 24 .630 1.000 .630
Severe/high (n = 84) versus
moderate/low (n = 78)
23 .373 .655 .718 21 .240 .881 .487
Severe/high/moderate (n = 143)
versus low (n = 19)
21 .618 .776 .842 20 .544 .860 .684
1218 Qual Life Res (2014) 23:1211–1220
123
psychometric properties of the scale will strengthen its role
in this respect.
Acknowledgments This study was funded by the North Stafford-
shire Medical Institute. The authors wish to thank the Dove Service
(North Staffordshire, UK), St Giles Hospice (Lichfield, UK) and the
Marie Curie Hospices (Belfast and Hampstead, UK) for their help
with data collection, Aisling Bartlam for help with data entry, and
Kelvin Jordan for advice on the manuscript.
References
1. Office for National Statistics (2012). Births and deaths in Eng-
land and Wales 2011. London: Office for National Statistics
www.statistics.gov.uk/statbase/Product.asp?vlnk=14409.
2. Prigerson, H. G., & Jacobs, S. C. (2001). Traumatic grief as a
distinct disorder: A rationale, consensus criteria, and a pre-
liminary empirical test. In M. S. Stroebe, R. O. Hansson, W.
Stroebe, & H. Schut (Eds.), Handbook of bereavement research:
consequences, coping and care (pp. 613–645). Washington, DC:
American Psychological Association.
3. Neimeyer, R. A. (2000). Searching for the meaning of meaning:
Grief therapy and the process of reconstruction. Death Studies,
24, 541–558.
4. Neimeyer, R. A. (2010). Grief counselling and therapy—the case
for humility. Bereavement Care, 29, 4–7.
5. House of Lords (2013). Select committee on public service and
demographic change—first report: Ready for ageing? http://www.
publications.parliament.uk/pa/ld201213/ldselect/ldpublic/140/14002.
htm.
6. Prigerson, H. G., & Maciejewski, P. K. (2006). A call for sound
empirical testing and evaluation of criteria for complicated grief
proposed by the DSM V. Omega, 52, 9–19.
7. Prigerson, H. G., Horowitz, M. J., Jacobs, S. C., Parkes, C. M., Aslan,
M., Goodkin, K., et al. (2009). Prolonged Grief Disorder: Psycho-
metric validation of criteria proposed for DSMV and ICD-11. PLoS
Medicine, 6(8), e1000121. doi:10.1371/journal.pmed.1000121.
8. Machin, L. (2001). Exploring a framework for understanding the
range of response to loss; a study of clients receiving bereavement
counselling. Unpublished PhD thesis: Keele University, UK.
9. Machin, L. (1980) Living with Loss. Research Report for the
Lichfield Diocesan Board for Social Responsibility (unpublished).
10. Machin, L. (2009). Working with loss and grief: A new model for
practitioners. London: Sage.
11. Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S.
(1978). Patterns of attachment: A psychological study of the
strange situation. Hillsdale, NJ: Lawrence Erlbaum.
12. Horowitz, M. J. (1997). Stress response syndromes, PTSD, grief,
and adjustment disorders (3rd ed.). Northvale, NJ: Jason Aronson.
13. Stroebe, M., & Schut, H. (1999). The Dual Process Model of
coping with bereavement: rationale and description. Death
Studies, 23, 197–224.
14. Martin, T. L., & Doka, K. L. (2000). Men don’t cry…women do:
Transcending gender stereotypes of grief. Philadelphia, PA:
Brunner/Mazel.
15. Mikulincer, M., & Florian, V. (1998). The relationship between
adult attachment styles and emotional and cognitive reactions to
stressful events. In J. A. Simpson & W. S. Rholes (Eds.),
Attachment theory and close relationships (pp. 143–165). New
York: Guilford Press.
16. Seligman, M. E. P. (1998). Building human strength: Psychol-
ogy’s forgotten mission. American Psychological Association
Monitor, 29, 1.
17. Greene, R. (2002). Holocaust survivors: A study in resilience.
Journal of Gerontological Social Work, 37, 3–18.
18. Machin, L., & Spall, R. (2004). Mapping grief: A study in
practice using a quantitative and qualitative approach to explor-
ing and addressing the range of response to loss. Counselling and
Psychotherapy Research, 4, 9–17.
19. Machin, L. (2007). The Adult Attitude to Grief Scale as a tool of
practice for counsellors working with bereaved people. Report
sponsored by Age Concern, Tameside and Keele University.
20. Stroebe, M. S., Folkman, S., Hansson, R. O., & Schut, H. (2006).
The prediction of bereavement outcome: Development of an
integrative risk factor framework. Social Science and Medicine,
63, 2440–2451.
21. Schumacker, R. E., & Lomax, R. G. (2010). A beginner’s guide to
structural equation modeling (3rd ed.). New York: Routledge.
22. Thompson, B. (2004). Exploratory and confirmatory factor
analysis: Understanding concepts and applications. Washington,
DC: American Psychological Association.
23. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in
covariance structure analysis: Conventional criteria versus new
alternatives. Structural Equation Modeling, 6, 1–55.
24. Marsh, H. W., Hau, K.-T., & Grayson, D. (2005). Goodness of fit
evaluation in structural equation modeling. In A. Maydeu-Oliv-
ares & J. J. McCardle (Eds.), Contemporary psychometrics: a
festschrift to Roderick P. McDonald (pp. 275–340). New York:
Routledge.
25. Blunch, N. J. (2008). Introduction to structural equation model-
ling using SPSS and AMOS. Thousand Oaks, CA: Sage.
26. Byrne, B. M. (2012). Structural equation modeling with Mplus:
Basic concepts, applications and programming. New York:
Routledge.
27. Marsh, H. W., Hau, K.-T., & Wen, Z. (2004). In search of golden
rules: Comment on hypothesis-testing approaches to setting cut-
off values for fit indexes and dangers in overgeneralizing Hu and
Bentler’s (1999) findings. Structural Equation Modeling, 11,
320–341.
28. Brown, T. A. (2006). Confirmatory factor analysis for applied
research. New York: Guilford Press.
29. DeMars, C. (2010). Item response theory. Oxford: Oxford Uni-
versity Press.
30. Kline, P. (1993). The handbook of psychological testing. London:
Routledge.
31. Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory
(3rd ed.). New York: McGraw-Hill.
32. Raykov, T. (1997). Estimation of composite reliability for con-
generic measures. Applied Psychological Measurement, 21,
173–184.
33. Robinson, J. P., Shaver, P. R., & Wrightsman, L. S. (1991).
Criteria for scale selection and evaluation. In J. P. Robinson, P.
R. Shaver, & L. S. Wrightsman (Eds.), Measures of personality
and social psychological attitudes (pp. 1–16). New York: Aca-
demic Press.
34. Kaplan, R. M., & Saccuzzo, D. P. (1997). Psychological testing:
principles, applications and issues (4th ed.). Pacific Grove, CA:
Brooks/Cole.
35. DeVellis, R. F. (2003). Scale development: Theory and applica-
tions (2nd ed.). Thousand Oaks, CA: Sage.
36. Spitzer, R. L., Kroenke, K., Williams, J. B., & Lowe, B. (2006).
A brief measure for assessing generalised anxiety disorder: The
GAD-7. Archives of Internal Medicine, 166, 1092–1097.
37. Kroenke, K., & Spitzer, R. L. (2002). The PHQ-9: A new
depression and diagnostic severity measure. Psychiatric Annals,
32, 509–521.
38. Lowe, B., Decker, O., Muller, S., Brahler, E., Schellberg, D.,
Herzog, W., et al. (2008). Validation and standardization of the
Qual Life Res (2014) 23:1211–1220 1219
123
Generalized Anxiety Disorder screener (GAD-7) in the general
population. Medical Care, 46, 266–274.
39. Kroenke, K., Spitzer, R., & Williams, J. (2001). The PHQ-9:
Validity of a brief depression severity measure. Journal of
General Internal Medicine, 16, 606–613.
40. Streiner, D. L., & Norman, G. R. (2008). Health measurement
scales: A practical guide to their development and use (4th ed.).
Oxford: Oxford University Press.
41. Youden, W. J. (1950). An index for rating diagnostic tests.
Cancer, 3, 32–35.
42. Cohen, J. (1988). Statistical power analysis for the behavioral
sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.
1220 Qual Life Res (2014) 23:1211–1220
123