The concurrent validity of the Global Assessment of Functioning (GAF)

British Journal of Clinical Psychology (2002), 41, 417–422
2002 The British Psychological Society
Brief report
The concurrent validity of the Global Assessment
of Functioning (GAF)
Mike Startup*, Mike C. Jackson and Sue Bendix
School of Psychology, University of Wales, Bangor, UK
www.bps.org.uk
Background. Few studies of the validity of the Global Assessment of Functioning
(GAF) have been published and none has shown how GAF ratings are associated with
concurrent ratings of symptoms and social functioning. This article provides such data.
Method. Patients suffering from schizophrenia were assessed at admission to hospital
and at six- and 12-month follow-up, using the GAF, the Scale for the Assessment of
Positive Symptoms (SAPS), the Scale for the Assessment of Negative Symptoms
(SANS) and the Social Behaviour Schedule.
Results. GAF ratings were highly correlated with ratings of symptoms and social
behaviour at both follow-ups but not at initial assessment, although the inter-rater
reliabilities for the measures were good.
Conclusions. The GAF can be rated reliably after minimal training. It provides a valid
summary of symptoms and social functioning among schizophrenic patients provided
they are not assessed when suffering from acute psychotic episodes.
The Global Assessment of Functioning (GAF), based on the widely used Global
Assessment Scale (Endicott, Spitzer, Fleiss, & Cohen, 1976), was included as axis V in
the revised third edition of the Diagnostic and Statistical Manual (DSM-III–R; American
Psychiatric Association, 1987) and has been retained with only minor revisions in DSM-
IV. It is a single rating scale for evaluating a person’s ‘psychological, social and
occupational functioning on a hypothetical continuum of mental health-illness’ and
ranges from 1, representing the hypothetically sickest individual, to 100, representing
the hypothetically healthiest. The scale is divided into 10 equal parts and provides
defining characteristics for each 10-point interval. The defining characteristics include
both symptoms and social functioning.
The Work Group on Multiaxial Issues (Goldman, Skodol, & Lave, 1992) considered
* Requests for reprints should be addressed to Prof Mike Startup, School of Behavioural Sciences, University of Newcastle,
Callaghan, NSW 2308, Australia (e-mail: mike.startup@newcastle.edn.au).
417

418 Mike Startup et al.
that simplicity and unidimensionality were two major advantages of the GAF, and
Skodol, Link, Shrout, and Horwath (1988) commended axis V for providing a summary
of many aspects of functioning that is useful for comparing the severity of illness
between groups of patients with different diagnoses and for measuring change over
time. However, few studies of the validity of the GAF appear to have been completed
and, crucially, there appears to have been no research on the degree to which GAF
ratings are associated with concurrent ratings of symptoms and social functioning. A
few studies have examined such associations for the closely related Global Assessment
Scale (GAS), but they have come to inconsistent conclusions. For example, Endicott et
al. (1976) were not dissatisfied with the modest zero-order correlations they obtained
between the GAS and independent ratings of signs, symptoms and overt behavioural
disorganization, but Holcomb & Otto (1988) concluded on the basis of similar
correlations with signs and symptoms that the GAS had ‘questionable concurrent
validity’ (p. 282). However, previous reports have not provided data on the reliabilities
of the ratings they correlated with the GAS, nor have they shown multiple correlations
between the ratings and the GAS. Thus, their results are difficult to evaluate since one
would not expect high correlations with unreliable measures, nor would one expect
high correlations between ratings of specific symptoms and a global scale if the latter
was designed to summarize information about symptoms and social functioning in
general.
The aims of the present study were, first, to assess the zero-order, partial and
multiple correlations between ratings of symptoms, social functioning and the GAF,
taking into account the reliabilities of the measures; and second, to assess how these
associations change between admission to psychiatric hospital during psychotic
episodes and follow-up at both 6 and 12 months.
Method
Participants were recruits to a controlled trial of cognitive therapy. Consecutive
admissions to three psychiatric hospitals were considered eligible for inclusion if they
were aged between 18 and 65, had received a clinical diagnosis of schizophrenia,
schizophreniform or schizoaffective disorder, were suffering an acute psychotic
episode, and if there was no evidence of organic mental disorder. The 214 patients
who were considered to be eligible were invited to participate when their psychiatrist
declared them to be capable of informed consent. Of these patients, 84 declined the
invitation. Those who accepted were then excluded if more than 28 days had passed
since they were admitted or if, during a baseline assessment, their diagnoses could not
be confirmed according to DSM-IV criteria, they had been dependent on alcohol or
illicit drugs during the past year, or their IQs were below 80. In all, 66 patients were
excluded according to these criteria, leaving 64 for the present sample.
At initial assessment and at both 6- and 12-month follow-up, patients were rated,
following a structured interview, on the Scale for the Assessment of Positive Symptoms
(SAPS: Andreasen, 1984) and the Scale for the Assessment of Negative Symptoms (SANS:
Andreasen, 1989). Following a structured interview with a ‘best informant’, usually a
qualified mental health professional, the patients were rated by the interviewer on the
Social Behaviour Schedule (SBS: Wykes & Sturt, 1986). This schedule covers 20
behaviour areas that have been found to describe the major difficulties experienced by
patients with long-term impairments. At the end of this interview the patients were

ated by both interviewer and informant independently, blind to each other’s ratings,
on the GAF. Only a brief guide to the GAF (5min maximum) was provided to informants
before they made their ratings. Ratings were based on either symptoms or social
functioning, whichever had been worse. All interviews referred to the preceding
month.
At 6- and 12-month follow-up, interviews could not be conducted with 16 and 19
patients, respectively and informants were unable to provide information on eight and
14 patients, respectively. Most of these data are missing because the patients had
moved out of the area or they were refusing contact with services.
In order to assess the reliabilities of the ratings, 12 interviews with patients were
sampled at random at each assessment point, with the constraint that no patient was
sampled more than once, and audio-recordings of these interviews were rated on the
SANS and the SAPS while blind to the interviewer’s ratings. These ratings were for the
global items of each subscale except for affective flattening and avolition. Since some of
the items of the latter subscales depend on observations of the patients’ behaviour
during interview, which could not be made from audio-recordings, independent ratings
of affective flattening were based on affective non-responsivity and lack of vocal
inflections, and ratings of avolition were based on impersistence and physical anergia.
Independent ratings of bizarre behaviour could not be made. Ten interviews with
informants were also sampled under the same constraints at each assessment point and
recordings of these interviews were rated on the SBS while blind to the interviewer’s
ratings. All of these independent ratings were made by the first author.
Results
Validity of the GAF
Inter-rater reliabilities (Table 1), assessed as intraclass correlations for two raters, were
computed for the GAF, the global items of the SANS and the SAPS, and for the three
dimensions that emerged from a principal components analysis of the SANS and SAPS
by Andreasen, Arndt, Alliger, Miller, and Flaum (1995): negative symptoms, psychotic
symptoms and disorganization. Reliabilities were also computed for a summary measure
and four subscales of the SBS. The ‘mild and severe behaviour’ (Wykes & Sturt, 1986)
score summarizes problematic social behaviour by counting how many of the 20
behaviours assessed by the scale receive ratings of at least 2 (indicating at least mildly
problematic behaviour). The SBS subscales were derived from a principal components
analysis of a large sample of patients with a diagnosis of schizophrenia (Harvey, Curson,
Pantelis, Taylor, & Barnes, 1996). Based on this analysis, the subscale ‘thought
disturbance’ is the average of ratings for the items ‘incoherence of speech’, ‘odd/
inappropriate conversation’ and ‘poor attention span’; the subscale ‘social withdrawal’
is the average for the items ‘little spontaneous communication’, ‘poor self-care’,
‘slowness’ and ‘underactivity’; the subscale ‘depressed behaviour’ is the average for the
items ‘depression’ and ‘suicidal ideas/behaviour’; and the subscale ‘anti-social
behaviour’ is the average for the items ‘hostility’, ‘socially unacceptable habits’ and
‘destructive behaviour’. Inter-rater reliabilities were at least adequate, and were usually
excellent, for nearly all of these measures at the three assessments (see Table 1).
Table 1 also shows the zero-order correlations between the measures and the
average of the GAF ratings made by the interviewer and the informant. These were
variable across assessments. For example, most of the measures of symptoms and social
behaviours correlated significantly with the GAF at 6-month follow-up, but only bizarre
419

420 Mike Startup et al.
Table 1. Inter-rater reliabilities of measures, sample sizes and correlations with the GAF
behaviour and ‘mild and severe behaviour’ correlated significantly at the initial
assessment. In order to assess the independent contributions of different kinds of
symptom and of social behaviour to global assessments, three multiple regression
analyses were conducted, one for each assessment. In each of these analyses the GAF
was the dependent variable and the three dimension scores from the SANS and SAPS,
together with the ‘mild and severe behaviour’ index, were entered as predictors. At
intake, the multiple correlation (R = .37) was only marginally significant
(F(4,59) = 2.37, p = .06), but at 6-month follow-up (R = .77, F(4,43) =15.9,
p < .0001) and at 12-month follow-up (R = .76, F(4,40) =13.8, p < .0001) the multiple
correlations were both large and highly significant. Table 1 shows the partial
correlations from these analyses. These reveal that all four predictors made significant
independent contributions at 6 months though only negative symptoms and social
behaviour had significant partial correlations at 12 months.
Discussion
Intake Six-month follow-up 12-month follow-up
ICC r pr ICC r pr ICC r pr
GAF .89 .94 .95
Symptoms: N 64 48 45
SANS total .88 –.12 –.10 .71 –.49* –.44* .96 –.63* –.54*
Affective flattening .86 –.06 .83 –.36* .95 –.35*
Alogia .94 .02 .00 –.31* .74 –.27
Avolition .80 –.18 .81 –.39* .93 –.49*
Anhedonia .63 –.13 .86 –.44* .97 –.71*
SAPS: psychotic dimension .94 .08 .10 .92 –.48* –.38* .97 –.15 –.07
Hallucinations .94 .11 .88 –.36* .96 –.04
Delusions .75 –.04 .93 –.57* .96 –.24
SAPS: disorganization .89 –.23 –.15 .84 –.40* –.40* .73 –.29 –.01
Thought disorder .99 –.08 .§ –.17 .§ –.24
Bizarre behaviour –.25* –.46* –.21
Social behaviour schedule: N 64 56 50
Mild and severe behaviour .96 –.31* –.27* .94 –.59* –.39* .98 –.64* –.46*
Thought disturbance .96 –.15 .98 –.54* .99 –.56*
Social withdrawal .97 –.16 .72 –.32* .96 –.39*
Depressed behaviour .63 –.14 .96 –.10 .92 –.51*
Anti-social behaviour .96 –.19 .96 –.20 .98 –.23
* p < .05.
Notes: ICC = intraclass correlations, all significant at p < .05 except for alogia at six months; r = zeroorder
correlations; pr = partial correlations from multiple regression analyses; SANS = Scale for the
Assessment of Negative Symptoms; SAPS = Scale for the Assessment of Positive Symptoms;
§ = insufficient variance to calculate ICC.
Inter-rater reliabilities for the GAF were excellent at all three assessments even though
the informants received only very brief introductions to the instrument. These results

confirm the finding of Jones, Thornicroft, Coffey, and Dunn (1995) that raters who are
mental health professionals need only a brief training in order to use the scale reliably.
The multiple correlations between the GAF and the measures of symptoms and
social behaviour were large and highly significant at the follow-up assessments, but not
at intake. This pattern is similar to that reported for the GAS (Endicott et al., 1976) and
may be explained largely by the reduced variability of the measures at intake; for
example, the variance of the GAF was at least three times greater at the follow-ups.
Although the multiple correlations were almost the same at the two follow-ups, the
patterns of partial correlations were somewhat different. At 6 months, all of the
variables, including the psychotic and disorganized dimensions, made significant,
independent contributions to the prediction of GAF scores, but at 12 months only
negative symptoms and social behaviour made significant contributions. This pattern
might be explained if there was a tendency for negative symptoms and social behaviour
to be more disabling than positive symptoms at 12 months but equally disabling at 6
months. However, this explanation cannot be tested with the present data.
In conclusion, the GAF provides a valid summary of symptoms and social functioning
among schizophrenic patients provided they are not assessed when suffering from
acute psychotic episodes. Since it has been found (Endicott et al., 1976) that global
measures are more sensitive to change than measures of single symptom dimensions,
the GAF has a valuable role in monitoring the effects of clinical interventions over time
and in outcome research with limited statistical power and, since it can be rated reliably
by mental health professionals after minimal training, such monitoring can easily be
integrated with routine community care. Jones et al. (1995) found that the GAF scores
of severely mentally ill patients are associated with their current support needs, and
that increases in support in the preceding month are associated with increases in GAF
ratings. Thus, there is also evidence that the GAF fulfils the hope that axis V ratings will
generally reflect patients’ current need for treatment or care.
Acknowledgements
This research was supported by grant RC/012 from the Wales Office of Research and
Development for Health & Social Care.
References
Validity of the GAF
American Psychiatric Association (1987). Diagnostic and statistical manual of mental disorders
(3rd ed., revised). Washington, DC: Author.
Andreasen, N. C. (1984). Scale for the Assessment of Positive Symptoms (SAPS). University of
Iowa: Iowa City.
Andreasen, N. C. (1989). Scale for the Assessment of Negative Symptoms (SANS). British Journal
of Psychiatry, 155 (Suppl. 7), 53–58.
Andreasen, N. C., Arndt, S., Alliger, R., Miller, D. & Flaum, M. (1995). Symptoms of schizophrenia:
Methods, meanings, and mechanisms. Archives of General Psychiatry, 52, 341–351.
Endicott, J., Spitzer, R. L., Fleiss, J. L., & Cohen, J. (1976). The Global Assessment Scale: A
procedure for measuring overall severity of psychiatric disturbance. Archives of General
Psychiatry, 33, 766–771.
Goldman. H. H., Skodol, A. E., & Lave, T. R. (1992). Revising axis V for DSM-IV: A review of
measures of social functioning. American Journal of Psychiatry, 149, 1148–1156.
421

422 Mike Startup et al.
Harvey, C. A., Curson, D. A., Pantelis, C., Taylor, J., & Barnes, T. R. E. (1996). Four behavioural
syndromes of schizophrenia. British Journal of Psychiatry, 168, 562–570.
Holcomb, W. R., & Otto, R. L. (1988) Concurrent validity of the Global Assessment Scale: What’s
in a number? Psychological Reports, 62, 279–282.
Jones, S. H., Thornicroft, G., Coffey, M., & Dunn, G. (1995). Abrief mental health outcome scale:
reliability and validity of the Global Assessment of Functioning (GAF). British Journal of
Psychiatry, 166, 654–659.
Skodol, A. E., Link, B. G., Shrout, P. E., & Horwath, E. (1988). The revision of axis V in DSM-III–R:
should symptoms have been included? American Journal of Psychiatry, 145, 825–829.
Wykes, T., & Sturt, E. (1986). The measurement of social behaviour in psychiatric patients: An
assessment of the reliability and validity of the SBS schedule. British Journal of Psychiatry,
148, 1–11.
Received 25 May 2001; revised version received 9 January 2002