The British Journal of Psychiatry (2015)

206, 456–460. doi: 10.1192/bjp.bp.114.147553

Causal and mediating factors for anxiety,

depression and well-being
Peter Kinderman, Sara Tai, Eleanor Pontin, Matthias Schwannauer, Ian Jarman and Paulo Lisboa

Background
The relationship between well-being and mental ill health 27 397 participants revealed different association pathways
is complex; people may experience very low levels of for the two constructs. Anxiety and depression were
well-being even in the absence of overt mental health associated with negative life events mediated by rumination;
problems. low levels of subjective well-being were associated with
material deprivation and social isolation, mediated by
Aims adaptive coping style.
This study tested the hypothesis that anxiety, depression and
well-being have different causal determinants and Conclusions
psychological mediating mechanisms. Our findings support the ‘two continua’ model of the
relationship between psychological well-being and mental
Method health problems, with implications for both treatment and
The influence of causal and mediating factors on anxiety, prevention.
depression and well-being were investigated in a cross-
sectional online questionnaire survey hosted on a UK Declaration of interest
national broadcasting website. None.

Results Copyright and usage

Multivariate conditional independence analysis of data from B The Royal College of Psychiatrists 2015.

The relationships between mental health, well-being and mental years (s.d. = 14.3), 61% were women, 93% were ‘White British’
illness are complex, and people may experience very low levels or ‘White other’ and 88% were from the UK. To determine
of well-being even in the absence of overt mental health whether the sample was representative of the UK population,
problems.1 Despite general agreement that these three concepts results were compared with national data for England and Wales.5
are important targets for health and social care, there is no
consensus about the relationship between them. This conceptual Procedure
distinction is best exemplified in the ‘two continua’ model,2 which The Stress Test was promoted via multimedia formats (television,
suggests that mental illness and mental well-being are related yet radio and online) and launched on BBC Radio 4’s All in the Mind,
distinct phenomena, rather than ends of a single measurement a flagship documentary focusing on issues of the human mind.
continuum. One important way to disambiguate ‘psychological The test’s URL (www.bbc.co.uk/labuk/experiments/stress/) was
well-being’, ‘mental health’ and ‘mental health problems’ is to publicised on subsequent episodes, related television broadcasts
explore their distinct causal determinants. We report here on an and made available on BBC web pages and social networking sites.
analysis of a unique database of 27 397 participants, using the Visitors to the Stress Test’s homepage accessed the test by signing
novel but powerful techniques of conditional independence in using BBC online membership. The test was ‘live’ for a 4-week
graphical analysis,3,4 testing whether the same patterns of these period. The test had 12 sections, which took approximately 20 min
causal factors and psychological mediators are implicated in the to complete in total. It was designed by authors P.K. and S.T. and
development of mental health problems and mental well-being. developed in collaboration with BBC Lab UK. Questionnaire
Although conditional independence mapping has been used to items were completed in a fixed order and answers were selected
analyse a range of data in both social and physical sciences, this from a drop-down menu. Some tasks were constrained within
is the first time that the technique has been used on multivariate time limits. On completion, an overview of scores was displayed
data such as these in such a substantial data-set. Conditional on a results home page and URL links for comprehensive and
independence mapping has similarities with covariance or tailored feedback based on test scores were presented. Once
structural equation modelling, but makes fewer a priori participants had completed the test they were not permitted to
assumptions about the expected relationships, allowing all retake it.
possible permutations of association to be tested simultaneously
in a single multivariate model. Identified significant associations Measures
can then be explored in more detail to explore the contingent Measures were selected on the basis of theoretical principles and
relationships between selected variables. empirical research to provide indicators of latent constructs
representative of the components of the biopsychosocial model
Method of mental health and well-being.6 Where measures were modified
or created for the purposes of this study, other analyses have
Participants were self-selected respondents to an online battery of demonstrated their post hoc construct validity.7 Demographic data
questionnaires hosted on the British Broadcasting Corporation collected included age, gender, ethnic group, occupation, gross
(BBC) website (the Stress Test). In total 32 827 respondents (age annual or weekly household earnings, highest level of formal
18–85 years) completed the test battery; their mean age was 40.5 schooling, occupational status, parents’ income, relationship

456
 Anxiety, depression and well-being

status and number of children. Measured variables to represent remaining associations are then directed according to the relative
the biological component of the theoretical model were a yes/no strength of conditional probabilities, and subjected also to
response to indicate participants’ reports of familial mental health additional constraints so that the final map is consistent and
diagnoses by a psychiatrist or general practitioner,8 and acyclic, so removing any closed loops. Specific associations
performance on two cognitive tests to detect response to negative between variables may be investigated further using covariance
feedback and negative and positive stimuli. These were the tables. Missing data were deleted listwise to produce complete data
‘delayed match to sample’ (DMS) and the ‘affective go/no go’ for 27 397 participants (retention of 83.5% of the original
(AGN) tasks adapted from the Cambridge Neuropsychological sample, n = 32 827). The neurocognitive data accounted for a large
Test Automated Battery (CANTAB).9–11 These tests were chosen proportion of missing data owing to invalid recording, probably
to measure neurological functioning previously associated with because of technical error. Missing case analysis revealed no
anxiety and depression,9,10 and therefore supplement the more significant difference between those with and without missing data
subjective participant report of familial history of mental health on demographic variables and a selection of measured variables.
problems. The social component of the model was represented The software used in this study, CiMAp version 1.0,3 imple-
by a novel 11-item questionnaire designed to indicate social mented within the MatLab (R2011a) 64-bit (win64) statistics
relationships with friends and family, and participation in social package, employs a computationally efficient algorithm,28 with
activities.12 These items were a combination of Likert scale and adjustments for repeated tests. It starts from a fully connected
yes/no responses. Indicators of the circumstantial component undirected map and deletes conditionally independent associations
included recent and historical life events. Recent life events were in multiple statistical tests. Conditional mutual information is
measured using the List of Threatening Experiences Questionnaire computed using the equation
(LTE-Q).13 Participants also indicated on a five-point Likert scale
whether they believed they had historically been physically, X pijA ðxi ; xj jxA Þ
sexually or emotionally abused, or bullied at school.14 The two ~I ði; jjAÞ ¼ pijA ðxi ; xj ; xA Þlog2
xi ;xj ;xA
pijA ðxi jxA ÞpjjA ðxj jxA Þ
key psychological processes (response style and attributional style)
were measured using an adapted Response Style Questionnaire
where xi, xj, xA represent values taken from variables Xi, Xj and XA,
(RSQ).15 In the modified version for the Stress Test, participants
respectively, and p is the joint mass function that is typically
indicated on a Likert scale their response to stressful situations
approximated with the joint estimate nijA/N.
from a list of coping strategies pertaining to rumination,
Larger values of Ĩ imply stronger dependencies, and significance
problem-solving/adaptive or dangerous activities. Rumination
tests assumed a w2 distribution for independent or conditional
was characterised by endorsement of statements such as ‘I think
independent variables.20 Further details of this equation and its
about my shortcomings, failings, faults and mistakes’; adaptive
use in conditional independence mapping can be found in more
coping was characterised by endorsement of statements such as
specialist publications.20
‘I make a plan to overcome a problem’; and dangerous activities
Conditional independence graphical analysis conducts
encompassed strategies such as ‘go shopping with no regard for
simultaneous tests of association on categorical data. Variables
the debts I may run up’ or ‘drink alcohol excessively’. Attributional
with more than 15 categories (attribution style, dangerous
style was measured using an adapted and modified version of the
activities response style, rumination response style, well-being
Internal, Personal and Situational Attributions Questionnaire
score and GAD score) were transformed into deciles. Frequency
(IPSAQ),16 to determine the degree to which individuals generate
of childhood abuse was transformed into six categories, owing
internal, personal or situational causes for an event or situation.
to the distribution of responses. Family history of mental health
Finally, mental health problems were assessed using the well-
problems was transformed to two simple ‘yes/no’ categories.
validated Goldberg Anxiety and Depression (GAD) scales,17 a
Other variables (highest level of formal schooling achieved,
paired nine-item self-report symptom inventory designed to assess
occupational status, total gross annual or weekly household
anxiety and depression. The GAD scale was chosen for its brevity
income, parents’ income while growing up, relationship status,
and because it has been found to be a valid and acceptable method
number of children, relationships with friends, relationships with
of detecting psychological disturbance in the general population.17
family, social activities, frequency of social contacts, total number
Well-being was measured using the BBC Subjective Well-Being
of life events, adaptive response style and dangerous activities
(BBC-SWB) scale,18 a 24-item questionnaire designed to measure
response style) were used in their raw form, with each integer
people’s subjective experiences across the wide breadth of domains
treated as a separate category. In the case of total gross annual
commonly included in the definitions of well-being: psychological
income, 12% of participants recorded either ‘don’t know’ or
well-being, physical health and satisfaction with relationships. The
declined to answer the question. As these responses are meaningful
BBC-SWB scale has been found to be a reliable and valid measure
(and were associated with occupational status as students or
for the online assessment of well-being in the general population
unemployed), these categories were included in the analysis rather
with good psychometric properties.19
than excluded as missing. A summary of the data processing
applied to each variable is given as an online supplement (Table
Statistical analysis DS1).
The data-set was subjected to conditional independence graphical Results
analysis,3,4 to test the conditional independence between paired
variables. Since mutual covariance is common in multivariate The majority of respondents held an undergraduate, professional
analyses – and especially common in the social sciences – the qualification or postgraduate degree (73%), were in full- or part-
algorithm used in this analysis ensures that each specific test for time employment (73%), were in a relationship (74%) and had no
significance of association is conditional on all associations with children (54%). Compared with national data for England and
other variables connected in the model. The resulting multivariate Wales,5 more UK respondents were White, had slightly higher
association structure is represented in a conditional independence earnings and were better educated than the general population,
map. Any pair of variables that do not carry significant although comparable on other demographic features. Demo-
information about each other are disconnected in the map. The graphic data for the full sample are presented in Table 1.

457
Kinderman et al

Table 1 Demographic details of whole sample and sample and mental health problems are unrelated, merely that conditional
after removal of missing data mutual information concerning their association is best represented
Missing data
through a pattern of indirect associations as presented in Fig. 1
Whole sample removed and involving several different pathways. The analysis does,
(n = 32 827) (n = 27 397) however, strongly suggest that the two concepts are unlikely to
% (n) % (n) be two poles of a single continuum. These associations were
Ethnic group
explored in greater detail using simple contingency tables based
White (British, Irish, other) 85.6 (28 105) 92.8 (25 434) on median splits. Table 2 presents the data from the BBC-SWS
Black minority ethnic 5.6 (1850) 5.9 (1612) and the GAD, with a clear inverse relationship between the two
Rather not say or missing 8.7 (2872) 1.3 (351) variables.
Highest level of schooling achieved
Did not complete schooling 2.3 (747) 2.2 (601)
In education until age 18 years 24.4 (8032) 24.7 (6766)
Different pathways for mental health problems
Degree or professional qualification 73.2 (24 048) 73.1 (20 030) and well-being
Occupational status Contingency tables also permit more detailed exploration of
In education 11.8 (3885) 11.4 (3109) three-way associations revealed by the conditional independence
In employment 73.1 (23 991) 73.7 (20 195) map. These analyses reveal clearly differentiated mediated pathways
Other 15.1 (4951) 14.9 (4093)
leading from causal factors to well-being and mental health
Total gross annual household income
problems. The contingent relationship between life events,
Up to £39 999 ($65 000) per annum 51.6 (16 936) 51.9 (14 206)
Above £39 999 ($65 000) per annum 34.2 (11 225) 36.0 (9851)
rumination and mental health problems is presented in Table 3,
Don’t know/prefer not to say 14.2 (4666) 12.2 (3340) with simple binomial categorisation to permit straightforward
Estimated parents’ income while growing up interpretation. For the whole sample the relationship between life
Lower than 50% population 51.2 (16 824) 50.8 (13 913) events and mental health problems was relatively weak, with only
Higher than 50% population 48.7 (16 003) 49.2 (13 484) a slightly disproportionate number of people falling into the cell
Relationship status representing both higher levels of mental health problems and
In a relationship 74.3 (24 389) 73.2 (20 062) more life events. The mediating effects of rumination are clear,
Single 25.7 (8438) 26.8 (7335) however, in the two succeeding parts of Table 3. For participants
Number of children reporting low levels of rumination, prevalence of mental health
None 54.5 (17 900) 53.7 (14 717) problems was low, even in the presence of more frequent life
One or more 45.6 (14 927) 46.3 (12 680)
events (17%). In contrast, for participants reporting high levels
of rumination, more frequent life events were associated with a
The results of the analysis are summarised in the conditional markedly higher rate of mental health problems (56%).
independence map (Fig. 1), which shows variables that are Similar contingency tables explored the three-way associations
significantly associated (P50.05), having tested for all possible between well-being, social activities and adaptive coping (Table 4).
associations. Each connecting line in Fig. 1 therefore represents For the whole sample the relationship between social activities and
the associated pairwise mutual information between the respective well-being was clear, with higher levels of social activities
variables. Higher mutual information corresponds to a stronger associated with higher levels of subjective well-being. Again,
pairwise association, so for this map the strongest pairwise however, adaptive coping acted as a mediating protective factor,
association is 0.35 and the weakest 0.01, but all are significant at as high levels of adaptive coping were associated with
the 95% confidence level. Inspecting the map in more detail disproportionately fewer people reporting low levels of subjective
reveals that the variable ‘mental health problems’ was directly well-being in the absence of social activity.
and significantly associated with two other variables: ‘rumination’
(0.35) and ‘number of negative life events’ (0.05). This means that, Discussion
given all the possible associations of ‘mental health problems’ with
every other variable in the data, it was only these two variables Our findings provide clear quantitative support for the hypothesis
that conveyed statistically significant conditional mutual that subjective well-being and mental health problems have
information. In contrast, ‘well-being’ was directly and significantly distinct causal pathways,1,2 with different causal factors and
associated with eight other variables: adaptive coping (0.1), psychological mediators, despite the high correlation between
frequency of social contacts (0.05), social activities (0.19), these concepts. These findings are strongly supportive of previous
relationships with friends (0.11), relationships with family reports that a family history of mental health difficulties, social
(0.05), parental income (0.02), household income (0.03) and deprivation and traumatic or abusive life experiences are
highest level of education (0.02). Surprisingly, well-being and associated with higher levels of anxiety and depression, mediated
mental health had no directly associated variables in common. by adaptive coping and rumination.7 The mediating role of
All the variables included in the analysis (listed in online Table psychological processes was strongly evidenced in the relationship
DS1) were represented on the map, with the exception of between life events and mental health problems, in that there was
‘attribution style’, which had no significant association with any a striking distinction between the rates of mental health problems
other variable. following frequent life events for people with low, as opposed to
Examination of the map of conditional associations reveals a high, levels of rumination. Low levels of subjective well-being were
cluster of social variables associated with well-being, accompanied associated with social isolation and low levels of adaptive coping.
by adaptive coping, previously identified as a mediating variable.7 On the other hand, mental health problems were associated with
In contrast, a cluster of variables including negative life events, negative life events and rumination.
abuse and a familial history of mental health problems was Because this study was conducted on a cross-sectional rather
associated with mental health problems (anxiety and depression), than a longitudinal basis, we were able to identify patterns of
accompanied by rumination, again previously identified as a association rather than direct causal relationships. For instance,
mediating variable.7 This analysis does not suggest that well-being rumination is often considered a symptom of depression rather

458
 Anxiety, depression and well-being

0.04 0.35
Frequency of Relationships Attributional
social contacts Social activities with friends style
0.05 0.19 0.11

Relationships 0.05
with family
Well-being
Occupational 0.12 Household
status income 0.03

0.04 0.12

0.01 0.10
Number of Highest level of
0.02
children education

0.35
Adaptive coping
Parental 0.02
Relationship
status income 0.03

Rumination

0.02 0.02 Familial mental 0.01 0.03 0.35

health history
Dangerous
0.03 activities

Severity of Number of life Mental health

childhood trauma 0.05 events problems
0.05

Fig. 1 Conditional independence map. Multivariate associations were calculated using mutual information, Ĩ, shown pairwise in mbits,
and mapped taking account of all orders of conditioning (P50.05). This removes pairwise associations that can be explained by other
variables.

Table 2 Contingent relationship between well-being and Table 3 Contingent relationship between mental health
mental health problems, based on median splits of each problems and life events, stratified by degree of rumination,
variable based on median splits of each variable
Low well-being High well-being Few life events Many life events
n (%) n (%) n (%) n (%)

All participants (n = 27 397) All participants (n = 27 397)

Mental health problems Mental health problems
Low 4563 (16.7) 10 956 (40.0) Low 4898 (17.9) 8546 (31.2)
High 9421 (34.4) 2457 (9.0) High 3252 (11.9) 10 701 (39.1)
Low levels of rumination (n = 11 834)
Mental health problems
Low 3548 (30.0) 5507 (46.5)
than a cause or a mediator. Our findings require further
High 755 (6.4) 2024 (17.1)
confirmation in a longitudinal, predictive, design. Nevertheless,
High levels of rumination (n = 15 563)
the conditional independence maps identified pathways that are Mental health problems
consistent with current understanding of causal mechanisms.7,21 Low 1350 (8.7) 3039 (19.5)
The external validity of the findings may also be subject to self- High 2497 (16.0) 8677 (55.8)
selection biases given the online nature of the survey. Responses
to questions about mental distress and well-being may be affected
by gender, education, socioeconomic status or a personal history levels of well-being in the presence of mental health problems
of mental health problems. The UK sample in this study was less and, conversely, relatively few people with low levels of well-being
ethnically diverse, had slightly higher earnings and was better were free from mental health problems, these two concepts were
educated than the general population in the UK, although other clearly not opposing poles on a single continuum, because they were
demographic features were comparable. associated with different causal processes. In one sense, this finding
follows from the nature of the phenomena under examination; our
measure of well-being incorporated physical, social and purely
The two-continua model psychological well-being, whereas the assessment of anxiety and
These findings are theoretically and clinically significant, and depression was more specifically psychological. It is perhaps therefore
validate the relatively novel analytical technique of conditional unsurprising that the two phenomena had different associations.
independence graphical analysis.3,4 They offer support in a novel However, although this makes the interpretation of the findings
analysis of multivariate data from a uniquely large data-set for readily comprehensible, it does not reduce their potential significance.
the ‘two continua’ model of mental health and well-being,2 as well These findings have practical as well as theoretical implications.
as support for the hypothesis that psychological processes mediate They can (at least partially) help explain why people with good
the impact of biological, social and circumstantial factors on objective well-being, who are financially secure and in good
mental health.7 Although relatively few people experienced high relationships, can become depressed. These findings imply that

459
Kinderman et al

Table 4 Contingent relationship between well-being and Peter Kinderman, PhD, Institute of Psychology, Health and Society, University
social activity, stratified by levels of adaptive coping, based of Liverpool; Sara Tai, ClinPsyD, School of Psychological Science, Manchester;
on median splits of each variable Eleanor Pontin, DClinPsych, Institute of Psychology, Health and Society, University
of Liverpool; Matthias Schwannauer, PhD, School of Health in Social Science,
Low level of High level of Edinburgh; Ian Jarman, PhD, Paulo Lisboa, PhD, School of Computing and
Mathematical Sciences, Liverpool John Moores University, Liverpool, UK
social activity social activity
n (%) n (%) Correspondence: Professor Peter Kinderman, Institute of Psychology, Health
and Society, University of Liverpool, Waterhouse Building, Liverpool L69 3GL, UK.
All participants (n = 27 397) Email: [email protected]
Well-being
First received 24 Feb 2014, final revision 30 Sep 2014, accepted 6 Oct 2014
Low 8995 (32.8) 4434 (16.2)
High 4259 (15.5) 9709 (35.4)
Low level of adaptive coping (n = 10 914)
Well-being Acknowledgements
Low 4875 (44.7) 1930 (17.7)
High 1396 (12.8) 2713 (24.9) We thank the staff of BBC Lab UK for their support for the development and conduct of the
study, and J. Rees for critical discussions and reading of the manuscript.
High level of adaptive coping (n = 16 483)
Well-being
Low 4120 (25.0) 2504 (15.2) References
High 2863 (17.4) 6996 (42.4)
1 Keyes CLM. Mental health and/or mental illness? Investigating axioms of the
complete state model of health. J Consult Clin Psychol 2005; 73: 539–48.

if bad things happen to them (they experience negative life 2 Westerhof GJ, Keyes CLM. Mental illness and mental health: the two continua
model across the lifespan. J Adult Dev 2010; 17: 110–9.
events), and if they ruminate about those events, they are much
3 Bacciu D, Etchells TA, Lisboa PJG, Whittaker J. Efficient identification of
more likely to become anxious or depressed. Clearly, this analysis independence networks using mutual information. Comput Stat 2013; 28:
can be extended to a wide range of other causes and mediators not 621–46.
included in this study. These findings can also partially explain 4 Spirtes P, Glymour C, Scheines R. Causation, Prediction and Search (2nd edn).
how people who do not show signs of specific mental health MIT Press, 2000.
problems such as anxiety and depression can have low levels of 5 Office for National Statistics. Labour Force Survey: Household Datasets,
well-being or fail to flourish. If socially isolated and relatively Spring Quarter. ONS, 2003.

lacking in adaptive coping responses (such as problem-solving 6 Kinderman P. A psychological model of mental disorder. Harv Rev Psychiatry
2005; 13: 206–17.
and emotional self-regulation), people are more likely to experience
7 Kinderman P, Schwannauer M, Pontin E, Tai S. Psychological processes
low levels of well-being, but are unlikely to experience mental health mediate the impact of familial risk, social circumstances and life events on
problems if otherwise not exposed to the causal mechanisms mental health. PLoS One 2013; 8: e76564.
described above. 8 Milne BJ, Caspi A, Crump R, Poulton R, Rutter M, Sears MR, et al. The validity
of the family history screen for assessing family history of mental disorders.
Am J Med Genet B 2008; 50: 41–9.

Implications of the study 9 Elliott R, Sahakian BJ, Herrod JJ, Robbins TW, Paykel ES. Abnormal response
to negative feedback in unipolar depression: evidence for a diagnosis specific
Practically, these findings suggest that interventions aimed at impairment. J Neurol Neurosurg Psychiatry 1977; 63: 74–82.
maximising well-being and interventions aimed at preventing or 10 Murphy FC, Sahakian BJ, Rubinsztein JS, Michael A, Rogers RD, Robbins TW,
treating mental health problems should be complementary but et al. Emotional bias and inhibitory control processes in mania and
depression. Psychol Med 1999; 29: 1307–21.
different. We found that people who ruminated were three times
11 Chamberlain SR, Sahakian BJ. The neuropsychology of mood disorders.
more likely to develop anxiety or depression following negative life Curr Psychiatry Rep 2006; 8: 458–63.
events, and people who lacked adaptive coping skills were three
12 National Development Team for Social Inclusion. Reviewing Measures
times more likely to have low subjective well-being in the face of of Social Exclusion. National Social Inclusion Programme, 2003.
social deprivation, but that these two pathways were independent 13 Brugha TS, Cragg D. The list of threatening experiences: the reliability and
of each other. These findings have important implications for validity of a brief life events questionnaire. Acta Psychiatr Scand 1990; 82:
current policies in health and social care. Specifically, they strongly 77–81.

suggest that conceptually distinct interventions are needed; 14 Bernstein DP, Fink L. Initial reliability and validity of a new retrospective
measure of child abuse and neglect. Am J Psychiatry 1994; 151: 1132–6.
services aimed at treatment and prevention of mental health
15 Nolen-Hoeksema S, Wisco BE, Lyubomirsky S. Rethinking rumination. Perspec
problems may be very different from those needed to promote
Psychol Sci 2008; 3: 400–24.
and improve subjective well-being. All such clinical decisions must
16 Kinderman P, Bentall RP. A new measure of causal locus: the internal,
be dictated by individual case formulations, but it seems personal and situational attributions questionnaire. Pers Individ Dif 1996;
reasonable to suggest that interventions and social policy aimed 20: 261–4.
at improving social circumstances, addressing financial equity 17 Goldberg D, Bridges K, Duncan P, Grayson D. Detecting anxiety and
and social justice, building local communities, supporting depression in general medical settings. BMJ 1989; 297: 897–9.
families, and facilitating relationships on the societal level, and 18 Kinderman P, Schwannauer M, Pontin E, Tai S. The development and
validation of a general measure of well-being: the BBC well-being scale.
developing individuals’ personal capacity to cope adaptively with
Qual Life Res 2011; 20: 1035–42.
stressful life circumstances may be best suited to developing
19 Pontin E, Schwannauer M, Tai S, Kinderman P. A UK validation of a general
subjective well-being. Local authority social services may have measure of subjective well-being: the modified BBC subjective well-being
an immediate responsibility here, but genuine progress may scale (BBC-SWB). Health Qual Life Outcomes 2013; 11: 150–9.
necessitate wider societal and political action. Interventions aimed 20 Goebel B, Dawy Z, Hagenauer J, Mueller JC. An approximation to the
at preventing abuse, mitigating the consequences of traumatic distribution of finite sample size mutual information estimates. IEEE Intl Conf
Comm South Korea 2005; 2; 1102–6.
personal life events, and evidence-based interventions targeting
21 Weich S, Brugha T, King M, McManus S, Bebbington P, Jenkins R, et al.
psychological processes such as rumination (including the now
Mental well-being and mental illness: findings from the Adult Psychiatric
ubiquitous cognitive–behavioural therapy) may be best suited to Morbidity Survey for England 2007. Br J Psychiatry 2011; 199: 23–8.
helping people recover from mental health problems.

460