Developing Holistic Wellness Model
Developing Holistic Wellness Model
Developing Holistic Wellness Model
RP DE LA REY
rdelarey@icon.co.za
Dept of Human Resource Management
University of Johannesburg
ABSTRACT
The objective of the present study was the development of a holistic theoretical wellness model that served as frame
of reference for the construction of a valid and reliable inventory that was considered suitable for the assessment of
wellness of employees of a South African life insurance organisation. The 5 Factor Wellness Inventory was developed
from existing proven wellness inventories. A non-random sample of 673 employees of the organisation concerned
participated in a cross-sectional survey by completing the Inventory and a biographical questionnaire. Goodness of
fit between the holistic wellness model and a data set derived from the application of the 5 Factor Wellness
Inventory was determined empirically. The structural equation modelling statistics produced acceptable goodnessof-fit indices albeit with some scope for improvement. The Root Mean Square Error of Approximation indices
supported acceptance of the holistic work-wellness model. The validity of the Inventory was also estimated.
Key words
Holistic, theoretical, wellness model, assessment of wellness
46
|----------------------------------------|------------------------------------------|
Illness
Health
Wellness
Problem statement
The fact that Wellness is not researched in a positive, holistic,
systemic and integrated manner is a cause for concern (Wissing
& Van Eeden, 1997, Wissing, 2000). Researchers in the health
and social sciences traditionally used a pathogenic paradigm as
point of departure for their investigations and interventions,
which implied an orientation towards abnormality, sickness,
disease and dysfunction as the direct opposite of wellness
(Strmpfer, 1995, 2002). Knowledge thus generated was used to
develop ways for treating and preventing each of these undesired
states of health. Businesses, institutions and professions,
ranging from everyday medical practice to pharmaceutical
companies, from the insurance industry to mass media, all
turned handsome profits by assisting clients to cope with
physical, emotional and behavioural threats as well as by
providing the required treatments (Saleebey, 1997). This onesided approach verifies the fact that attention should also be
given to wellness research an approach implied by the opposite
side of the illness-health-wellness continuum (Antonovsky, 1987;
Seligman & Csikszentmihalyi, 2000; Strmpfer, 1995; Wissing,
2000). This second approach would thus focus on developing
the positive aspects and strengths of human behaviour.
Recently there has been a growing trend among organisations to
realise the importance of the so-called human factor (Snyder &
Lopez, 2002). Various disciplines highlight the importance of a
good life and attend to the total well-being or wellness of
people. Multidisciplinary research merges different theoretical
perspectives into a more realistic wellness paradigm.
Interdisciplinary teams are further better equipped to achieve
holistic integration and a better understanding of wellness
issues. Members of study disciplines other than psychology, all
focus on wellness from their own school of thought and could
contribute meaningfully to joint theory building. Dietary,
nutritional, health and fitness management are examples hereof
but they all tend to avoid a holistic interdisciplinary approach
(Adams et al., 1997). Research into and promotion of holistic
health and wellness requires an integrated and multidisciplinary
approach, even in the case of existing inputs by research
psychologists.
Psychology as a profession acknowledges the importance of a
holistic understanding of the strengths, coping patterns,
adaptive abilities and growth potential of individuals
(Antonovsky,
1987;
Strmpfer,
1995;
Seligman
&
Csikszentmihalyi, 2000). It is crucial for psychologists to assist
people to achieve higher levels of psychological well-being and
wellness (Dunn, 1961, 1977; Wissing & Van Eeden, 1997). This
requires acceptance of the origins of health and psychological
strengths as a point of departure. Behavioural activities and
personality characteristics that promote wellness and prevent
sickness require sound scientific research.
Research should focus on the nature, dimensions and dynamics
of wellness at individual, group, organisational and general
population levels. It is important to understand how these
manifest themselves in various phases of the human life cycle
(childhood, adolescence, adulthood, old-age) in different
contexts (cultural, interpersonal, work or technological
surroundings) and by taking into account internal as well as
external risks and protective factors, demographics and other
moderating or mediating variables (Dunn, 1961). Socially
responsible and ethically respectable programmes based on
scientifically validated knowledge (theories and models) must be
developed, and the effectiveness and efficacy thereof analysed
and evaluated.
47
48
ELS, DE LA REY
49
RESEARCH DESIGN
Research approach
In general research of excellence requires a careful validation
of any new conceptual model (or theory) to ensure that all
relevant constructs, components, concepts and principles are
fully integrated into it before an attempt is made to put into
practice. The aforementioned brief guidelines were followed in
the current project. In practice, research ideals are sometimes
not met while limitations become apparent as the research
process evolves. These and other confounding factors and
events in the current project were acknowledged, accepted and
duly reported on.
The research project under consideration was broadly conducted
in four phases. At first, it was investigative by nature and
consisted of an extensive survey of available literature on work
wellness. In the second stage, the outcome of this overview of
literature was used to assess the complexity, nature and
interrelatedness of components of wellness and also to identify
measurable variates (a variate is a variable and a set of individual
test scores linked to it) that were likely to be considered for
inclusion in a preliminary model of employee wellness. Thirdly,
this prior model of employee wellness was converted into a brief
wellness questionnaire that was researched and validated by
means of a cross-sectional survey design (Bethlehem, 1999).
Finally, a new holistic wellness model was formulated from the
results of an extensive empirical investigation involving
advanced statistical methods. The current research project thus
contributed towards an expansion of both theoretical and
applied behavioural science by initially identifying and mapping
variables linked to a holistic wellness model and by finally
integrating these into an empirically verified model that could
be used in working environments.
The cross-sectional survey design lent itself to the examination
of stable, long-term states or conditions and allowed the
researcher to make inferences from the sample that could be
generalised to different populations. The empirical study
identified connections between variates and wellness constructs
that were not ascribable to any planned intervention as these
were measured and assessed by means of a validated and
unbiased work-wellness inventory.
Practical research problems that were addressed in the
design included errors of measurement. Respondents for
instance did not understand specific questions in the
measuring instrument, processing errors occurred, coding
mistakes were made during data capturing and the thirdvariable problem (where a high correlation between two
variables could have been attributed to a third variable that
correlated significantly with them) had to be managed. One of
the profound practical problems of the chosen design was the
fact that it was almost impossible to establish causation
between co-related variates. A pilot study was conducted to
identif y important sources of error of measurement. The
Statistical Consultation Services of the University of
50
ELS, DE LA REY
Participants
The study population of the current research project
consisted of employees of a major life insurance company in
South Africa. The staff complement of this organisation
consisted of 2 523 employees. A non-random sampling
strategy based on availability, determined the composition of
the sample. All participants, drawn from the different
business units within the organisation under consideration,
volunteered to partake in the present study. None of them had
previously participated in wellness studies. Altogether 673
employees (26.70% of the total staff count) were included in
the actual sample. The sample of participants was judged as
being representative of the staff complement of the
organisation. Detailed assignment of subjects from the
population to the sample approximated the distribution of
biographic attributes of the former in terms of race, marital
status, gender, qualifications and years of service.
Measuring instrument
A single questionnaire, the 5 Factor Wellness Inventory (5F-Wel
Inventory) was used to measure and assess the different wellness
factors. The validity of this measuring device was empirically
assessed in terms of its rationale, development, contents,
psychometric properties, administration, scoring and reliability.
The wellness-measuring device took approximately thirty
minutes to complete. Variables such as the age of participants
and others could have had an effect on the results of the study.
A biographical questionnaire was therefore included to control
for possible bias effects from variables such as age, gender,
length of service and educational background.
SAMPLE
70
51
133
139
156
121
3
22
402
161
58
10
20
Home language
African
English
Afrikaans
Other
Missing
183
202
276
9
3
Race
African
White
Coloured
Indian or Asian
Missing
125
402
76
63
7
Marital status
Single
Married
Separated
Divorced
Widowed
Missing
219
377
6
57
10
4
Male
Female
Missing
245
424
4
Gender
51
Research procedures
The scientific objective of the current research project required
a dedicated research design that was based on ten
methodological considerations that also included several
practical steps namely:
Step 1: The researcher did a survey of literature to extract
knowledge that would lead to a clear holistic
conceptualisation of wellness and further the identification
of constituent components and elements of wellness that
influenced the behaviour and functioning of employees.
Step 2: Retrieved information about the components and
elements of wellness guided the choice of items from existing
psychometric tests of general wellness. Items that were
provisionally judged as suitable for inclusion in a wellness
test appropriate for use in a working environment, were
identified. Selected items were mostly taken from the original
5 Factor Wellness Inventory and included in a shortened
psychometric test called the 5F-Wel Inventory. Availability
and administration of the 5F-Wel Inventory guided several
aspects of the empirical study, such as the research design as
well as the construction, assessment and validation of the
final version of the holistic model.
Step 3: A formal sample was drawn; the research group was
introduced to the researcher, informed of the purpose,
method, and procedures of the study while participants
consent to data access were obtained.
Step 4: The 5F-Wel Inventory was taken down individually.
Participants responded to all variables included in this
measuring device by answering all questions in the inventory
in a single test session.
Step 5: Test-generated raw data, namely the original
individual responses of sampled participants to the test items
in the 5F-Wel Inventory, was captured in a data set. This data
set of 5F-Wel Inventory variates was observable and
unfactored, since it had not yet been subjected to structural
change through statistical manipulation.
Step 6: The unfactored set of variates was statistically
analysed to generate provisional feedback for participants.
Participants personal results were confidential and handled
as such.
Step 7: The unfactored set of variates next was transformed by
means of principal-axis factor analysis into a factored data set
that included the factor scores of participants. This method
extracts latent factor structures that underlie an original set of
variates. Latent scores are not directly observable in an
unfactored data set. The extracted latent factor structures and
weights or loadings, produced factor scores for each
participant.
Step 8: The factor structures produced by the principal-axis
factor analysis were also used to construct and describe the
new wellness model that was the principal scientific objective
of the current study. The validity of the constructs linked to
the new wellness model was also verified by means of visual
and verbal inspection of the wellness structures. Reliability
estimates, such as the consistency of factor scores produced
by the 5F-Wel Inventory, provisionally were made by using
Cronbachs formula for the calculation coefficient alpha.
Step 9: The validity of the wellness model was also
determined by means of statistical structured equation
modelling. This approach to modelling determined the
goodness of fit between the observable or unfactored sets of
variates and the latent or factored data set. From a statistical
perspective, good fit between the two data sets would
confirm the validity of the holistic wellness model.
Step 10: All available research information on wellness was
integrated into a single body of scientific knowledge that was
handed over to the South African insurance organisation that
became involved in the research project. A number of
Statistical analysis
Parts of the statistical analysis were done with the SAS
programme (SAS Institute, 2000). Three types of statistics were
calculated with the SAS programme, namely descriptive
statistics, correlation coefficients and Cronbach estimates of
reliability. All factor analytic calculations that were needed to
examine factor structures were done by means of the SPSS
Windows programme (Ferguson & Takane, 1989). The extraction
method used to determine the factor structure was that of
principal-axes factoring. The Oblimin approach with Kaisers
normalisation was used for the purpose of rotation. The AMOS
programme (Arbuckle, 1999) was used for the structural
equation model. The statistical methods were explained in more
detail elsewhere.
The scientific study of wellness is a difficult assignment. The
construct is multifaceted; studying it involves a multitude of
variables and variates as well as requiring use of advanced
multivariate statistics. High-level research focuses on
examining and understanding of patterns of co-relatedness or
interdependence between sets of variates. Several indices of corelation are available to describe and analyse patterns of
interrelatedness. Inspection of any large matrix of indices of corelation immediately shows that no simple intuitive
interpretation of the pattern of co-relations among a set of
variables is possible (Ferguson & Takane, 1989). Two issues
complicate analysis. Firstly, two kinds of co-relationships or
covariance occur. Variates might co-relate with several other
variates in a data set. Variates of this type appear to measure
part of a common construct since they share communality.
Their mere presence in a data set suggests duplication of
observed variates. A particular variate might not co-relate at all
with any other remaining variates. It thus might measure a
unique component of a construct that is limited to one variate
in the set. Secondly, aggregates of variates tend to form clusters
or patterns of co-relationship: some of these patterns are
observable while others underlie or are latent in a set of variates
and thus are difficult to observe. Factor analysis is a statistical
method that assumes that any set of m variates contains both
communality and uniqueness. The method employs these two
properties to convert the original scores linked to the variates
(raw or unfactored data) into factor scores (factored data) and
uses these to reduce the variables to create a simplified
structure. Reduction implies that the patterns of the original
variates and transformed factor scores differ noticeably: the
converted set carries the latent structure (Harris, 1975).
Factor analysis requires an unfactored or raw data set consisting
of an aggregate of individual test scores linked to a set of m
variates. The statistical method converts the original raw data
into either an m m intercorrelation matrix or variancecovariance matrix. Statistical data that forms the elements of a
matrix is obtained by pairing off each variable with every other
in the set to calculate their communality or uniqueness. The
elements of a matrix thus reflect the patterns of co-relation
between all of the available m variates. The m x m matrix next
is transformed into a factor matrix. An index, such as R2 that
estimates communality of variates, is selected from the contents
of a matrix. One variate at a time is compared with all
remaining variates in the set. Their indices of co-relatedness in
the matrix are then converted into factor scores by multiplying
the element values by the estimated index of communality to
produce a set of weights or loadings that describes the latent or
underlying structure of the original variates in the derived
factor. Summation of the factor loadings produce sets of Eigen
values or l values. The calculation procedures are iterative.
Once a set of factor loadings for a particular variate is extracted,
the procedure reduces the intercorrelation or variancecovariance matrix before a next set of factors for a second
52
ELS, DE LA REY
RESULTS
The 5 Factor - Wellness Survey was used to measure the degree
to which the test responses of employees in this study complied
with the theoretical definition of wellness, as it is defined by
The Indivisible Self: An Evidence-Based Model (Myers &
Sweeney, 2005). The reported 5F-Wel descriptive statistics and
53
THE
(5F-WEL)
Factor
Mean
SD
Skewness
Kurtosis
13,65
1,77
-0,25
-0,80
0,70
II
10,45
3,11
0,02
-0,65
0,81
III
8,01
2,86
1,22
1,96
0,86
IV
9,36
1,52
-0,38
1,24
0,79
7,06
2,10
0,99
0,56
0,77
VI
6,62
1,04
1,00
2,85
0,64
VII
12,10
2,32
-0,13
0,30
0,60
VIII
9,36
2,51
0,38
0,24
0,73
IX
1,88
0,61
0,31
0,66
0,58
27,97
2,61
0,12
-0,12
0,64
XI
10,41
3,08
0,21
-0,28
0,82
XII
7,31
1,10
-0,03
0,47
0,47
XIII
10,72
2,57
0,28
-0,22
0,57
XIV
8,81
1,46
-0,37
0,06
0,18
XV
11,77
2,03
-0,39
0,68
0,68
XVI
5,51
1,54
0,40
0,32
0,56
XVII
11,82
2,81
0,99
0,56
0,75
XVIII
12,55
2,54
0,12
-0,12
0,66
XIX
9,94
2,01
-0,03
0,47
0,40
c2
c2/df
NFI
TLI
CFI
16 828,18
4,32
0,27
0,32
0,33
RMSEA
0, 07
It could be seen from Table 3 that the RMSEA confirmed that the
variates in the data set showed good fit to the model. Browne and
Cudeck (1993) regard this value as an acceptable and reasonable
error of approximation. The c2-based goodness-of-fit statistic
was sensitive to the dimensions of the moment matrix derived
from a data set of large proportions: this anomaly produced an
unreliable index of model fit. Most c2 tests and c2-based indices
are N-sensitive, that is sensitive to sample size. The calculated
54
ELS, DE LA REY
c2/df value deviated significantly from the expected value for the
theory that was being assessed. The numerical value was
therefore rejected as being statistically untenable. In some
instances, non-significant chi-square values are desired in order
to prove the null hypothesis. The NFI value (0,27) indicated that
the model fit could be improved substantially. A NFI of 1.00
would have indicated a perfect fit. The Tucker-Lewis Index (LTI)
and Comparative Fit Index (CFI) both pointed to poor fit with
values lower than 0,90 (Arbuckle and Wothke, 1999). The most
important test, namely RMSEA, indicated a good fit with a value
of 0,07. Browne and Cudeck (1993) stated that a value of 0,08 or
less for the RMSEA indicated a reasonable error of
approximation.
Using the 5FWel model as theoretical point of reference, the
final wellness model was presented in Figure 2.
The proposed model was an explorative model that was
constructed to serve as basis for the development of a holistic
work-wellness model. A single value that was produced by an
important goodness-of-fit index from the available range of
indices, confirmed the validity of the preliminary model. The
proposed multivariate wellness model was arrived at from the
factor analytic structure produced by a principal-axes factor
analysis of the original data set. The goodness of fit between this
factor analytic description of a wellness construct and the
theoretical approach to it that was represented in the original
data set had been tested with the AMOS Basic programme. The
results had confirmed and validated good fit for the purposes of
future application. The structural model in Figure 2 showed
that the insurance organisation might consider all sixteen
wellness factors as well as the three contextual factors as
components of an independent wellness model for future
interventions.
A final word of warning, however, was necessary. As was stated
earlier in explaining the statistical analysis, the usual provision
of Eigen or l values with the results of the principal-axes factor
analysis and use of these to eliminate trivial factors from the
model, were done away with for the purposes of the current
study to ensure a holistic model of wellness. The negative
impact of the two decisions mostly would have affected the
latter factors in the wellness construct. From a statistical point of
view, this approach would cast some doubt on the tenability of
the contextual variates, possibly the social identity variate and
perhaps one or two others as well.
CONCLUSIONS
Conclusions in terms of the results on this section of the study
were based on the statistics from the factor analysis and the
goodness-of-fit statistics. The factor analysis, with some
reservations, pointed to love, leisure, local safety, institutional
concern, nutrition, social identity, stress management,
spirituality, exercise and self-worth as reliable factors as
measured by the 5FWel instrument. Goodness-of-fit statistics
confirmed good fit (RMSEA) for a wellness model for use in an
insurance organisation. However, some indices suggested that
modification might be considered to improve the chi-square
value, NFI, the Tucker-Lewis Index and the Comparative Fit
Index (CFI).
Recommendations
Based on the literature and empirical findings in this research
some recommendations for the organisation and future research
were made.
The current study did not consider wellness within the context
of selection of staff. Further research was thus required to
validate the appropriateness of wellness in this particular
organisational context. Measurement of wellness might be used
in conjunction with other measures of staff selection to ensure
that healthy and well employees are appointed to the
organisation.
With regard to the existing staff members of the organisation
that had participated in this study, the results indicated that
REFERENCES
Adams, T.B., Bezner, J.R. & Steinhardt, M. (1997). The
conceptualisation and measurement of perceived wellness:
integrating balance across and within dimensions. American
Journal of Health Promotion, 12 (3), 380-388.
55
56
ELS, DE LA REY