View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by University of Southern Queensland ePrints

Investigating the Structural Validity of Ryff’s Psychological

Well-Being Scales across Two Samples

Richard A Burns* & M. Anthony Machin

Department of Psychology
University of Southern Queensland
Toowoomba Queensland,
AUSTRALIA

* Corresponding author
Richard Burns
Department of Psychology
University of Southern Queensland
Toowoomba
Queensland,
AUSTRALIA
Email: burns@usq.edu.au

Abstract
Ryff’s (1989b) Psychological Well-Being (PWB) scales measure six related constructs of human
functioning. The present paper examined the validity of Ryff’s 6-factor PWB model, using data
from a life events study (N = 401) and an organisational climate study (N = 679). Previous
validation studies, using Confirmatory Factor Analysis (CFA), have identified alternative PWB
models, but limitations include the use of shorter scale versions with items relating to a number
of life domains within the same PWB factor, and failure to examine the influence of participants’
socio-demographic characteristics on PWB. In this study, Exploratory Factor Analysis (EFA)
mostly found consistency in the PWB items and structure between the two studies whereby a 3-
factor model delineated between items relating to Autonomy, Positive Relations and a super-
ordinate factor comprising the other PWB factors. Using CFA, Goodness of Fit indices reached
acceptable levels for the adjusted PWB model identified by the EFA, whilst differences between
adjusted models of PWB previously identified in the literature were hardly evident. Post-hoc
analysis by gender demonstrated socio-demographic effects on the structure and items that
comprise PWB. Further development of PWB measures is needed to reflect its hierarchical and
multi-dimensional nature. In the scales’ current form, the construct validation of the PWB factors
will continue to be problematic and will fail to adequately evaluate the nature and impact of
PWB.

Decades of research have related notions of positive mental and physical health with the
absence of such adverse states as depression, anxiety, and physical illness. However, a number of
2

researchers have proposed that well-being is not necessarily the antithesis to these constructs of
ill-being (Kahneman, 1999; Ryff, 1989a). A major challenge is to identify relevant dimensions of
well-being and to understand how these dimensions are shaped by various human experiences.

Ryan and Deci (2001) have described two distinct, yet related approaches with which
most psychological theories of well-being could be aligned. The Hedonic, or Subjective Well-
Being (SWB) approach, focused on immediate human functioning and experience, and was
associated with perceptions of pleasure, displeasure, satisfaction, and happiness. With
philosophical roots in antiquity and more recently in Priestley and Bentham’s ‘greatest happiness
principle’, models of SWB have perhaps been the most frequently reported on within the well-
being literature (Diener, Suh, Lucas, & Smith, 1999). Alternatively, a Eudaimonic or
Psychological Well-Being (PWB) model emphasises those mechanisms that are associated with
healthy human functioning and adjustment. Whilst daily SWB fluctuates with life experiences
(Headey 2000, Heady & Wearing, 1989), PWB is a relatively stable construct that captures those
aspects of human functioning more likely to lead to adaptive human functioning and positive
experiences (Ryan & Deci, 2001). Such theories are not new. As with SWB, PWB is grounded in
ancient philosophical works (e.g., Epictetus) whilst more recent proponents have included the
existential writings of Kierkegaarde and Tillich, both of whose discourses into the experience of
melancholia and anxiety led to the conclusion that the good life is one not free of ‘angst’, but one
that is lived in spite of it.

Ryff’s (1989b) Psychological Well-Being model drew on gerontological and life-span

research and reflects one construct-oriented approach to PWB. Its theoretical underpinnings
stemmed from a wide range of influences including Allport’s (1961) concept of the mature
personality, Rogers’ (1961) fully-functioning individual, and Maslow’s (1968) self-actualisation,
and led to the formulation of six dimensions of PWB: Autonomy, Positive Relations with Others,
Environmental Mastery, Personal Growth, Purpose In Life, and Self-Acceptance (Ryff, 1989a;
Ryff, 1989b). With intuitive appeal and widespread interest, the use of the PWB model, its
domains and items, have been applied to a number of different applied psychological areas
(Clarke, Marshall, Ryff & Wheaton, 2001; Fava, Ruini, Rafanelli, Finos, Conti, & Grandi, 2004),
despite unresolved questions relating to its validity (e.g. Springer & Hauser, 2006; Springer,
Hauser, & Freese, 2006). Also, Keyes, Shmotkin and Ryff (2002) have identified relationships
between PWB and SWB variables which raises questions about the degree to which PWB and
SWB are distinct constructs. We propose to investigate the dimensionality of PWB in order to
allow a better understanding of how PWB relates to SWB.

Abbot et al.’s (2006) recent review noted that most psychometric analyses of the PWB scales
occurred almost a decade after the first publication of the PWB scales (Ryff, 1989b) with a
number of different findings being reported. Whilst the ‘a priori’ correlated 6-factor model has
received some support (e.g. Ryff & Keyes, 1995), a number of studies (e.g. Clarke, Marshall,
Ryff & Wheaton, 2001) have indicated a high degree of correlation between four of the PWB
variables: Environmental Mastery (E), Personal Growth (G), Purpose in Life (P), Self-
Acceptance (S) (EGPS), such as to warrant analysing these factors as one super-ordinate factor.
Further analyses of the PWB scales have supported this structure with separate first order factors
for Autonomy and Positive Relations with Others, and one-second order factor containing the
EGPS variables (Abbot et al., 2006).
3

Limitations of existing studies

A number of the studies reviewing the structure of the PWB constructs (e.g. Kafka &
Korma, 2002; Ryff & Keyes, 1995; van Dierendonck et al., 2007) have been fraught with some
methodological limitations. Whilst the original model (Ryff, 1989b) included 120 items, shorter
versions have included 84, 54, 42 and 18 items, generally with equal numbers of items per PWB
variable. Most analyses have tested the factorial validity of the PWB model with the smallest 18-
item (3 items per variable) scale, though two recent analyses used a 42-item scale (Abbot et al.,
2006; Springer & Hauser, 2006) or an amended 39-item scale (van Dierendonck et al., 2007).
There are issues relating to the validity of these findings since there is a lack of consistency in
the items that comprise the shorter versions of the PWB scales. Whilst the 84-item version
comprises all items used in the 54-item scales, there is considerably less overlap in the items
used between the shorter versions, with only 6 common items between the 18 and 42-item scales.
Van Dierendonck (2004) analysed the 84, 54, and 18 item scale versions and found support for a
6-factor model with a second-order PWB factor. Although internal consistencies were high,
Goodness of Fit Indices (GFI) indicated poor fit for the two larger scales.

Further issues relate to the methodology employed in developing the PWB scales. Initial
development of the original 120-item version is explained fully elsewhere (Ryff, 1989b; Ryff &
Singer, 2006), but in summary, an initial pool of some 80-items per variable were reduced to 32
items per variable. Ryff (1989b) then analysed the bi-variate correlations of items to their
respective variable and retained items (20 per variable) with the strongest correlations, as long as
an item’s strongest correlation was reported between the item and its parent variable. Even so,
this process means that some items which scored most highly on their respective variables will
likely fail to discriminate between other variables if they also reported lesser but still very strong
correlations with other variables. This process certainly explains why high correlations (e.g., van
Dierendonck et al., 2007) and cross-loading of items across PWB variables (e.g., Springer &
Hauser, 2006) have been reported, and why internal consistency of the PWB variables is often
quite high (e.g., Ryff, 1989b).

Most published factor analyses of the Ryff PWB scales have used Confirmatory Factor
Analysis (CFA) though one study (Kafka & Kozma, 2002) used an Exploratory Factor Analysis
(EFA) approach to assess the dimensionality of the original 120-item version, and supported a
one general PWB factor. However, the authors first extracted all factors with eigenvalues greater
than one and then the ‘a priori’ 6-factor model using principal components analysis (PCA), with
an orthogonal (Variamax) rotation, a process which is generally described as a data reduction
process. It is not surprising that most of the PWB items loaded onto the first factor. Given the
frequently reported high degree of correlation between the PWB variables, a Principal Axis
Factoring (PAF) method with an oblique rotation would seem most appropriate to identify a
correlated PWB factor structure. A re-analysis of the original item pool with PAF, using an
oblique rotation, should provide a more coherent and defensible set of dimensions.
We propose that the longer scales, at least the 84 and 54 item scale versions, be used to
test the validity of Ryff’s 6-factor structure of PWB. A significant amount of meaningful data is
lost when only 3 or 7 of the original 20 items per variable are used in the data collection as it is
likely that the influence of sample characteristics, like gender (Marks & Lambert, 1998), age
(Ryff & Keyes, 1995), and culture (Ryff, Keyes, & Hughes, 2004), all of which have been
4

demonstrated to have some effect on PWB, will be reflected on the PWB factor structure,
particularly when using a smaller item pool. It may be that sampling characteristics influence
particular response patterns to items of different content. Given these effects, the validity of
Ryff’s (1989b) original development of the model must also be considered with caution since
60% of the original sample (N = 321) were female, and the sample was stratified by three age
groups.

However, despite the weaknesses related to certain aspects of the scales’ initial
construction, and the limitations of some subsequent analyses, considerable evidence (Ryff &
Singer, 2006) does relate PWB to a range of outcomes including biological indicators (Ryff,
Singer, & Love, 2004), successful transitions in later life (Smider, Essex, & Ryff, 1996) and
better counselling interventions (Fava, Ruini, Rafanelli, Finos, Conti, & Grandi, 2004),
supporting the utility of the construct and its operationalisation using Ryff’s PWB model.

Aims of the current study

The current paper seeks to test the factor structure of the PWB model with two of the
larger scale versions (84- and 54-items), on two separate studies: a life events study with an
Australian community sample (N = 401) and a cross-national organisational climate study with
teachers (N = 679). Unlike previous validation studies, we seek to identify a stable structure
underlying Ryff’s (1989b) model of PWB using EFA and to determine whether this structure is
consistent across our studies. In addition, the availability of a SWB measure in each of these
studies allows us to test associations between measures of PWB and SWB across our studies.
Finally, using just those items identified in the EFA, a CFA will compare a range of Goodness of
Fit Indices (GFI) for the following models: a) the ‘a priori’ 6-factor correlated model (Ryff,
1989b), b) a 1-factor model (Kafka & Korma, 2002), c) the structure identified in the EFA, and
d) a model that combines four of the PWB variables: Environmental Mastery, Personal Growth,
Purpose In Life, and Self Acceptance as a second-order factor (Abbott et al., 2006). These
models will be tested with two adjustments previously tested in the PWB literature: the inclusion
of two method variables (Abbott et al. 2006), and the inclusion of correlated error terms (Singer
and Hauser, 2006).

Method
Participants and Design
Data from two studies were used for the analyses in this paper. Study 1 was a Life Events
Study (N = 401) comprising undergraduate students from the Department of Psychology at the
University of Southern Queensland (USQ). Participation in departmental projects is a
requirement of enrolment in some psychology courses. Participants were predominantly female
(83%), and unlike most undergraduate populations, studied part-time (55%) with a relatively
equal distribution from late teens to late forties. These sampling characteristics can be attributed
to the provision of unique educational services by several universities in Australia, like USQ,
which recognise that many do not necessarily follow the traditional route of entering university
within a year or so of having completed their high school qualification. With the impediments
(e.g. family and work responsibilities) associated with entering higher education later in life,
USQ provides opportunities for students to undertake most of their courses on a part-time and
external basis, in addition to the traditional full-time and on-campus modes.
5

Study 2 was an organisational climate study (N = 679) comprising three samples of

schoolteachers, drawn from privately-funded schools in the Australian Capital Territory,
Australia (n = 253), school teacher members of the Norwegian teacher union (n = 250), and from
schools worldwide which designated themselves as being International Schools (n = 176).
Predominantly female (63%), most participants (46.2%) were aged between 30 to 55 years of
age, though 63.2% of the Norwegian sample was aged 45 years and older.

Procedure
Both studies included two measures of well-being, the data from which were analysed for
this study. Ryff’s (1989b) Psychological Well-Being scales assesses six dimensions of PWB:
Environmental Mastery (E), Personal Growth (G), Purpose in Life (P), Self-Acceptance (S),
(EGPS); Autonomy (A); and Positive Relations (PR). An 84-item version was used in Study 1
and a 54-item version was used in Study 2. Unlike shorter versions of the PWB scale, all the
items of the 54-item version are included in the larger 84-item version. Therefore, items from the
54-item version were extracted from the 84-item version used in the life events study, to allow
for comparison with the 54-item version that was used in the organisational climate study.
Individuals indicated their response on a 6-point Likert-type scale, with higher scores on each
scale indicating greater well-being on each dimension. The PANAS (Watson, Clark, & Tellegen,
1988) assessed SWB with 20-items relating to positive affect (PA) and negative affect (NA), and
was assessed in both studies. Individuals indicated their response on a 5-point Likert-type scale,
with higher scores on each scale indicating greater well-being on that dimension.

Both studies were undertaken between June 2006 and June 2007. A high number of
participants in both studies did not live in the immediate vicinity of the university, so therefore
participants accessed the survey through a secure web facility which is run and monitored by the
technical services staff within the Department of Psychology. The University’s Human Research
Ethics Committee provided approval for both studies.

Analyses were undertaken separately for both studies and then compared. In addition,
post-hoc analyses analysed cohort effects in the organisational climate study analysis, and gender
effects for both the life events and organisational climate studies.

Results
Principal axis factoring with an oblique rotation attempted to differentiate between PWB
items. Parallel analysis (O'Connor, 2000) was first used to identify the number of factors to be
extracted for factoring. For each study, the parallel analyses indicated extracting between 6 and 9
factors, however convergence failed when extracting either 6, 7, 8, or 9 factors. Extracting 4 or 5
factors led to a number of items loading across more than one factor, so items were deleted from
the analysis if they loaded above .30 on more than one factor, or failed to achieve this level on
one factor. For both studies, items now loaded onto three factors: Autonomy (A); Positive
Relations (PR); and a first-order factor (EGPS), comprising items relating to Environmental
Mastery (E), Personal Growth (G), Purpose In Life (P), and Self-Acceptance (S). This supports
Abbot et al.’s (2006) findings of a higher order factor, EGPS, however, in this instance, our
EGPS variable reflected a first-order factor. The items and their respective factor loadings are
displayed in Table 1 for each study. Inspection of the item loadings reveal mostly moderate
loadings and indicate a fair degree of consistency in the items that load onto their respective
6

factor. Few differences in the size of loading scores and in those items identified as significant
indicators, are reported.

INSERT TABLE 1

Further analyses sought to differentiate between PWB and SWB. Therefore, the
aforementioned analyses were extended to include 20 items from the PANAS scale. Principal
Axis Factoring and an oblique rotation clearly delineated between the three PWB factors and two
SWB factors: PA and NA, for both studies. Some cross-loading of the SWB items on PWB
factors did occur, but all remained below the .30 criterion cut-off.

Although Factor Analysis differentiated between PWB and SWB items, bi-variate
correlations (Table 2) indicated mostly moderate to strong correlations between all the well-
being factors, with findings generally consistent between the studies. Few differences between
studies were reported and related only to the size of association. All correlations were significant
(p < .001).

INSERT TABLE 2

Confirmatory Factor Analysis (CFA, Table 3) of the items identified in the initial
Exploratory Factor Analysis (EFA, Table 1) was undertaken to assess whether the PWB model
identified in the EFA reported better Goodness of Fit Indices (GFI) than the ‘a priori’ 6-factor
model, and a number of alternative models identified in the literature. Four main models were
tested: Model 1 tested the ‘a priori’ correlated 6-factor model (Ryff, 1989b); Model 2 tested a
general 1-factor model (Kafka & Korma, 2002); Model 3 tested for results identified in the EFA
reported earlier in this paper whereby a first-order factor (EGPS) comprised items relating to
Environmental Mastery, Personal Growth, Purpose In Life, Self-Acceptance; and Model 4
replicated previous findings (Abbot et al, 2006) which identified EGPS as a second-order factor.
CFA analyses were performed using the items identified for each study from the original EFA.

INSERT TABLE 3

In addition to these four main models, we tested additional methodological factors, which
have also been identified in the literature. For example, Springer and Hauser (2006) and Abbott
et al. (2006) introduced several adjustments to test for methodological effects. Springer and
Hauser introduced a latent variable to account for reverse-scored items, which they found
significantly improved fit. Abbott et al. found strong support for the introduction of two method
factors reflecting positive (non-reversed scores) and negative (reversed scores) method factors.
Springer and Hauser found further support for methodological effects by correlating the error
terms of adjacent items and items with similar content. The inclusion of these paths is not usually
recommended unless there is a strong theoretical basis for doing so, such as when item content is
similar, when there is a likelihood of social response bias/desirability, where a model omits the
inclusion of an exogenous variable, and in repeated measures designs where items are measured
on two or more occasions (Aish & Jöreskog, 1990; Byrne, 2001). Based on these findings, we
tested the effect of including two method factors, as well as significant error covariances.
However, we did not believe testing for correlated adjacent items was warranted since the
7

structure of the PWB scales includes intermittent use of items that require reverse scoring,
negatively and positively phrased items, as well the systematic ordering of items so that no item
from the same variable is placed adjacent to each other. Our rationale for including significant
correlated error terms assumes that Springer and Hauser’s findings reflect other artifact such as
response bias, which is common in attitude surveys and when items are similar in content. These
additional effects were tested in Models 5-8 (two method factors) and Models 9-12 (significant
error covariances).

In models 1-4, the pattern of findings was identical across both studies for all models.
The six-factor model (Model 1) was clearly a better fitting model than the single factor model
and those models with the first and second order EGPS factor, however GFI were far from
acceptable. Models 5-8, which included the two latent method variables, performed better than
Model 1, though the six-factor model with the additional two method factors (Model 5),
performed better than either of the other models. Models 9-12 tested the effect of including
significant covariances between correlated error terms. Positive covariances were included if
they reported Modification Index values above 4, and if the association was significant (p = .05).
All four of these models performed better than previous models though differences in GFI
between models 9 to 12 were less apparent with different GFI indicating different models as best
fitting. Many of the models that tested the method variables consisted of paths, between the
method variables and the items, that failed to achieve significance (p > .05), whilst the models
with error covariances only included significant associations. The significant correlated error
terms included in the analyses varied between the studies and this may reflect differences
between participants, where socio-demographic characteristics might be related to different PWB
items.

It was surprising that the model with the first order EGPS did not report the best fit
considering this was the factor identified in the EFA and that only those items identified in the
EFA for each sample were included in the CFA. The authors have previously suggested that
sampling characteristics, in particular age and gender, may influence the structure of PWB and
this may reflect differences between participants. Unfortunately, any post-hoc analysis of the
factor structure by a number of socio-demographic variables was limited by the design of the
studies and by the variables that were operationalised for the original purpose of each study. This
precluded post-hoc analyses of a number of socio-demographic effects on PWB, such as age. For
example, age groupings were not comparable between the Life Events study and the
Organisational Climate study as they had been designed to reflect the age range of the targeted
study participants. Furthermore, there was a preponderance of young to middle aged adults in the
life events study, and middle-aged to late middle-aged adults in the organisational climate study
which precluded a sub-groups analysis of PWB by age within each study. However, a sub-groups
analysis of the original items (Table 1) by gender (Table 4) for both studies, and by cohort in the
organisational climate study (Table 5), was possible. Both of these analyses found some support
for the PWB structure reported in the initial EFA findings. In particular, the results for the
Australian and International teacher cohorts were considerably similar to the findings of the
original EFA. However, analysis by the different teacher cohort is still likely to demonstrate the
effect of participant characteristics and these findings need to be considered in this light.
8

Not all items were consistently reported with equal loading by gender and cohort, and
several items cross-loaded onto other factors. This may explain why the factor structure with a
first-order EGPS variable did not outperform the alternative and ‘a priori’ models. Whilst
females reported items with factor loadings that more closely mirrored the overall results for
both samples, this is quite likely a consequence of the greater proportion of females in both
studies. However, consistent differences between males and females for several items (e.g.
Environmental Mastery item 53, and Autonomy item 16) between studies, does support the
notion that perhaps there are differences between gender on items that comprise PWB.

INSERT TABLE 4

INSERT TABLE 5

Discussion
In two studies, PAF with oblique rotation delineated three PWB variables: Autonomy,
Positive Relations, and EGPS, a first-order factor first previously identified as a second-order
factor by Abbott et al. (2006), and comprising the Environmental Mastery, Personal Growth,
Purpose in Life and Self Acceptance items. The inclusion of a SWB measure identified two
SWB factors: Positive Affect and Negative Affect, which were distinct from the PWB variables,
although significant correlations between all the SWB and PWB variables were reported. Whilst
some differences between studies in the items constructing the PWB variables were reported, the
structure of PWB was consistent between the studies. Post-hoc analysis of the different teacher
cohorts within the organisational climate study and by gender for both studies, revealed that
sampling characteristics appear to influence both the structure and items that comprise PWB.

These results support a number of previous findings which have postulated either a
simple 1-factor model, a correlated 6-factor model, as well as first or second-order factors which
incorporated the EGPS variables reported in this study and elsewhere (Abbott et al., 2006).
Initial GFI of the unmodified models were poor, though the six-factor model was the preferred
model. Two types of adjustments were assessed and included the addition of method factors
(Springer & Hauser, 2006; Abbott et al., 2006) and reported much better fit. Despite some
concern about the methodological, theoretical and statistical implications, a second adjustment
expanded on previous findings (Springer & Hauser, 2006) which allowed for correlated error
terms. Results demonstrated acceptable and comparable fit for all four models where significant
paths between correlated error terms were included.

Based on our findings we recommend the use of EFA techniques with larger item pools,
and to remove less important items, or items that are related to more than one factor, in
subsequent analysis. This would appear to be a happy medium where larger scales improve
internal consistency and shorter scales that are suited to factor analysis (Van Dienrendonck et al.,
2004). We recognise that the items and structure of PWB will reflect particular characteristics of
the sample, but would hypothesise that a larger item pool will increase the likelihood of
identifying a consistent structure to the PWB model, though gender, age and other socio-
demographic effects on the structure of PWB are to be expected.
9

In relation to Ryff’s (1989b) original model development, Principal Axis Factoring, using
an oblique rotation may have proved a more fruitful methodological approach. Whilst the final
item pool may have resulted in a multi-dimensional model of PWB, a correlational approach
fails to consider item content which enables greater differentiation between highly related
constructs, rather leading to the inclusion of items that differ in the extent to which they assess
specific versus general judgements of well-being. For example, the Environmental Mastery items
cover a wide range of areas of personal control, from control of daily life responsibilities (item
#17), to control of time and demands (item #36), to control of personal finances (item #29) and
participants’ responses will surely reflect the importance of each particular issue for people at
different ages or stages in their lives. Items that comprised the Personal Growth variable also
reflect a mixture of items that relate to one’s personal growth through life to date (item #s 37. 45
and 50), or reflects on the prospect of continuing to face the challenges to one’s growth and
development (item 21). Clearly, people of different ages, who are at different stages of their
lives, may relate to these questions in quite different ways. For instance, older participants may
relate more easily to the reflective questions, whilst younger participants may, in comparison,
have shorter temporal contexts within which to reflect on such issues. In contrast, findings for
the future-oriented questions may be more important for younger participants.

Such issues are not new to models of self-referent beliefs and attitudes. Decades of
research into self-concept failed to consider the implications of generating items within scales
that comprise a mixture of items whose content fail to distinguish between global and context-
specific judgments. It wasn’t until reviews of the existing self-concept measures of the day (e.g.
Wylie, 1974; Burns, 1979), that these weaknesses in self-concept surveys were summarily
identified, revealing that they failed to address these very same issues. Consequently, Shavelson,
Hubner and Stanton (1978) proposed a multidimensional and hierarchical nature of self-concept
that reflected this structure, whilst Marsh’s (1992) construction of the Self Description
Questionnaires (SDQ) operationalised Shavelson et al.’s model, and has since identified the
utility of a multi-dimensional and hierarchical structure to self-referent beliefs.

In a similar vein, we would propose a model of well-being that is both multi-dimensional

and hierarchical in nature. Whilst well-being variables at a higher order level may certainly have
differential predictions on a number of outcomes, we believe that further development of well-
being models are needed and must consider the domain and level of specificity that is being
assessed. Such a model would incorporate the eudaimonic processes that Ryff has sought to
address at a general level, the hedonic states captured by SWB measures (e.g. PANAS), as well
as physical and biological health correlates. Support for such a model has previously been
indicated (Keyes, Shmotkin, & Ryff, 2002) where analysis delineated distinct yet related
associations between PWB and SWB variables, a finding supported in this study whereby factor
analysis of PWB and SWB items delineated differences at the item level, but with high
correlations between the variables indicating highly related constructs. We would also propose
that the relationship between these different dimensions is reciprocal, though stronger causal
paths from PWB to SWB may be expected, and that the strength of this reciprocal nature is
reflected by the level of the hierarchy at which the association is investigated.

More recent analyses of the PWB scales (Abbottt et al., 2006; Springer & Hauser, 2006;
Van Dierendonck et al., 2007) have used PRELIS or MPlus to provide polychoric correlation
10

estimates as previous methods have perhaps incorrectly assumed PWB responses to reflect
continuous data, which can bias estimates. However, it is common to assume that Likert scales
that consist of at least 5 points can be analysed as if reflecting a continuous scale (Dollan, 1994).
Still, regardless of sample size (Jöreskog & Sörbom, 1996), the use of Pearson correlation
matrices in Factor Analysis appears to underestimate the degree of association between variables
and consequently results in reduced factor loadings (DiStefano, 2002).

Approaches that use polychoric correlations may be warranted in some circumstances,

but such techniques may only prove to be more stringent. Whilst computing polychoric and
tetrachoric matrices is certainly possible in some statistical packages, its use in personality
research, for example, where scales frequently comprise Likert scales similar to the PWB scales
and where most factor analysis has typically used bivariate correlation matrices, has indicated
that this approach generally fails to produce dissimilar results from traditional methods.
Holgado-Tello, Carrasco-Ortiz, Victroria del Barrio-Gandara and Chacon-Moscoso (2007) tested
the veracity of the Five-Factor Personality Model using polychoric estimates and concluded that
the polychoric estimation approach produced results comparable to previous non-polychoric
approaches. It is perhaps for these reasons that so few commercially available statistical
packages allow for these sorts of techniques. Since the use of such methods in personality
research has contributed little, we do not think that their use in well-being research would expand
our knowledge of the structure of PWB other than, as previously said, to provide a more
stringent approach to estimating the correlation matrices. Instead, we believe that the issues
relating to the use of the larger scales and the extent to which sample characteristics have
influenced the previous results, are much more important issues to consider.

The use of the larger 84- and 54-item scales in this paper is an improvement on previous
validation studies that have used the shorter scale versions, which comprise far fewer similar
items, and has resulted in considerable confusion about the efficacy of Ryff’s PWB scales. As
well, previous PWB validation studies have typically reported internal reliability and CFA
techniques, whilst the methodology employed in the one EFA study (Kafka & Korma, 2002) has
serious limitations. CFA procedures are conceptually different from EFA techniques, being
generally theory rather than data driven, and this study has addressed these concerns by using an
EFA technique to analyse the larger scale versions, resulting in a revised 3-factor model of PWB,
supporting Abbot et al.’s (2006) CFA findings that four of the PWB variables are highly
interrelated. To the authors’ knowledge, this is the first paper that has undertaken an EFA
approach that has correctly used a PAF procedure with an oblique rotation. Importantly, at the
item level, we found this procedure could delineate PWB constructs from two broad valence
SWB constructs, positive and negative affect, whilst an oblique rotation reported moderate
associations at the factor level.

Some limitations to our studies relate to the relatively small sample sizes for the CFA to
generate reliable parameter estimates in Structural Equation Modelling. In addition, these studies
were designed independently of each other and therefore a number of socio-demographic
variables were classified differently from each other. For instance, age groupings reflected the
target population of each study and as such a sub-groups analysis of the PWB scales based on
age and other demographic variables was not possible. The preliminary sub-groups analysis of
gender did reveal some differences and may explain why there is such debate over the validity of
11

Ryff’s model of PWB as differences in the findings of previous validation studies may reflect
these sampling characteristics. However, the unequal distribution of gender in both our studies
prohibits us from placing too much weight to this claim. The post-hoc analysis by cohort within
the teacher study revealed greater consistency in the items and structure identified by EFA and
suggests that some demographic effects may be consistent amongst Western schoolteachers
though differences are still apparent. It should be noted that the Abbot et al. (2006) analysis
comprised a birth cohort sample who were all female, therefore age and gender effects could not
have been an issue for their findings which identified the second order EGPS variable as a better
fitting model than the 6-factor model. Still it does provide support for the multi-dimensional
properties purported to be measured by Ryff’s (1989) PWB scales, but also a hierarchical
structure which we suggest needs to be investigated further.

Conclusion
The purpose of this paper has not been to discredit the value of the Ryff PWB scales, nor
the previous validation studies cited in this paper. Rather, the authors have sought a back-to-
basics approach to test the dimensionality of the PWB scales. The complexity in drawing out a
satisfactory conclusion on the structural validity of PWB, leads us to suggest that further
development into the nature and structure of well-being, which recognises the multiple domains
and hierarchical structure inherent to self-referent attitudes, is warranted. Further analyses should
identify the extent to which socio-demographic characteristics may influence the structural
validity of the PWB scales. Finally, we would propose that Ryff’s (1989b) PWB scales are
limited by item content that comprise both general and context-specific judgements of well-
being. However, we recognise the link between PWB in its current form and a number of health
outcomes, and would conclude that Ryff’s PWB scales are an appropriate tool for assessing
distinct aspects of PWB at a general level, though the extent to which this can be replicated
across populations will be influenced by sampling characteristics.

References
Abbot, R. A., Ploubidis, G. B., Huppert, F. A., Kuh, D., Wadsworth, M. E. J., & Croudace, T. J.
12

(2006). Psychometric evaluation and predictive validity of Ryff´s psychological

wellbeing items in a UK cohort sample of women. Health and Quality of Life Outcomes,
4, 76.
Aish, A. M., & Jöreskog, K. G. (1990). A panel model for political efficacy and responsiveness:
an application of LISREL 7 with weighted least squares. Quality & Quantity 24: 405-
426, 1990.
Allport, G. W. (1961), Pattern and growth in personality. New York: Holt, Rinehart and
Winston.
Bernstein , I. H., Garbin, C., & Teng, G. (1988 ). Applied Multivariate Analysis. New York:
Springer-Verlag.
Burns, R. B. (1979). The self-concept theory in measurement, development and behavior.
London: Longman.
Byrne, B. M. (2001). Structural Equation Modeling With AMOS: Basic Concepts, Applications
and Programming. New Jersey: Lawrence Erlbaum Associates.
Clarke, P. J., Marshall, V. W., Ryff, C. D., & Wheaton, B. (2001). Measuring psychological
well-being in the Canadian Study of Health and Aging. International Psychogeriatrics,
13 Supp 1:79-90.
Costa, P. T. & McCrae, R. R. (1985). The NEO Personality Inventory. Odessa: PAR.
Diener, E., Suh, E M., Lucas, R. E., & Smith, H. L. (1999). Subjective Well-Being: Three
Decades of Progress. Psychological Bulletin, Vol: 125(2), p 276–302.
DiStefano, C. (2002). The impact of categorization with confirmatory factor analysis. Structural
Equation Modeling, 9, 327–346.
Dollan, C. V. (1994). Factor analysis of variables with 2, 3, 4 and 7 response categories: A
comparison of categorical variable estimators using simulated data. British Journal of
Mathematical and Statistical Psychology, 47, 309–326.
Fava, G. M., Ruini, C., Rafanelli, C., Finos, L., Conti, S., & Grandi, S. (2004). Six-year outcome
of cognitive behavior therapy for prevention of recurrent depression. American Journal of
Psychiatry 161, 1872–1876.
Headey, B. (2000). Subjective Well-Being: Revisions to Dynamic Equilibrium Theory using
National Panel Data and Panel Regression Methods, Social Indicators Research, 79:
369–403
Headey, B.W., & Wearing, A. J. (1989). Personality, life events and subjective well-being:
Toward a dynamic equilibrium model, Journal of Personality and Social Psychology, 57,
731–739.
Holgado-Tello, F. P., Carrasco-Ortiz, M. A., Victroria del Barrio-Gandara, M. & Chacon-
Moscoso, S. (2007). Factor analysis of the Big Five Questionnaire using polychoric
correlations in children. Quality and Quantity, DOI: 10.1007/s11135-007-9085-3, 202-
214.
Jöreskog, K.G., & Sörbom, D. (1996). PRELIS 2: User’s Reference Guide. Chicago: SSI
Kafka, G. J., & Kozma, A. (2002). The construct validity of Ryff's scales of psychological well-
being (SPWB) and their relationship to measures of subjective well-being. Social
Indicators Research, 57:171-190.
Kahneman, D. (1999) Objective happiness, in D. Kahneman, E. Diener and N. Schwarz (eds.),
Well-Being: The Foundations of Hedonic Psychology. New York: Russell Sage. pp. 3–
25.
Keyes, C. L. M., Shmotkin, D., & Ryff, C. D. (2002). Optimising Well-being: The empirical
13

encounter of two traditions. Journal of Personality and Social Psychology, Vol. 82, No.6,
1007 – 1022.
Marks, N. F., & Lambert, J. D. (1998). Marital status continuity and change among young and
midlife adults: longitudinal effects on psychological well-being, Journal of Family Issues
19, pp. 652–686.
Marsh, H. W. (1992). Self Description Questonnaire (SDQ) III: A theoretical and empirical
basis for the measurement of multiple dimensions of late adolescent self-concept: An
interim test manual and research monograph. Macarthur, NSW, Australia: Faculty of
Education, University of Western Sydney.
Maslow, A. (1968). Toward a psychology of being (2nd ed.). New York: Van Nostrand.
O'Connor, B. P. (2000). SPSS and SAS programs for determining the number of components
using parallel analysis and Velicer's MAP test. Behavior Research Methods,
Instrumentation, and Computers, 32, 396-402.
Rogers, C. R. (1961). On becoming a person. Oxford: Houghton Mifflin,
Ryff, C. D. (1989a). Beyond Ponce de Leon and life satisfaction: New directions in quest of
successful aging. International Journal of Behavioral Development, 12, 35–55.
Ryff, C. D. (1989b). Happiness is everything, or is it? Explorations on the meaning of
psychological well-being. Journal of Personality and Social Psychology, 57, 1069-1081.
Ryan, R., Deci, E. L. (2001). On happiness and human potentials: a review of research on
hedonic and eudaimonic well-being. Annual Review of Psychology, 52:141-166.
Ryff, C. D., & Keyes, C. L. M. (1995). The Structure of Psychological Well-Being Revisited.
Journal of Personality and Social Psychology, Vol. 69(4), 719-727.
Ryff, C. D., & Singer, B. H. (2006). Best news yet on the six-factor model of well-being. Social
Science Research, 35, 1103–1119.
Ryff, C. D., Keyes, C. L. M., & Hughes, D. L. (2004) Psychological well-being in MIDUS:
profiles of ethnic/racial diversity and life course uniformity. In: O.G. Brim, C.D. Ryff
and R.C. Kessler, Editors, How Healthy Are We? A National Study of Well-being at Mid-
life, The John D. and Catherine T. MacArthur Foundation Series on Mental Health and
Development: Studies on Successful Midlife Development, Chicago: University of
Chicago Press, 398–424.
Ryff, C. D., Singer, B. H., & Love, G. D. (2004) Positive health: connecting well-being with
biology, Philosophical Transactions of the Royal Society of London B 359, 1383–1394.
Shavelson, R. J., Hubner, J. J., & Stanton, G. C. (1976). Self-concept: Validation of construct
interpretations. Review of Educational Research, 46, 407-441.
Smider, N. A., Essex, M. J., & Ryff, C. D. (1996). Adaptation of community relocation: The
interactive influence of psychological resources and contextual factors. Psychology and
Aging, Vol 11(2), 362-372.
Springer, K. W., & Hauser, R. M. (2006). An assessment of the construct validity of Ryff’s
scales of psychological well-being: Method, mode, and measurement effects. Social
Science Research, 35, 1080–1102.
Springer, K. W., Hauser, R. M., & Freese, J. (2006). Bad news indeed for Ryff’s six-factor
model of well-being. Social Science Research, 35, 1120 – 1131.
Van Dierendonck, D. (2004). The construct validity of Ryff’s scales of psychological well-being
and its extension with spiritual well-being. Personality and Individual Differences, 36,
629–643.
van Dierendonck, D., Díaz D., Rodríguez-Carvajal, R., Blanco, A. & Moreno-Jiménez, B.
14

(2007). Ryff’s Six-factor Model of Psychological Well-being, A Spanish Exploration.

Social Indicators Research DOI 10.1007/s11205-007-9174-7
Watson, D., Clark, A. & Tellegen, A. (1988). Development and validation of brief measures
of positive and negative affect: The PANAS scales, Journal of Personality and Social
Psychology 54, 1063–1070.
Wylie, R. C. (1974). The self-concept: A critical survey of pertinent research literature. Lincoln,
NE: University of Nebraska Press.

Acknowledgements
The authors thank the participants from both studies, Mr Ross Bool and the Technical Services
staff for organizing the online questionnaires, and Dr. Tim Windsor for feedback on an earlier
manuscript.
15
Table 1 A comparison of the item loadings of the 54 item PWB scale by study

Life Events Study Teacher Study

PWB Variable item # Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3

02 .446 - - - - -
07 - - - - - -
12 - - - - - -
17 .706 - - .640 - -
20 - - - - - -
Environmental Mastery 29 .540 - - .408 - -
36 .643 - - .477 - -
49 - - - - - -

53 - - - .520 - -
03 - - - - - -
18 - - - - - -
21 .487 - - .322 - -
26 - - - - - -
Personal Growth 37 .776 - - .581 - -
41 - - - - - -
45 .669 - - .458 - -
50 .546 - - .419 - -
54 - - - - - -
8 - - - - - -
13 .373 - - - - -
22 - - - - - -
27 .444 - - - - -
Purpose In Life 30 .604 - - - - -
33 .806 - - .511 - -
38 .838 - - .693 - -
42 .630 - - .483 - -
46 - - - - - -
4 - - - - - -
9 - - - - - -
14 - - - - - -
23 .395 - - .411 - -
Self-Acceptance 28 - - - .368 - -
31 - - - - - -
43 - - - - - -
48 - - - - - -
51 .460 - - .486 - -
1 - - - - - -
5 - .649 - - .527 -
10 - .747 - - .707 -
15 - - - - - -
Positive Relations 24 - .620 - - .754 -
32 - .739 - - .595 -
34 - - - - - -
39 - .728 - - .501 -
47 - .527 - - - -
6 - - - - - .402
11 - - .618 - - .309
16 - - .520 - - .480
19 - - - - - -
Autonomy 25 - - .638 - - .469
35 - - - - - -
40 - - .501 - - .671
44 - - .658 - - .434
52 - - .426 - - -
**Teacher study was assessed using Wave 1 data only. - Item either cross-loaded onto more than one factor or
loaded weakly (<.30) onto one factor. $Italics indicate negatively worded items
16
Table 2 A comparison of the correlations* between PWB and SWB variables by Study

1 2 3 4 5
Life Events Study

1. Positive Affect 1 -.243 .638 .317 .433

School Sample
2. Negative Affect -.224 1 -.405 -.453 -.362

Combined
+
3. EGPS .589 -.237 1 .361 .527
4. Positive Relations .182 -.386 .314 1 .306
5. Autonomy .250 -.293 .305 .247 1
*All correlations significant at p < .001, +EGPS comprises items relating to Environmental Mastery, Personal
Growth, Purpose In Life, and Self-Acceptance.
17
Table 3 Confirmatory Factor Analysis by study testing several structural models of PWB using the items identified
by Exploratory Factor Analysis

Model Study Cmina Df GFI CFI RMSEA (95% CI)

Life Events Study 708.32 284 .877 .889 .061(.055 - .067)
Model 1
Combined Teacher Samples 1010.14 260 .891 .808 .065(.061- .070)
Life Events Study 1721.70 300 .690 .630 .109(.104 - .114)
Model 2
Combined Teacher Samples 1980.81 275 .778 .570 .095(.091 - .099)
Life Events Study 897.41 296 .842 .843 .071(.066 - .077)
Model 3
Combined Teacher Samples 1088.19 272 .882 .792 .067(.062-.071)
Life Events Study 769.65 292 .867 .876 .064(.058 - .069)
Model 4
Combined Teacher Samples 1036.65 268 .898 .804 .065(.061 - .069)
Life Events Study 477.77 257 .918 .943 .046(.040 - .053)
Model 5
Combined Teacher Samples 820.70 237 .913 .851 .060(.056 - .065)
Life Events Study 924.38 272 .829 .830 .077(.072 - .083)
Model 6
Combined Teacher Samples 1085.57 249 .884 .786 .070(.066 - .075)
Life Events Study 644.04 269 .883 .902 .059 (.053 - .065)
Model 7
Combined Teacher Samples 905.90 248 .902 .832 .063(.058 - .067)
Life Events Study 548.34 265 .905 .926 .052(.046 - .058)
Model 8
Combined Teacher Samples 876.00 245 .906 .839 .062(.057 - .066)
Life Events Study 471.49 258 .920 .944 045(.039 - .052)
Model 9
Combined Teacher Samples 547.30 222 .940 .917 .046(.042 - .051)
Life Events Study 444.87 241 .923 .947 .046(.039 - .053)
Model 10
Combined Teacher Samples 489.27 216 .946 .930 .043(038 - .048)
Life Events Study 511.27 263 .913 .935 .049(.042- .055)
Model 11
Combined Teacher Samples 536.79 232 .942 .922 .044(.031 - .049)
Life Events Study 496.50 265 .916 .940 .047(.040- 053)
Model 12
Combined Teacher Samples 559.637 232 .939 .916 .046(.041 - .050)
Model 1: ‘a priori’ 6 Correlated factors; Model 2: 1 PWB Factor; Model 3: EGPS 1st order factor correlated with A and
nd
PR; Model 4: EGPS 2 order factor correlated with A and PR. Model 5 = Model 1 with correlated method variables;
Model 6 = Model 2 with correlated method variables; Model 7 = Model 3 with correlated method variables; Model 8 =
Model 4 with correlated method variables. Model 9 = Model 1 with significant correlated error terms; Model 10 =
Model 2 with significant correlated error terms; Model 11 = Model 3 with significant correlated error terms; Model 12 =
Model 4 with significant correlated error terms. a all chi square statistics were significant p = .000.
18
Table 4 A comparison of the item loadings of PWB by gender and by study of the items extracted from the original EFA

Teacher Study (N = 679 ) Life Events Study (N = 401)

Item Male (n = 252) Female (n = 427) Male (n = 68) Female (n = 333 )

PWB Variable No. Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3

02 - - - - - - .531 - - .440 - -

17 .690 - - .629 - - .672 - - .634 - -

Environmental
29 .586 - - .304 - - .453 - - .545 - -
Mastery
36 .432 - - .418 - - .594 - - .557 - -

53 .595 .335 - .522 - - .540 .378 - .432 - -

21 - - .412 .317 - - .557 - - .436 - -

26 - - - - - - .319 - - .402 - -
Personal
37 .588 - - .586 - - .837 - - .741 - -
Growth
45 .626 - - .314 - - .726 - - .613 - -

50 .535 - - .338 - - .455 - .402 .557 - -

13 - - - - - - - .412 .430 .520 - -

27 - - - - - - - - .380 .466 - -

30 - - - - - - .343 - .473 .695 - -

Purpose In Life 33 .413 .379 - .424 - - .742 - - .777 - -

38 .768 - - .613 - - - - - - - -

42 .420 - - .469 - - .592 - - .586 - -

46 - - - - - - - - .306 - - -

Self-Acceptance 23 .414 - - .374 - - .599 - - .341 - .311

28 .436 - - .416 - - .536 - - .373 - -

19
51 .400 - .492 .491 - - .529 - - .468 - -

05 - .575 - - -.478 - - .836 - - .604 -

10 - .694 - - -.766 - - .734 - - .753 -

Positive 24 - .643 - - -.768 - - .623 - - .604 -

Relations 32 - .311 .414 - -.659 - - .693 - - .747 -

39 - - - - -.547 - - .596 - - .742 -

47 - - - - - - .487 .493 -.312 - .523 -

06 -.344 .346 .785 - - .394 .444 - - - - .551

11 - - - - - .387 - -.302 .452 - - .662

16 - - .445 - - .504 - .323 .317 - - .540

Autonomy 25 - - .502 - - .467 - - .770 - - .569

40 - - .301 - - .720 - - .545 - - .490

44 - - .325 - - .447 - - .772 - - .587

52 - - - - - - - - .475 - - .423

- Item either cross-loaded onto more than one factor or loaded weakly (<.30) onto one factor.
20
Table 5 A comparison of the item loadings of PWB by teacher cohort of the items extracted from the original EFA

Teacher Study (N = 679 )

International Teacher Cohort (n = Norwegian Teacher Cohort Australian Teacher Cohort

Item 176) (n = 250) (n = 253)

PWB Variable No. Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3

17 - - - - -
.672 .448 .303
.704
29 - - - - - - -
.646 .528
Environmental Mastery
36 - - - - - - -
.593 .422
53 - - - - - -
.549 .553 .563
21 - - - - - - -
.448 .390
37 - - - - - -
.674 .333 .478
Personal Growth
45 - - - - - -
.556 .435 .360
50 - - -
.370 .314 .355 -.385 .450 .330
33 - - - -
.571 .334 .453 .361
.334
Purpose In Life 38 - - - - -
.807 .545 .413 .551
42 - - - - - -
.559 .441
23 - - - - - -
.585 .468 .353
Self-Acceptance 28 - - - - - -
.539 .407
51 - - - - - -
.557 .538 .534
Positive Relations 05 - - - - - -
.583 .374 .605
10 - - - - - -
.717 .587 .657
24 - - - - - -
.691 .606 .657
32 - - - - - -
.587 .544 .505
21
39 - - - - - -
.486 .533 .406
06 - - - - -
.359 .304 .348 .414
11 - - - - - - -
.345 .493
16 - - - - -
.496 .409 .365 .526
Autonomy
25 - - - - - -
.446 .370 .440
40 - - - - -
.796 .351 .328 .780
44 - - - - - -
.350 .311 .437
- Item loaded weakly (<.30) onto factor.