O C C U P A T IO N A L H E A L T H A N D W E L L - B E I N G

A longitudinal study of stress and psychological distress in nurses and

nursing students
Roger Watson, Eric Gardiner, Richard Hogston, Helen Gibson, Anne Stimpson, Robert Wrate and
Ian Deary

Aim. The aim of this study was to investigate how differences in life events and stress contribute to psychological distress in
nurses and nursing students.
Background. Stress is an issue for nursing students and qualified nurses leading to psychological distress and attrition.
Design. A longitudinal study using four time waves was conducted between 19941997.
Methods. Measures were taken of stress, life events and psychological distress in addition to a range of demographic data. Data
were analysed using descriptive statistics, linear modelling and mixed-effects modelling. The study was set in Scotland, UK and
used newly qualified nurses and nursing students from four university departments of nursing over four years. The study was
initiated with 359 participants (147 nurses and 212 nursing students) and complete data were obtained for 192 participants.
Results. Stress levels, psychological distress and life events are all associated within time and across time. At baseline, life events
and stress contributed significantly to psychological distress. The pattern of psychological distress differed between the nursing
students and the newly qualified nurses with a high level in the nurses after qualifying and starting their career.
Conclusion. Stress, individual traits, adverse life events and psychological distress are all interrelated. Future lines of enquiry
should focus on the transition between being a nursing student and becoming a nurse.
Relevance to clinical practice. Stress and psychological distress may have negative outcomes for the retention of nursing students
in programmes of study and newly qualified nurses in the nursing workforce.
Key words: nurses, nursing students, psychological upset, psychology, quantitative approaches, stress
Accepted for publication: 20 June 2008

Introduction
Stress in nurses, studied extensively in recent decades, is an
international phenomenon (Chang et al. 2005). There has
also been an increasing amount of research on stress in
student nurses (Jones & Johnson 1997, 1999, 2000). Nursing
is an occupation with a constellation of circumstances leading
to stress. The relevant literature was reviewed recently by
Chang et al. (2005) and the factors leading to stress include a
Authors: Roger Watson, PhD, RN, FAAN, School of Nursing and
Midwifery, The University of Sheffield, Sheffield, UK; Eric Gardiner,
BSc, MSc, PhD, Hull York Medical School, The University of Hull,
Hull, UK; Richard Hogston, BA, MSc, RGN, PGDipEd, Faculty of
Health, Leeds Metropolitan University, Leeds, UK; Helen Gibson,
BA, MA, Faculty of Health and Social Care, University of Hull, Hull,
UK; Anne Stimpson, BA, RGN, Faculty of Health and Social Care,
University of Hull, Hull, UK; Robert Wrate, MBBS, DPM,

270

highly demanding job with poor support, rapidly changing

circumstances, shortage of resources and staff, and dealing
with death and dying. These factors are intrinsic to nursing
and are compounded by environmental factors such as
difficult patients and their families, relationships with physicians, low institutional commitment to nursing and the
delivery of poor quality care (Chang et al. 2005). In addition,
the need to undertake continuing professional development
by nurses (Hogston 1995) and part-time education (Timmins
FRCPsych, Edinburgh Young Peoples Unit, Royal Edinburgh
Hospital, Edinburgh, UK; Ian Deary, BSc, PhD, MB, ChB, FRCPE,
FRCPsych, FRSE, FBA, FMedSci, Department of Psychology,
University of Edinburgh, Edinburgh, UK
Correspondence: Ian Deary, Department of Psychology, University of
Edinburgh, Edinburgh EH8 9JZ, UK. Telephone: +44 131 6503452.
E-mail: ian.deary@ed.ac.uk

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278
doi: 10.1111/j.1365-2702.2008.02555.x

Occupational health and well-being

& Nicholl 2005) can also be stressful. Nursing is an

emotionally demanding job and this contributes to interactive
stress and to the daily stress of nurses work (Mann &
Cowburn 2005). In addition, stress leads to more emotion
focused coping which is not constructive (Deary et al.
1996) and causes nurses to seek social support (LeSergent
& Hanley 2005). Stress is a normal reaction to a range of
circumstances (Smith & Fawcett 2006) but can have negative
consequences, one of which is emotional exhaustion; it has
been demonstrated that factors leading to stress, such as
increasing patient load, also lead to emotional exhaustion
(Maslach & Jackson 1986).
Stress among nursing students includes some of the above
aspects, as students experience clinical practice but also
includes stress from sources such as separation from home,
financial worries, regular clinical and educational assessment
and frequently changing clinical environments (Deary et al.
2003). Deary et al. (2003) showed that nursing students do
experience increasing levels of stress and psychological
distress throughout their nursing programme. Newly qualified nurses, entering the clinical areas for the first time, seem
to experience a high level of stress associated with lack of
confidence and exposure to unfamiliar circumstances (Chang
& Hancock 2003, Chang et al. 2005).

The study
The aim of the present study was to examine how a range of
demographic, personal, circumstantial/environmental factors
and stress contribute to differences in psychological distress
in newly qualified nurses and nursing students.
Stress is a process. Studies which examine stress in a crosssectional manner, with a limited number of relevant variables, only reveal part of the process, and at worse, simply
report correlations among scales with overlapping content
(Deary et al. 1996). In addition to the inclusion of the
appropriate variables, studies capture the process of stress
better if they are longitudinal because the outcomes of the
stress process can then be assessed. The transactional model
of work-related stress (Deary et al. 1996) recognises that
there are various factors that act as antecedents to stress.
There are personal factors such as demographics, personality
traits and environmental life events. These can have an
influence on the amount of stress reported at work, which
should also be measured. But this work-related stress is not
the outcome of the process. In fact, there might be more
than one outcome, such as psychological distress. Following
this process model, the present study includes most of the
above-mentioned elements, and is longitudinal. The antecedent variables collected included demographic and social

Stress in nurses and nursing students

circumstance data, as well as environmental life events.

Work-related stress was recorded. Psychological distress was
the outcome. The life events and psychological distress were
collected on four separate occasions, over a period of four
years. Thus, the present study is a prospective, longitudinal
study of nurses and nursing students in Scotland carried out
between 19941997. The combination of stress-process
model variables and the multiple longitudinal data collections
make it one of the most comprehensive studies of nursing
stress undertaken to date. In addition, maximum use was
made of the rich data by employing mixed-effects modelling.
It is acknowledged by the authors that considerable time
has elapsed between the collection of data for this study and
the present analysis. This is explained, in part, by the death of
the original principal investigator and the fact that the
analysis applied here, making use of multi-level modelling,
was not widely used at the time of the original study. The
original research questions for the study and the analysis
applied were not concerned with longitudinal analysis and
the funders of the study, The Chief Scientist Office of The
Scottish Office, approached The University of Edinburgh to
undertake a retrospective study using more sophisticated
statistical methods. The context of the study was a period of
considerable stability in nursing education in Scotland
following the introduction of Project 2000 in the late
1980s, whereby the whole of nursing education moved into
the universities in the early 1990s.

The research questions

The research questions addressed in this study were:
Is there a relationship between a range of demographic and
circumstantial factors and stress and psychological distress
in nurses and nursing students?
What are the individual characteristics of nurses and
nursing students and the circumstantial factors leading to
stress and psychological distress?

Design
The design was a longitudinal cohort study with an existing
dataset, which came from a three-year longitudinal study by
Working Minds (see Acknowledgements), with four waves of
data collection (Baldwin 1999). The subjects were newly
qualified nurses and nursing students in Scotland. Participants entering the study as students were recruited in the
classroom at the beginning of their programme of study and
participants entering the study as newly qualified nurses were
recruited in the classroom at the end of their programme of
study. Written consent was obtained from the students to be

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278

271

R Watson et al.

interviewed and to complete the questionnaires for the study.

This study took place prior to the requirement for formal
institutional ethical approval to involve students; permission
was obtained from their universities to meet them and
introduce the study.

Sample
Initially, 220 nurses were approached and 67% were
recruited (n = 147); 285 students were approached and
74% (n = 212) were recruited. The study was, therefore,
initiated with 359 participants and there were complete data
for 192 (depending upon which outcome is studied). Eighteen
nurses were men (129 women) and 38 students were men
(174 women). The median age of the nurses was 26 (range
2052) and of the students 22 (range 1744).

Questionnaires
Demographic and circumstantial factors Information was
available on sex, age, marital status, having a close friend or
not (at each of the four waves), status (staff or student) and
having a dependent child or not (at each of the four waves).
The brief life events inventory measures stressful life events
in the last year. It is an eight item questionnaire which asks
respondents to answer yes or no to a series of life events
such as financial problems or job applications. It then asks the
subjects to respond, using a five point Likert scale, to how
much this has affected them with possible responses ranging
from Not at all stressful (0)Extremely stressful (4) through
Slightly stressful, Moderately stressful and Very stressful.
The total score was used here. It was administered four times,
in all four waves: in 1994, 1995, 1996 and 1997. The internal
consistency of the brief life events inventory measured at first
use in the study using Cronbachs alpha was 053.
The work-stress inventory (Firth-Cozens 1992) used in the
Working Minds project was a 15-item work-related stress
questionnaire which enquired about both frequency and level
of stress with items such as Difficult relations with senior
nurses and Making decisions. Frequency of stressors was
responded to on a three point Likert scale from Never (0)
through Occasionally (1)Frequently (2). The level of
stress for each stressor was responded to on a five point Likert
scale from Not at all stressful (0)Extremely stressful (4)
through Slightly stressful, Moderately stressful and Very
stressful. Total stress frequency and level were used here. It
was administered once, at wave 4 in 1997. The internal
consistency of the work-stress inventory frequency scale was
068 and for the level of stress was 076 measured using
Cronbachs alpha.
272

The General Health Questionnaire-28 (GHQ: Goldberg &

Williams 1988) was used to measure current psychological
distress. It has 28 questions, each of which has a four point
response scale, which are scored from zero (least distress)3
(most distress). It asks questions about somatic symptoms,
anxiety, insomnia, social dysfunction and depression. Higher
scores on items measuring these aspects indicate greater
levels of psychological distress. The total score was used
here. It was administered four times, in all four waves: in
1994, 1995, 1996 and 1997. The internal consistency of the
GHQ measured at first use in the study using Cronbachs
alpha was 087.

Analysis
Descriptive statistics and linear modelling
Distributions (means and standard deviations) are described
for the principal variables. Malefemale and staffstudent
differences are examined (unpaired t-tests). Differences over
the four waves of testing are examined (paired t-tests).
Pearsons correlations (pairwise and listwise) are examined
between variables and within variables across time waves.
The GHQ scores are modelled as a function of life
events, sex, stress and other demographic variables (general
linear modelling: ANCOVA ). These analyses were run on SPSS
version 14.
Mixed-effects modelling
The principal novel analysis used mixed-effects models: statistical models with both fixed and random effects. Because
the data were longitudinal, they allow an understanding of
both the mean levels of stress and the changes in stress over
time in these groups of nurses and student nurses. The GHQ
and the life events scales were administered on four occasions. There are missing data, with not every subject
continuing until the end of the study. We were also able to
use, in the mixed-effects modelling, information on whether
or not the person was present at each wave. Therefore,
dropout from each wave was used as information, actually
part of the statistical modelling. Demographic and circumstantial/environmental factors may be modelled using fixed
effects. Further, individual variation may be modelled using
random effects, in particular, random individual intercepts
and random individual slopes over time. Life events can also
be introduced as a random effect because they were measured
repeatedly over time. These analyses were conducted using
R 220 for Windows (R Development Core Team 2005).
Because mixed-effects modelling is still used and understood by relatively few health researchers, and because it is
especially well suited to the data we are examining here, a

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278

Occupational health and well-being

Stress in nurses and nursing students

short explanation of the method is provided. Mixed-effects

modelling is widely used in other disciplines such as medicine,
education and sociology (http://www.soc.surrey.ac.uk/sru/
SRU23.html; retrieved 19 June 2006) and should be used
more in academic nursing. All statistical models contain at
least one random variable, reflecting the impossibility of a
complete explanation of values taken by outcome measures.
It is often useful, however, to distinguish between different
sources of random variation. In this context, two potential
sources are differences between individuals in their initial
outcome values and differences in their changes over time.
These are additional to the remaining random variation,
often labelled measurement error. The significance of such
sources can be tested using mixed-effects models. Mixedeffects models have significant theoretical advantages over
other older methods of analysing longitudinal data, such as
repeated measures analysis of variance. In particular: (i) all
available data can be analysed, rather than only data from
individuals present at all time points and (ii) considerable
mathematical flexibility is available for modelling the correlation of outcome measures at different time points within
individuals.

Results
Descriptive statistics
Mean (SD) values and numbers are shown for stress
frequency and level (1997), GHQ (from 19941997) and
life events (from 19941997) (Tables 1 and 2). These are
shown for all subjects and according to sex-specific and
status-specific (student or newly qualified nurses) subgroups.
Table 1 shows the pairwise values and Table 2 shows the

listwise values. Thus, Table 1 affords the most powerful

between-sex and between-status comparisons. Women scored
higher than men on stress level (p = 0044). There were no
other significant differences between men and women. Newly
qualified nurses scored higher than students for stress
frequency, stress level, GHQ 1994 and life events 1994
1996 (see footnote of Table 1 for p-values). Students scored
higher than newly qualified nurses for life events in 1997
(p < 0001). Table 2, with listwise data, affords a comparison of the same peoples GHQ scores across time. The main
result was that the final wave (1997) showed a significantly
lower GHQ scores than the other three waves.

Correlations
Pearsons correlations are shown between stress frequency
and level (1997), GHQ (from 19941997) and life events
(from 19941997) (Table 3). Because of the large numbers of
participants in this study, r-values equal to or greater than 02
are significant at p < 001. Listwise correlations are shown
above the diagonal and pairwise below, and they are very
similar. Self-esteem, tested in 1994, is associated significantly
with GHQ across all four waves (r-values from 045 to
021) but correlates progressively lower as the GHQ is more
distant in time.
General health questionnaire shows moderate stability
across the four waves with coefficients between 027048 (all
p < 0001). Life events show similar across-wave stability
with coefficients ranging from 021055 (all p < 0001). The
correlations between GHQ and life events were also examined and four of the GHQ-life events correlations are based
on GHQ and life events measures taken contemporaneously;
the mean of these is 031 and the range is 026039. Twelve

Table 1 Descriptive statistics (pairwise) for the variables used in the Working Minds study

Stress frequency*
Stress level*
GHQ 1994*
GHQ 1995
GHQ 1996
GHQ 1997
Life events 1994*
Life events 1995*
Life events 1996*
Life events 1997

All subjects

Male students

Female students

Male staff

Female staff

122
186
196
213
221
172
72
57
50
72

114
164
169
220
224
169
70
49
54
86

108
168
177
221
226
172
66
35
39
86

119
156
184
171
191
142
70
54
41
51

138
215
230
206
219
176
79
87
64
58

(38) 237
(80) 232
(101) 359
(118) 318
(131) 289
(108) 291
(49) 357
(54) 306
(43) 282
(48) 266

(45) 22
(84) 22
(96) 38
(130) 30
(146) 30
(97) 26
(47) 36
(55) 27
(45) 30
(57) 24

(33) 103
(70) 101
(85) 174
(109) 154
(122) 130
(109) 134
(46) 174
(39) 146
(38) 127
(49) 114

(29) 13
(57) 13
(97) 18
(85) 16
(118) 15
(66) 15
(33) 18
(41) 16
(38) 14
(38) 15

(36) 99
(85) 96
(114) 129
(129) 118
(140) 114
(115) 116
(54) 129
(58) 117
(44) 111
(43) 113

Numbers within each cell are mean (SD) n.

GHQ General Health Questionnaire-28.
*Staff > students (p < 0001 for all variables except life events 1994 where p = 0032), female > male (p = 0044). No other sex differences,

students > staff (p < 0001).

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278

273

R Watson et al.
Table 2 Descriptive statistics (listwise) for the variables used in the Working Minds study

Stress frequency
Stress level
GHQ 1994*
GHQ 1995*
GHQ 1996*
GHQ 1997*
Life events 1994
Life events 1995
Life events 1996
Life events 1997

All subjects
(n = 192)

Male
students
(n = 18)

Female
students
(n = 85)

Male
staff
(n = 12)

Female
staff
(n = 77)

122
188
190
192
207
167
72
56
48
73

113
161
179
228
213
161
66
45
54
92

108
167
164
194
215
166
66
33
37
85

120
158
158
157
177
128
70
58
30
49

138
221
223
187
202
176
80
83
61
60

(38)
(83)
(105)
(99)
(119)
(105)
(50)
(52)
(40)
(50)

(47)
(87)
(109)
(153)
(138)
(96)
(53)
(49)
(49)
(62)

(33)
(73)
(82)
(80)
(108)
(96)
(44)
(34)
(36)
(50)

(30)
(59)
(80)
(61)
(119)
(43)
(27)
(45)
(25)
(42)

(37)
(85)
(121)
(105)
(127)
(121)
(57)
(58)
(41)
(44)

Numbers within each cell are mean (SD) n.

GHQ General Health Questionnaire-28.
*For all subjects, pairwise comparisons of GHQ scores: 1996 > 1994 (p = 003); 1997 < 1994 (p = 0005); 1997 < 1995 (p = 0001);
1997 < 1996 (p < 0001).

Table 3 Correlations pairwise (below the diagonal) and listwise (above the diagonal; n = 192)

Stress frequency
Stress level
GHQ 1994
GHQ 1995
GHQ 1996
GHQ 1997
Life events 1994
Life events 1995
Life events 1996
Life events 1997

Stress
frequency

Stress
level

GHQ
1994

GHQ
1995

GHQ
1996

GHQ
1997

Life
events
1994

Life
events
1995

Life
events
1996

Life
events
1997

076 (231)
034 (237)
022 (227)
019 (222)
031 (237)
016* (237)
035 (219)
036 (217)
022 (232)

079

033
021
028
037
021
032
033
028

036
036

041
049
034
035
046
034
020

024
023
038

055
029
024
028
031
026

018*
030
048
044

041 (253)
035 (287)
026 (269)
034 (282)
030 (247)

034
039
035
027
039

018 (291)
025 (261)
016* (247)
032 (266)

014
021
033
017*
030
022

039 (304)
033 (280)
025 (266)

032
034
056
027
030
030
043

054 (263)
016* (245)

034
032
036
020
026
013
029
055

027 (242)

021
029
027
019*
030
039
028
021
029

(232)
(222)
(218)
(232)
(232)
(214)
(214)
(227)

(318)
(289)
(291)
(357)
(305)
(282)
(266)

(279)
(271)
(317)
(304)
(272)
(254)

GHQ General Health Questionnaire-28.

*001 < p < 005, 005 < p < 010. All other correlations are significant at p < 001.

of the GHQlife events correlations are based on measures

taken between one and three years apart and the mean of
these correlations is 027 with a range from 013056.
Therefore, there is no evidence for stronger correlations when
GHQ and life events are measured on the same occasion.

General linear modelling

General health questionnaire scores at each wave were
modelled using life events scores as covariables in a series
of general linear models, with sex as a fixed effect (Table 4).
We originally had the following variables in the models but
these were dropped as they were not significant contributors
to GHQ score variance: age, having a close friend or not,
274

status (newly qualified nurse or nursing student) and having a

dependent child or not. Sex had no significant effect on GHQ
score at any wave. Life events contributed significantly to
GHQ scores at all waves (all p < 0001), contributing
between 59% of the variance.
For GHQ measures in 1997, there was, in addition, a stress
level score. Therefore, in accordance with the process model
of stress, we examined contributors to reported stress levels
(Table 4). Contemporaneous life events and sex made
significant contributions (10, 7 and 3% of the variance,
respectively). Second, we modelled GHQ in 1997 as a
function of sex, self-esteem, life events (1997) and stress level.
The significant contributors were life events and stress level,
accounting for 7 and 8% of the variance, respectively.

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278

Occupational health and well-being

Table 4 General linear modelling of
outcome variables in the Working
Minds study

Stress in nurses and nursing students

Life events
GHQ
GHQ
GHQ
GHQ
GHQ
Stress

1994
1995
1996
1997
1997*
level

372,
154,
256,
265,
165,
155,

<0001,
<0001,
<0001,
<0001,
<0001,
<0001,

Sex
009
005
009
009
007
007

40,
01,
02,
09,
01,
60,

Stress level
0047, 001
072, 000
067, 000
033, 000
071, 000
0015, 003

180, <0001, 008

Within each cell numbers are F, p, g2.

GHQ General Health Questionnaire-28.
*Includes stress level as a covariable.

Mixed-effects analyses
Data were grouped into the following dropout time categories
for GHQ scores: Missing at all time points (n = 74; not used
in the analysis), Dropout after first time point (n = 19),
Dropout after second time point (n = 13), Dropout after
third time point (n = 34) and Completer or intermittent
responder (n = 293). Marital status was recoded into two
categories (Married/living with someone vs. Single/divorced/
separated/widowed).
First stage of analysis
The purpose of this stage of the analysis was to determine
what factors were associated with dropout from the study as
a necessary prelude to the next stage of the analysis. This
stage was carried out on SPSS 13 for Windows. Multinomial
regression of four dropout time categories (not assuming any
ordering) was used to determine whether status (newly
qualified nurses or nursing student), age, life event score,
marital status and sex were associated with missingness category. Only status was statistically significant (likelihood
ratio v2 = 10285, df = 3, p = 0016): newly qualified nurses
were more likely to be completers or intermittent responders
than nursing students. Ordinal regression, taking into
account the natural ordering in the dropout categories, gave
similar results; only status was statistically significant.
Second stage of analysis
The purpose of this stage of the analysis was to determine
what was associated with, or predicted, GHQ scores and
change in GHQ scores. This stage of the analysis was carried
out on R version 220. Linear mixed models were fitted
assuming missing at random data. In this missing at random
analysis, the following were examined:
1 A random slope for time of GHQ measurement (year,
treated as continuous) and a random slope for life events
(treated as continuous);
2 A variety of correlation structures for the repeated GHQ
scores;

3 Fixed (main) effects of status (newly qualified nurses

or nursing student), age, life event score, marital status and
sex; and interactions involving all these covariates (status age, etc.), treated as fixed effects, with time (year)
treated as continuous and fixed.
The random effects in (1) were not statistically significant.
This means that there were no significant differences in the
trajectories of GHQ and life events over the three years (four
waves) which may be due to the fact that the trajectories were
too short with a maximum of four time waves. Therefore, the
differences in individual GHQ scores cannot be explained as
being due to random differences from an underlying linear
change of GHQ score over time nor can they be due to
random differences from an underlying linear relationship
between GHQ score and life event scores.
For (2), we conjectured that the autoregressive covariance
(AR1) structure might be appropriate. In this structure, GHQ
scores at 1994 and 1995 are assumed to have the same
correlation as those at 1996 and 1997 and these correlations
are assumed to be weaker than those between GHQ scores at
1994 and 1997, for example. However, an unrestricted
correlation structure with different variances at each time
point fitted the data best, according to likelihood ratio tests.
Examination of the modelling output suggested that the fixed
effect estimates did not change much when different correlation structures were tried.
For (3), status time was the only statistically significant
interaction. This means that the pattern of GHQ changes
over the three years (four waves) was significantly different
for newly qualified staff and students. Parameter estimates,
standard errors, Walds t-values, p-values and 95% confidence intervals for the model containing this interaction and
all other main effects are presented in Table 5. From Table 5,
interpretation of the significant interaction is that newly
qualified nurses GHQ is higher to begin with than nursing
students GHQ but reduces over time, so that differences are
less marked in 1997 than in 1994. Additionally, life event
score and age are statistically significant; as each of these

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278

275

R Watson et al.

Parameter

Estimate

Intercept
Age
Life events
Married (married/living
with someone)*
Sex (female)*
Status (staff)
Time
Status (staff) time

SE

t-Value

p-Value

95% CI
(5543,
(0086,
(0172,
( 2048,

9560
0211
0294
0533

2047
0063
0062
0772

467
333
473
069

<0001
0001
<0001
049

1940
4036
0089
1234

1162
1263
0288
0426

167
320
031
290

0095
0001
076
0004

1358)
0335)
0415)
0981)

Table 5 Random effects modelling of

GHQ scores in the Working Minds
study

( 0340, 4221)
(1558, 6513)
( 0477, 0654)
( 2069, 0398)

Parameter estimates and standard errors assume missing at random data

GHQ General Health Questionnaire-28.
*This is compared with the reference category of the categorical covariate/factor. Thus, the
interpretation is that GHQ is predicted to be 194 points higher for females than for males (the
reference category) when other terms in the model are kept fixed although this particular
difference isnt statistically significant, the interpretation is that GHQ was predicted to be 404
points higher for staff than students in 1994 if all other variables are equal. However, this
difference decreases by 1234 points every year, so that in 1995 GHQ was predicted to be 2802
points higher for staff than students for example.

increases, GHQ score is predicted to increase. Other covariates, although not statistically significant, point in the
expected directions. The details of a validity check on the
data are appended (Appendix 1).

Discussion
Previous work by the present investigators (Deary et al. 2003)
and others has demonstrated that stress and psychological
distress are related in nurses and nursing students. This has
implications for work-related illness and attrition in both
groups. The present study, using longitudinal data and
advanced statistical analysis, has added significantly to this
knowledge base by demonstrating that, in addition to stress,
there is an effect on psychological distress of life events and
the trait of self-esteem. In addition, there is a combined effect
of age and life events on psychological distress. Measured
over the same time, from start of nursing education and start
of nursing career, this study has shown that psychological
distress is higher initially in newly qualified nurses but that
this falls to the same level as nursing students over four years.
Possible limitations of this study include the use of a nonstandard measure of stress making direct comparisons with
other studies difficult; the lack of a comprehensive measure of
personality traits at baseline which could have provided
insight into subsequent individual differences and the fact
that the transition from nursing student to newly qualified
nurses nurse was not followed through in individuals.
Despite limitations, in terms of the transactional model
of stress which was the theoretical underpinning for
the present study the longitudinal design with repeated
276

measures allowed us to study the influence of antecedent

factors such as demographics and life events on the development of stress and the relationship between stress and
psychological distress. A cross-sectional design at only one
time point would only have provided insight into correlations
without the possibility of establishing causal mechanisms.
The present study used mixed-effects models, which are, as
yet, uncommon in nursing research. They were especially
useful here in view of the longitudinal nature of the data.
They could make use of the missing data, which are treated as
information rather than inconvenience as they are in other
approaches. Although it was not found here, these models
afford the possibility of finding significant random slopes; this
means that they can detect whether people not only have
different mean values in variables such as GHQ, but whether
there are significant differences in the patterns of change over
time. The models can also find the determinants of these
different patterns of change over time thereby providing
further insight into the relationship between outcomes such
as stress and their determinants. Where someone begins in
terms of their demographics, psychological distress and
environmental factors may only provide partial insight into
their subsequent levels of stress. For example, their response
to life events and work-related factors over a given time may
also determine their ultimate outcomes.
There were missing data in this study and, while it may be
assumed that some of these individuals were lost to nursing,
either during education or in employment, no data on
attrition were available. The implications for retention of the
nursing workforce lie in the recognition that events outside
nursing lead to psychological distress and may also play a

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278

Occupational health and well-being

role, in addition to stress. The fact that psychological

distress is higher at the start of employment and then
decreases points to a critical period at the start of employment as a nurse where psychological distress may lead to
work-related illness and attrition. While steps may be taken
to reduce workplace stress (Ryan & Watson 2004), it is
impossible to control for life events; in addition, nurses have
to start their employment at some point and this initial
phase of inculturation against the backdrop of a steep
learning curve is probably unavoidable. Nevertheless, the
consequences for human resources are obvious and profound: newly qualified nurses and nursing students should
have staff support services available to them and be
encouraged to use these under circumstances where they
are more likely to experience psychological distress. Clinical
supervision is one strategy that has been shown to be
effective in reducing such psychological distress (Hyrkas
2005). Continuing professional development in relation to
the causes of psychological distress and the recognition of its
causes is required and services should be targeted at the
initial stages of nursing employment in first- and second-year
staff nursing.

Acknowledgements
Working Minds was established by Robert Wrate and the late
Pamela Baldwin who was Chief Investigator for the present
project which was funded by the Chief Scientist Office for
Scotland. Data were collected and initially entered onto SPSS
by Marjory Dodd. Funding from the Chief Scientist Office for
Scotland for this present study is acknowledged. Ian Deary is
the recipient of a Royal Society-Wolfson Research Merit
Award.

Contributions
Study design: RWr, ID, RWa; data collection and analysis:
RWr, AS, HG, RW, ID, EG and manuscript preparation:
RWa, ID, RH.

References
Baldwin PJ (1999) Nursing. In Stress in Health Professionals (FirthCozens J & Payne RL eds). John Wiley & Sons Ltd, London, pp.
93104.
Chang EM & Hancock K (2003) Role stress and role ambiguity in
new nursing graduates in Australia. Nursing Health Science 5,
155163.
Chang EM, Hancock KM, Johnson A, Daly J & Jackson D (2005)
Role stress in nurses: review of related factors and strategies for
moving forward. Nursing and Health Sciences 7, 5765.

Stress in nurses and nursing students

Deary IJ, Blenkin H, Agius RM, Endler NS, Zealley H & Wood RA
(1996) Models of job-related stress and personal achievement
among consultant doctors. British Journal of Psychology 87, 329.
Deary IJ, Watson R & Hogston R (2003) A longitudinal cohort study
of burnout and attrition in nursing students. Journal of Advanced
Nursing 43, 7181.
Firth-Cozens J (1992) The role of early experiences in the perception
of organisational stress: fusing clinical and organisational perspectives. Journal of Occupational and Organisational Psychology
65, 6165.
Fitzmaurice G, Laird N & Shneyer L (2001) An alternative parameterization of the general linear mixture model for longitudinal
data with non-ignorable dropouts. Statistics in Medicine 20,
10091021.
Goldberg D & Williams P (1988) A Users Guide to the General
Health Questionnaire. HFER-Nelson, Windsor.
Hogston R (1995) Nurses perceptions of the impact of continuing
professional education (CPE) on the quality of nursing care.
Journal of Advanced Nursing 22, 586593.
Hyrkas K (2005) Clinical supervision, burnout, and job satisfaction
among mental health and psychiatric nurses in Finland. Issues in
Mental Health Nursing 26, 531556.
Jones MC & Johnson DW (1997) Distress, stress and coping in firstyear student nurses. Journal of Advanced Nursing 26, 475482.
Jones MC & Johnson DW (1999) The derivation of a brief Student
Nurses Stress Index. Work and Stress 13, 162181.
Jones MC & Johnson DW (2000) Reducing stress in first level and
student nurses: a review of the applied stress management literature. Journal of Advanced Nursing 26, 475482.
LeSergent CM & Hanley CJ (2005) Rural hospital nurses stressors
and coping strategies: a survey. International Journal of Nursing
Studies 42, 315324.
Mann S & Cowburn J (2005) Emotional labour and stress within
mental health nursing. Journal of Psychiatric and Mental Health
Nursing 12, 154162.
Maslach C & Jackson SE (1986) Maslach Burnout Inventory manual, 2nd edn. Consulting Psychologists Press, Palo Alto, CA.
R Development Core Team (2005) R: A Language and Environment
for Statistical Computing. R Foundation for Statistical Computing,
Vienna.
Ryan D & Watson R (2004) A healthier future. Occupational Health
56, 2021.
Smith GD & Fawcett JN (2006) Stress. In Nursing Practice: Hospital
and Home, 3rd edn. (Alexander MF, Fawcett JN & Runciman PJ
eds). Churchill Livingstone, Edinburgh, pp. 693714.
Timmins F & Nicholl H (2005) Stressors associated with qualified
nurses undertaking part time degree programmes some implications for nurse managers to consider. Journal of Nursing Management 13, 477482.

Appendix 1
The purpose of this third stage of analysis is mainly as a
statistical validity check. Stage 2 attempts to answer
primary research questions but assumes that the probability
of dropout does not depend on the missing GHQ scores
that would have been observed. As mentioned, stage 2

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278

277

R Watson et al.

makes what is known as the missing at random assumption. Stage 3 attempts to see whether the findings from
stage 2 change when this assumption is not made. This
stage of the analysis was also carried out on R version
2.2.0. We proceeded with the mixed model containing the
fixed main effects of age, life event score, sex, status, time
and marital status, the fixed status time interaction, no
random effects and a general covariance structure (no
constraints on correlations or variances of the repeated
GHQ scores). Sensitivity analyses were performed by
adding interaction terms involving the dropout times and
covariates as additional terms to this linear mixed model to
obtain pattern mixture models (Fitzmaurice et al. 2001). In
these models, missing data are no longer assumed to be
missing at random: the possibility that the likelihood of
data being missing is dependent on the unobserved GHQ
scores is permitted. Fitzmaurice et al. (2001) indicate that
the dropout time may be treated as a factor, and this
was performed here. However, the standard errors reported

278

by software from fitting these pattern mixture models as

linear mixed models are not correct. Fitzmaurice et al.
(2001) suggest bootstrapping as a means of obtaining
standard errors. This was used here, carrying out 2000
bootstrap samples for each pattern mixture model fitted.
Another choice required is the manner in which the
occurrence probabilities for the dropout times are estimated. Fitzmaurice et al. (2001) state that these can be
allowed to depend on covariates and so the results from
the first stage were used, giving one set of probabilities for
newly qualified nurses and another set for nursing students.
The pattern mixture sensitivity analyses led to similar
conclusions to those reported above for the second stage
with the missing at random assumption. Results from the
sensitivity analyses (pattern mixture modelling and varying
the correlation structure) are encouraging because they give
us consistent answers to the primary research questions
under a variety of assumed conditions.

 2008 The Authors. Journal compilation  2008 Blackwell Publishing Ltd, Journal of Clinical Nursing, 18, 270278