You are on page 1of 7

British Journal of Social and Clinical Psychology (1979), 18, 237-243 Printed in Great Britain 237

Expressed emotion: A review


Liz Kuipers

Expressed emotion refers to emotional aspects of speech for which ratings have been derived. In the
relatives of discharged schizophrenic patients, these ratings are predictive of outcome suggesting the
importance of social and family factors on the prognosis of schizophrenicsliving with their families. The
development, measurement and current use of the concept are discussed.

I think I can detect undertones in the voice of the person I’m speaking with - I have a very sensitive
ear.. .The voice is a dangerous instrument. I don’t mean the timbre of the voice which may be high,
low, melodious or grating. I’m not talking about the sound but about the inner world from which it
springs.. .(Knut Hamsun, Mysteries).
The measurement of emotional states from speech has a relatively long history. Such
measurements have attempted to analyse not only the content of speech but also aspects of the
voice that give expression to utterances. In 1938 Eisenberg & Zalowitz wrote ‘Here we are not
interested in the content of speech but in the manner and qualities of speech, for example, loud
or-soft, fast or slow, smooth or hesitant, even or uneven tempo, etc.’ (Eisenberg & Zalowitz,
1938, p. 260). Similarly Sokin (1953) distinguished between two channels of speech. He
described a verbal channel which carried potential semantic information and a vocal channel
which carried potential affective information.
A particular use of ‘vocal’ attributes has been in rating the emotions expressed by relatives of
the mentally ill. This work, by Brown and colleagues (Brown et al., 1958), started in the 1950s.
Initially Brown et al. (1958) and Brown (1959) in a study of 229 discharged male patients,
two-thirds of whom were diagnosed as schizophrenic, found that patients’ success or failure in
the community seemed to be associated with the kind of living group they returned to.
Specifically, those returning to large hostels, parents and marital partners were less able to
survive in the community in the following year than those returning to live with siblings or in
lodgings. Prolonged contact with the living group made relapse more likely - if, for instance,
both patient and parent were unemployed and so spent most of the time together. One
limitation of this study was the heterogeneous subject group. A second was that it was
retrospective. Thus, in a more rigorous prospective study, Brown et al. (1962) attempted to
specify which aspects of the living group contributed to the result. The male patient group
followed up on this occasion were all diagnosed as schizophrenic (n = 128), and both the patient
and the key (female) relative were interviewed at or shortly before discharge, 2 weeks after the
patient arrived home and at 1 year follow-up. At these interviews the ‘emotion expressed’, the
level of ‘dominance’ and degree of ‘hostility’ were measured for both patient and relative. In
practice, the ratings of the relative alone were most useful. Patients were placed in ‘high’ or
‘low emotional involvement ’ groups according to whether the key relative was given a high
rating on ‘expressed emotion’ or on ‘hostility’. Dominance ratings did not add any extra
information. On this basis Brown et al. (1962) again found that patients returning to homes rated
high in ‘emotional involvement ’ had higher relapse rates than those in low ‘emotional
involvement ’ homes. There was some relationship between high ‘emotional involvement ’,
deterioration in the patient, level of unemployment and severity of symptoms. Also, those with
high involvement but with reduced contact had lower rates of relapse; this applied only if the
patient’s behaviour was moderately or severely disturbed at discharge.
The ratings of emotion in these studies were still very crude. There was no audiotaping of the
interviews, and affect was rated on four-point scales - ‘made by commonsense signs such as

OOO7-1293/79/0602-0237$02.00/0 @ 1979 The British Psychological Society


238 Liz Kuipers
content of speech, tone of voice and gesture’ (Brown et al., 1962, p. 57) - observed during the
interviews. Interestingly, one of the original findings of the 1958 work was not confirmed, as
patients returning to lodgings deteriorated to the same extent as those going to live with parents.
It was postulated that lack of supportive ties may be just as ‘harmful’ as ties involving tension
and hostility. This discrepancy in the two studies escapes further comment because the later
research only assessed those patients returning to relatives. It is also in this paper that the term
‘expressed emotion’ (EE) first appears.
. After this broad confirmation of part of the original finding Brown and his co-workers became
justifiably more concerned with techniques of measurement. Up to that point, only poorly
defined concepts had been established. For instance it was not clear which aspects were
sufficient to define a key relative as ‘highly involved’. There was also the issue of whether such
measures were reliable and, if so, whether they were valid. The subsequent two papers (Brown
& Rutter, 1966; Rutter & Brown, 1966) were exclusively concerned with these issues. A
non-schedule standardized interview developed on 80 families and called the Camberwell family
interview, was designed as a basis for ratings. [A non-schedule, standardized interview is one
with certain set questions but no predetermined order of questioning; the interviewer is free to
follow the content of the respondents’ replies as long as all questions are eventually covered
(Richardson et al., 1965).] The authors were interested in three different kinds of ratings: first,
the ‘objective’ rating of events and activities; second, a subjective but reliable rating of aflect,
using vocal and semantic aspects of speech; and third, a summary score for relationships.
Optimum ways of achieving these were discussed in detail. Data from interviews with 30
families are presented and show that interjudge reliabilities for all scales were generally high
(Pearson’s r 2 0.8).
At this stage the use of the vocal channel to make ratings of affect is just beginning to emerge.
Brown & Rutter (1966) appear to be particularly influenced by an early study of Hamburg et al.
(1958). Hamburg and his colleagues rated emotion during unspecified stress experiments, using
speech and non-verbal cues (such as grimaces) rather than vocal cues as such. While the ethics
of such stress experiments are very questionable, particularly as quite severe emotions seemed
to have been elicited (the experiments were apparently performed on military personnel and
funded by the U.S. Army), the methodology of the emotional rating was impressive and the
investigators obtained high interjudge reliabilities on scales of anger, depression and anxiety (all
reliability coefficients 20.78).
The authors attribute these results firstly to a long sampling period - at least 3 hours per day
over 4 days to allow a wide range of behaviour in a variety of situations; secondly they tried to
minimize the amount of inference needed to decide on ratings, i.e. they tried to make the ratings
as objective as possible; thirdly, they had precise definitions of the unit being observed;
fourthly, they had well-trained raters. Rutter & Brown (1966) give two examples of the
emotional scales they developed, i.e. (1) critical remarks and (2) warmth. There is a descriptive
account of methods used to elicit and score emotional and attitudinal content in the interview.
The main sources of data are spontaneous expressions of emotion by the relative while factual
information is being sought. The ratings are based on several factors, best described by the
authors. ‘The content of what was said was taken into account, but more emphasis was laid on
the way things were said. Interviewers were expected to recognize emotions by observing
differences in the speed, pitch and intensity of speech. To a lesser extent expression and gesture
were taken into account’ (Rutter & Brown, 1966, p. 45). Reliabilities between raters were
generally high (Pearson’s r 2 0.8 range r = 0.6 to r = 0.92). This was attributed to: the training of
raters; lack of inference in scales; specification of cues used; uniform interview stimuli and
unipolar scales to reduce halo effects. They concluded that (1) reliable effect ratings can be
produced and (2) vocal channels are as important as verbal ones; the tone, pitch and rhythm of
the voice are necessary aspects of emotional ratings.
Expressed emotion: A review 239
Rutter & Brown (1966) also claim to be particularly concerned with the issue of the validity
of ratings. This is a more difficult area. They equate validity with consistency across situations,
and they then compare ratings from two different interviews. Rutter & Brown later admit that
this ‘is not strictly speaking a measure of validity’ (p. 46). However as the across-interview
correlation for an affect scale is quite high between the two different interviews (Pearson’s
r = 0.68) this is taken as supportive evidence of ‘good validity ’ (p. 46). In fact this exercise does
not conform to any traditional definition of validity, and it is more accurately described as a
demonstration of equivalent form reliability. Rutter & Brown (1966) themselves leave the issue
at this however.
In order to test the validity of expressed emotion ratings, it would be necessary to use some
external criterion. One way of doing this is to show that a measure of EE in relatives is related
to some measurable external response in the schizophrenic patient. This has been possible in the
psychophysiological area. Spontaneous skin conductance fluctuations of the schizophrenic in the
presence of a high or low EE relative compared to when alone have been taken (Tarrier et al.,
1979). The results of this preliminary recent work suggest that high EE relatives prevent the skin
conductance response of the schizophrenics from habituating in this novel situation, whereas low
EE relatives potentiate habituation. The study indicates that low EE relatives may be having
some ‘calming’ effect on the patient, whereas high EE relatives do not. If the work could be
replicated this would provide some evidence for the concurrent validity of EE ratings.
An alternative possibility is to measure independently some other characteristics of relatives
and then see if they are related to the EE ratings. Some behavioural aspects of relatives’
responses to patients while interacting with them are currently being investigated, and this may
eventually provide evidence for content validity: evidence about the particular behaviour of
relatives towards the patient that a measure of EE is reflecting. It is worth noting however that
while adequate evidence for the validity of the EE rating has not yet been demonstrated, the
consistent finding over the succeeding 10 years and with different samples that ratings of EE are
predictive of schizophrenic outcome (to be discussed), does make it increasingly likely that the
ratings do reflect an important and subtle aspect of the home atmosphere.
In 1972 Brown et al. completed a further follow-up study using the more sophisticated
methodology and specific rating criteria that had been developed. The relatives of 101
schizophrenics returning to their families were interviewed at admission, 2 weeks after
discharge, and at 9 months after discharge (or at the time of relapse if this was earlier). Patients
and relatives were interviewed separately, except at 2 weeks after discharge when a special joint
interview, designed to get the whole family talking about their recent medical or social services
contact, was conducted. Of all these interviews the one with the relative alone at admission was
found most predictive of subsequent relapse rate. Both initial diagnosis and relapse were defined
in terms of a standard psychiatric interview, the Present State Examination (Wing et al., 1974).
This paper is the first one to describe and define clearly all the affect ratings. Both critical and
positive remarks were counted and overall ratings of criticism, hostility, warmth, emotional
overinvolvement and dissatisfaction were made. Of these, critical remarks and warmth were the
ones most dependent on vocal cues. The term, ‘emotional involvement’ was replaced by that
of ‘expressed emotion’. An index of EE was obtained either from a rating of marked
emotional overinvolvement, seven or more critical remarks, hostility, or any combination of
these. In fact critical remarks were the most predictive accounting for 77 per cent of high EE
families. The results of this study confirmed their earlier findings (Brown et al., 1962). Patients
returning to a relative rated high on the EE index tended to relapse more than those returning to
low EE families; 58 per cent relapsed in the high EE group and 16 per cent in the low EE group.
The difference in relapse rate was independent of the patients’ prior disturbed behaviour. Finally
those patients in high contact with high EE relatives relapsed more frequently than those in low
contact. (‘High contact ’ was defined as more than 35 hours per week in the same room.)
240 Liz Kuipers

The affect scales were explained in some detail. The ratings were now made from audio-tapes
of the interview. They relied on raters having been trained for at least 3 months using
well-established criteria. All the scales depend upon speech content but the unique feature is
their use of changes in vocal qualities: changes in the tone, pitch and speed of the voice. This is
particularly necessary for the rating of a critical remark. A full understanding of the vocal cues
relies on a training procedure involving listening to tapes. Some idea however of the concepts
involved can be conveyed by written description. For instance, the words ‘Every morning he
goes to buy the newspaper’ have innocuous content but can be said in a variety of ways to give
different emotional meanings. If they were intended to be critical this would be conveyed in
writing by stress marks - ‘Every morning he goes to buy the newspaper ’. A critical remark can
be defined as a comment that shows that the respondent dislikes, resents or disapproves of the
patient o’r his behaviour; to qualify as critical, it must have a vocal component more intense than
mere dissatisfaction (which itself is not predictive of relapse). The basis for the method is an a
priori agreed system of deciding when sufficient tonal change has occurred to justify a rating as a
critical remark. The decision can be made because sufficient listening experience enables an
individual to recognize affective aspects of speech reliably.
The definition of hostility is as a more extreme kind of critical remark: in contrast to criticism
it is based on content alone, e.g. ‘he is an evil person’. As hostility is associated with critical
comments it was later discarded as an independent definition of a high EE family (Vaughn &
. alternative feature of a high EE family is the presence of considerable
Leff, 1 9 7 6 ~ )The
emotional overinvolvement. This is defined as a tendency to overprotect, to overdramatize
incidents, go into excessive and inordinate detail, and show emotional distress in the interview.
It occurs in parents who may treat the patient as a child again. Spouses tend to resist such
dependency in a partner, and do not usually show emotional overinvolvement.
The relationship of ‘expressed emotion’ to relapse was further refined in a replication of the
1972 study by Vaughn & Leff ( 1 9 7 6 ~ )Using
. a shortened version of the Camberwell Family
Interview, as it was found that the number of critical remarks was independent of interview length
(Vaughn & Leff, 1976b), they repeated the work on a group of 37 discharged schizophrenics and
30 neurotic depressives. Their results confirmed the original findings: schizophrenic patients
returning to high EE homes had a higher likelihood of relapse. Again this was independent of the
patients’ prior behaviour disturbance. Further, when the subjects were combined with the earlier
(1972) study it was possible to see that there were other protective factors operating. Patients
returning to high EE families, who remained on medication or were in low contact with the
family, relapsed less often. If both factors operated the effects were additive and relapse rates,
even in high EE families, were only I5 per cent in the following 9 months - a risk as low as that
in a low EE family. On the other hand if neither factor was operating the relapse rate was 92 per
cent. It is important to note that the EE index used here (Vaughn & Leff, 1976a) omitted
hostility (which was not predictive by itself), and was based on a reduced threshold of six rather
than seven critical remarks. This meant that high EE families were those with a rating of
marked emotional overinvolvement or six or more critical remarks. This was a post hoc decision
and produced a better cut-off point in terms of relapses in defined high EE families. Its use was
argued in terms of the initially arbitrary decision to use seven critical remarks as a dividing line
between high and low E E families. The decision to use a cut-off point at all remains arbitrary
as it is possible that critical remarks occur on a continuum which is related to the length of time
the patient stays well. However this is a hypothesis which remains to be tested.
Over the 20 years encompassed by this research the concept of ‘expressed emotion ’ has
evolved and changed. From being a crude measure of emotional response in relatives of
schizophrenic patients it has become a highly specific one. Aspects of the original concept not
found empirically to be predictive have been jettisoned. It is now a measure of the number of
critical remarks and the level of overinvolvement, spontaneously expressed by the relative in the
Expressed emotion: A review 241

course of a factual interview. (The relative is asked about various aspects of the illness, its
effects on themselves and other household members, together with questions about the
household routine, and amount of time spent in the same room.) EE has been shown to be a
useful way of quantifying aspects of the family situation when a schizophrenic family member is
admitted to hospital (Brown et al., 1972; Vaughn & Leff, 1976~).Critical remarks remain the
most predictive of the component ratings; hostility is not independent of criticism while marked
emotional overinvolvement is an infrequent occurrence. In Vaughn & Leff (1976a) a rating of
marked emotional overinvolvement alone only added 5 per cent more families to the high EE
category.
EE remains an esoteric measure. Although it is possible to assess critical remarks and
emotional overinvolvement reliably, it is still impossible to say what exactly the measure is
reflecting. This is partly because EE is defined purely in terms of vocal and semantic qualities. It
does not as yet have any external definition in terms of what it is that the relatives are, or are
doing that is having the effect on schizophrenic outcome. Some clues to this are only just
appearing, for instance in the psychophysiological data discussed earlier (Tarrier et al., 1979).
Aspects of high and low EE families can however be described in terms of what they say and
how they seem to cope with similar situations. One technique has been to do a content analysis
of the critical remarks made by relatives (Vaughn, 1977). One-third of the relatives made no
critical remarks at all. Of those that did, two-thirds of the remarks were related to longstanding
attributes of the patient rather than to recent changes in behaviour, i.e. there was no
differentiation between pre- and post-illness behaviour, or even an awareness of an illness
process. ‘She’s always been selfish and spoilt.’ ‘You never could do a thing with her.’ In other
words high EE families tended to blame the person, and not to see the changes in behaviour as
due to anything but an intensification of previously noted faults. Low EE relatives were better at
recognizing behaviour as due to illness, and attributing problems to this. Their ability to refer the
behaviour to some cause seemed to reduce the emotional flavour of their responses in the
interview. ‘I knew something must be wrong, she never used to be like this.’
High EE families appear as ones who cope least well with the crises and are most worried and
upset. They may have an overintense relationship with the patient or may be resentful of his
behaviour. One characteristic of these families is that they emphasize the personal impact of the
episode, rather than its impact on the patient. The following examples from a current study
illustrate this. One wife said, ‘For four years I’ve had him hanging round my neck like this
microphone.. .There’s no way I can lose him he’s like a sore or a mole’ (013). A mother said,
‘I’ve been on edge. I watch him upstairs, I follow him, I can’t help it. I’m just watching him all
the time: it’s nearly broke me’ (001). Or a husband: ‘I was tensed up all the time, wondering
what I was going to do that was going to be wrong next ’ (007). Parents are particularly likely to
become emotionally overinvolved and to start treating the patient as a child again. One mother,
talking about her 30 year old son, said ‘He and I we’re very close.. .he usually gives me a ring
during the day (from work). . .if anything I feel even more protective ’ (019).
In contrast, low EE families are more tolerant and cope with incidents calmly. A son
announced one night that he had a mission to kill his father. The father wasn’t unduly disturbed
by this: ‘It didn’t worry me, I thought if he goes for me I’ll put him in his place’ (072) (Vaughn,
1977). One mother whose son had been exposing himself at the railway station and was in
trouble with the police said: ‘He went out the other night and I thought I’ll give it an hour and if
he doesn’t come back then I’ll call the police. But he did come back - he’d just been for a walk’
(003). These examples indicate the different methods which relatives adopt in order to cope with
potentially upsetting and disturbed behaviour in a family member.
Thus ratings of EE appear to be a way of assessing different attitudes and coping responses in
relatives when faced with disturbed behaviour in a schizophrenic family member. It is of note
that despite the inference of work on ‘causal family influences ’ in schizophrenia - that families
242 Liz Kuipers
are instrumental in causing schizophrenic breakdowns (Bateson et al., 1956; Laing & Esterson,
1964) - half of the families seen in the Brown studies coped remarkably well with extremely
difficult situations and were generally accepting of changes in the relative. In this sense the
family reactions, whether high or low in EE, are like those of families coping with, say, a
handicapped child. The main difference is that in schizophrenia the disability tends to occur
relatively suddenly to an adult family member.
Apart from such descriptive data, expressed emotion in relatives cannot be further illustrated
as yet. There is some recent evidence however that Greenley (1979) has been tapping similar
family influences, suggesting that the concept is not totally esoteric. Greenley followed up and
interviewed relatives of 3 1 functional psychotic patients 4 years after their first admission. He
found that families who expected the patient ‘to be like a ten year old child’, ‘expected more
symptoms ’, ‘did not like the patient ’ or ‘did not want him home ’, had more frequent
readmissions. In this study, unfortunately, there was no independent way of checking if
readmissions were equivalent to relapses as used in the Brown studies. However, Greenley
concludes ‘. . .that frequently readmitted patients have families who expect them to be
dependent, and at the same time do. not like the patient’ (p. 21) -the essence of what is rated as
emotional overinvolvement, and criticism.
The use of listeners to rate vocal qualities and to judge emotional states reliably in these
families has progressed considerably since the original findings in the 1950s. The technique is
now sophisticated, and with sufficient training it appears possible for English speakers with a
research or clinical background and good hearing to become reliable in the ratings. (Two Hindi
speakers, with English as a second language, have also been reliably trained in EE using the
English tapes.) However at present it is still a research tool. It remains time consuming to
administer the interview and rate E E even with the new shortened version (Vaughn & Leff,
19766) - an average of 4 hours per individual is required. Other uses of the rating are just
beginning to be explored. Work comparing EE in relatives and skin conductance ratings in
patients in the presence of the relative has produced provocative results discussed earlier
(Tarrier et al., 19791, which need to be extended. Vaughn & Leff (1976a) looked at EE in two
patient groups, schizophrenics and neurotic depressives, and found that its effects on relapse
were different: the neurotic depressives were even more sensitive to criticism and relapsed at a
lower EE threshold. Further work on other patient groups might clarify this area. Preliminary
studies on the cultural specificity of EE are in progress (World Health Organization Project 5.2).
There is also some evidence that EE ratings could be re-analysed using structural voice analysis
(e.g. Alpert et al., 1963) rather than the ratings of listeners.
Finally the most immediately practical use of EE would be as a clinical screening device; to
determine which families might be most likely to benefit from intervention to reduce relapse
rates. The earlier studies make clear the immediate targets for a possible hierarchy of
interventions: ensuring patients stay on medication, reducing the amount of face-to-face contact
with high EE relatives and, if necessary, modifying levels of E E itself. Pilot work in this area is
being attempted. It is to be hoped that this useful predictive method will prove equally relevant
in helping to improve the outcome for schizophrenics living with their families.

Acknowledgements
I would like to thank Paul Bebbington, Christopher Tennant, Julian Leff and Joe Kuipers for reading and
commenting on earlier drafts of the paper.
Expressed emotion: A review 243

References
ALPERT,M., KURTZBERG, R. & FRIEDHOFF, A, LAING,R. D. & ESTERSON,D. (1964). Sanity,
(1963). Transient voice changes associated with Madness and the Family. London: University of
emotional stimuli. Archives of General Psychiatry, London Press.
8 , 362-365. RICHARDSON, S. A., DOHRENWEND, B. S. & KLEIN,
BATESON,G., JACKSON, D., HALEY,J. & D. (1965). Interviewing: Its Forms and Functions.
WEAKLAND, J. (1956). Towards a theory of New York: Basic Books.
schizophrenia. Behavioral Science, 1, 251-264. RUTTER,M. & BROWN,G. W. (1966). The reliability
BROWN,G. W. (1959). Experiences of discharged and validity of measures of family life and
chronic schizophrenic mental hospital patients in relationships in families containing a psychiatric
various types of living group. Millbank Memorial patient. Social Psychiatry, 1 , 38-53.
Fund Quarterly, 37, 105-131. SOSKIN, W. F. (1953). Some aspects of
BROWN,G. W., BIRLEY,J. L. T. & WING,J . K. communication and interpretation in
(1972). Influence of family life on the course of psychotherapy. Paper read at American
schizophrenic disorders: A replication. British Psychology Association, Cleveland, September.
Journal of Psychiatry, 121, 241-258. TARRIER, N., VAUGHN,C., LADER,M. H. & LEFF,
BROWN,G. W., CARSTAIRS, G. M. & TOPPING, G. C. J. P. (1979). Bodily reactions to people and events
(1958). The post hospital adjustment of chronic in schizophrenia. Archives of General Psychiatry
mental patients. The Lancet, ii, 685489. (in press).
BROWN,G. W., MONCK,E. M., CARSTAIRS, G. M. VAUGHN,C. E. (1977). Patterns of interactions in
& WING,J. K. (1962). The influence of family life families of schizophrenics. In H. Katschnig (ed.),
on the course of schizophrenic illness. British Schizophrenia: The Other Side. Vienna: Urban &
Journal of Preventative Social Medicine, 16, 55-68. Schwarzen berg.
BROWN,G. W. & RUTTER,M. L. (1966). The VAUGHN,C. E. & LEFF, J. P. (1976a). The
measurement of family activities and influence of family and social factors on the
relationships. Human Relations, 19, 241-263. course of psychiatric patients. Brifish Journal of
EISENBERG,P. & ZALOWITZ,E. (1938). Judging Psychiatry, 129, 125-137.
expressive movement: 111. Judgements of VAUGHN,C. E. & LEFF, J. P. (1976b). The .
dominance feeling from phonograph records of measurement of expressed emotion in the families
voice. Journal of Applied Psychology, 22, 620-63 1. of psychiatric patients. British Journal of Social
GREENLEY, J. R. (1979). Family symptom tolerance and Clinical Psychology, 15, 157-165.
and rehospitalisation experiences of psychiatric WING,J. K.,COOPER,J. E. & SARTORIUS,N.
patients. Personal communication. (1974). The Description and Classigcation of
HAMBURG, D. A., SABSHIN,M., BEARD,F. A., Psychiatric Symptoms: A n Instruction Manual for
GRINKER, R. R., KORCHIN, S. S., BASOWITZ,H., PSE and Catego System. London: Cambridge
HEATH, H. & PERSKY,H. (1958).'Classification University Press.
and rating of emotional experiences. Archives of
Neurology and Psychiatry, 79, 415-426.

Received 20 November 1978

Requests for reprints should be addressed to Liz Kuipers, MRC Social Psychiatry Unit, Institute of
Psychiatry, London SE5 8AF.

You might also like