Open Journal of Social Sciences, 2014, 2, 96-99
Published Online May 2014 in SciRes.
How to cite this paper: Ladstätter, F., et al. (2014) Neural Network Analysis of Nonlinear Effects of Hardiness on Burnout in
Chinese Nurses. Open Journal of Social Sciences, 2, 96-99. . 2014. 25019
Neural Network Analysis of Nonlinear
Effects of Hardiness on Burnout in
Chinese Nurses
Felix Ladstätter1, Eva Garrosa2, Junming Dai3
1Department of Psychology, IE University, Segovia, Spain
2Faculty of Psychology, UAM, Madrid, Spain
3School of Public Health, Fudan University, Shanghai, China
Received January 2014
Substantial research attention is evident in the hardiness and related literature concerning the
topic of moderational effects of hardiness on work-related stressors and strains. In this research
mostly linear methods have been used to analyze these moderational effects. However, it is not
very likely that these effects are purely linear. The present study uses a neural network, a method
which can model nonlinear relationships, to analyze the effects of hardiness. A cluster analysis of
268 Chinese nurses based on their self-ratings in the hardiness dimensions of commitment, chal-
lenge, and control was performed. Two groups of individuals were identified, consisting of (1)
those who scored above average and (2), those who scored below average on all hardiness dimen-
sions. On the basis of these clusters, a multi-layer neural network was used to analyze the data.
Hardiness, Cluster Analysis, Artificial Neural Network, Burnout, Nursing
1. Introduction
Nursing is a profession which is highly vulnerable to stress, and it is considered that the stressors nurses are ex-
posed to a problem that affects the practice worldwide [1]. During the past decade, the health care system has
experienced major changes (e.g. rising readmission rates, the ever-growing emphasis on efficiency, etc.).
The Peoples’ Republic of China (PRC) faces a unique set of challenges in health service delivery. The rapid
transformation during the last 30 years in the PRC towards a modern working life is associated with increasing
demands for learning new skills, the need to adapt to new types of work, pressure for higher productivity and
quality of work, time pressure, hustle and growing psychological workload and stress among the workforce [2],
especially in the health professions.
Burnout is understood as a specific occupational stress in human services, resulting from the psychologically
and emotionally demanding relationships between caregivers and their patients. The operational definition of
F. Ladstätter et al.
burnout in human services has three dimensions: emotional exhaustion, depersonalization, and lack of personal
accomplishment. Based on these three dimensions, the experience of burnout can be alleviated by the availabili-
ty of personal resources.
Ever since Kobasa [3] introduced the concept of hardiness as an important personality characteristic affecting
the relationship between stressors and strains, many studies showed its relevance for health and performance [4].
Hardiness has been characterized by the three interrelated dimensions (3Cs) of commitment, control, and chal-
lenge. Extant evidence shows that hardy people perform better and stay healthier in the face of stress [5].
2. Method
The Nursing Burnout ScaleShort Form (NBS-SF) was used to evaluate the process of burnout. The survey of-
fers measures of specific job stressors in nursing as antecedents of burnout (16 items), burnout (12 items), har-
diness (12 items), and consequences of burnout (12 items), totaling in 52 items.
3. Results
Cronbach alphas were calculated to ensure the reliability of the NBS-SF scale (Table 1).
We used the k-means cluster analysis to categorize participants on the basis of their mean z-scores on each of
the hardiness dimensions. Clusters that would make sense if obtained according to the existing theoretical back-
ground on the role of hardiness include patterns comprising above average or below average scores on each har-
diness dimension. The mean z-scores of each hardiness dimension by cluster and the percentage of participants
in each cluster are reported in Table 2. Cluster 1 consists of hardy individuals and cluster 2 of non-hardy indi-
viduals. To confirm that the clusters extracted from the data indeed consist of individuals with different charac-
teristics besides their distinct hardiness profile, comparisons between clusters were conducted to determine
whether or not they were higher on variables that have been proposed to be associated with hardiness, using
one-way ANOVA. Mean scores on all measures for both clusters are shown in Table 1.
In order to further improve our understanding of how hardiness produces its effect and, especially, to show
that the relationships between the stressors and strains, as well as the moderating effect of hardiness are nonli-
near, artificial neural networks, a relatively new methodology that was found to be superior to regression in nu-
merous problem domains [6] was used. Specifically, a three-layer feed-forward network was used for the hardi-
ness model approximation. After training and simulation of the network, a linear regression was performed be-
tween the network outputs and the desired outputs for each of the three types of consequences, separately and
To assess the predictive capacity of the burnout model, a network validation was executed after the training.
All results including t-test statistics and their significance levels are summarized in Table 3. The t-test was used
to compare the predicted outcome of the neural network with the desired outcome.
Visual analyses revealed that the relationships between stressor variables (antecedents), burnout, and conse-
quences were not only nonlinear but also different for hardy and non-hardy individuals. Figure 1 exhibits a
contour plot in which physical consequences are shown as a function of work overload and contact with death
and pain for hardy and non-hardy individuals.
The contour plot graphically illustrates that the stressors role ambiguity and contact with death and pain are
not linearly related to organizational consequences because the surface is not a flat plane but instead a curved
surface. It furthermore reveals the nonlinear influence of hardiness in this relationship because the curved sur-
face displays different shapes for hardy and non-hardy individuals.
4. Discussion
The clusters we identified are suggestive of possible moderational effects on the relation observed between
stressors and strains. Specifically, hardy people who score above average on all hardiness dimensions score be-
low average on all stressors, burnout dimensions and on psychological consequences. Non-hardy individuals
who score below average on all hardiness dimensions score above average on all stressors, burnout dimensions
and on organizational and physical consequences (see Table 1).
The neural network analysis revealed nonlinear relations (Figure 1) between the stressors and strains as well
as a moderational effect of hardiness.
F. Ladstätter et al.
Table 1. Cronbach alphas, means and standard deviations of associates for both hardiness clusters and F-statistic of the
one-way ANOVA.
Variables Alpha Cluster 1 Cluster 2 F(1266)
Role ambiguity 0.73 2.22 0.466 1.86 0.469 37.60***
Contact with death and pain 0.75 2.66 0.449 2.23 0.479 43.33***
Troubled interaction 0.76 2.62 0.539 2.33 0.536 15.81**
Work overload 0.78 2.79 0.522 2.48 0.561 17.91**
Emotional exhaustion 0.81 2.58 0.566 2.35 0.598 10.33**
Depersonalization 0.79 2.22 0.476 1.87 0.472 15.28***
Lack of personal accomplishment 0.74 2.40 0.533 1.93 0.471 37.00***
Psychological C. 0.77 2.27 0.483 1.99 0.509 20.21***
Organizational C. 0.85 2.90 0.540 2.52 0.570 31.24***
Physical C. 0.78 2.74 0.527 2.51 0.485 13.60***
Commitment 0.79 2.32 0.356 2.98 0.400 201.15***
Challenge 0.80 2.46 0.445 3.08 0.361 100.48***
Control 0.83 2.49 0.373 3.14 0.308 179.76***
**p < 0.01. ***p < 0.001.
Table 2. Number of participants per cluster. Percentage of the sample and mean z-scores for all three hardiness dimensions.
1 2
N 145 123
% of total (N = 268) 55.1% 44.9%
Hardiness dimensions
Commitment 0.60 0.74
Challenge 0.51 0.63
Control 0.53 0.65
Table 3. Determination coefficients (R2) and t-Test statistics (t, p) for the neural network.
Data sets Psychological Organizational Physical
R2 t p R2 t p R2 t p
Training (N = 227, df = 226) 0.57 0.305 0.698 0.67 1.197 0.263 0.49 0.299 0.765
Validation (N = 49, df = 48) 0.37 1.361 0.225 0.39 0.646 0.521 0.36 0.889 0.394
For both, hardy and non-hardy individuals, there is a relatively flat area for contact with death and pain levels
of 1 - 3 and work overload levels of 1 - 3. This tells us that within this area, even if the input variables change
within the above mentioned ranges, psychological consequences keep constant (and low). However, if the levels
of contact with pain and death increase (3 - 4) we find a very steep area which indicates that in this area the va-
riable contact with pain and death has a severe effect on psychological consequences, namely, they increase
dramatically especially in the case of non-hardy persons. This means that there is a point in the range of contact
F. Ladstätter et al.
Figure 1. Effect of Hardiness on the Relation between Stressors and Strains.
with pain and death levels until which change has little to no effect on psychological consequences. However, if
this point is exceeded, the psychological consequences increase severely.
An interesting issue for further research would be if other hardiness profiles besides hardy and non-hardy
ones might deepen the understanding of how hardiness affects the relation between stressors and strains.
[1] Schaufeli, W.B. and Enzmann, D. (1998) The Burnout Companion to Study and Practice: A Critical Analysis. Taylor
& Francis, London.
[2] Garrosa, E., Moreno-Jiménez, B., Ladstätter, F. and Liang, Y. (2006) Globalization of Occupational Health: China’s
Development and Reform. Medicina y Seguridad del Trabajo, 202, 39-44.
[3] Kobasa, S.C. (1979) Stressful Life Events, Personality, and Health: An Inquiry into Hardiness. Journal of Personality
and Social Psychology, 37, 1-11.
[4] Eschleman, K.J., Bowling, N.A. and Alarcon, G.M. (2010) A Meta-Analytic Examination of Hardiness. International
Journal of Stress Management, 17, 277-307.
[5] Bartone, P.T., Eid, J., Johnsen, B.H., Laberg, J.C. and Snook, S.A. (2009) Big Five Personality Factors, Hardiness, and
Social Judgment as Predictors of Leader Performance. Leadership and Organization Development Journal, 30, 498-
[6] Ladstätter, F., Garrosa, E., Badea, C. and Moreno, B. (2010) Application of Artificial Neural Networks to a Study of
Nursing Burnout. Ergonomics, 53, 1085-1096.