Soc Indic Res (2009) 92:489–496

DOI 10.1007/s11205-008-9303-y

A Note on the Dimensionality of Quality of Life Scales:

An Illustration with the Satisfaction with Life Scale
(SWLS)

Suzanne L. Slocum-Gori Æ Bruno D. Zumbo Æ Alex C. Michalos Æ Ed Diener

Accepted: 11 August 2008 / Published online: 29 August 2008

Springer Science+Business Media B.V. 2008

Abstract A case is made that measures used in quality of life and happiness research will
be essentially unidimensional: inherently tapping minor dimensions. This is illustrated
using Diener’s Satisfaction with Life Scale (SWLS). It is shown that the SWLS does not
meet the standard of strict unidimensionality, but that the interpretation of the total scale
score is not compromised because the additional dimensions are relatively minor. In the
context of the example, a multi-step strategy is described that allows researchers to test for
essential unidimensionality. Throughout the article, essential unidimensionality is con-
trasted with the received view of strict unidimensionality and confirmatory factor analysis
methods.

Keywords Psychometrics
Quality of life
Dimensionality
Measurement

Validity

The most common format for quality of life scales, and other psychological and health
scales, is to ask the respondents to rate a set of statements as they apply to her/his life and
to then compute a total scale score from the responses. For example, as can be seen in
Fig. 1, Diener’s Satisfaction with Life Scale (Diener et al. 1985; Pavot and Diener 1993a)
presents respondents with five statements and instructs them to use a 7-point rating scale.
The researcher then sums the item responses to obtain a total ‘‘scale’’ score on the measure.

S. L. Slocum-Gori
B. D. Zumbo (&)

Department of ECPS (Program Area: Measurement, Evaluation, and Research Methodology),
University of British Columbia, Scarfe Building, 2125 Main Mall, Vancouver, BC V6T 1Z4, Canada
e-mail: bruno.zumbo@ubc.ca

A. C. Michalos
University of Northern British Columbia, Prince George, BC, Canada

E. Diener
University of Illinois, Champaign, IL, USA

123
490 S. L. Slocum-Gori et al.

Fig. 1 The SWLS Below are five statements that you may agree or disagree with. Using the 1 - 7 scale
below indicate your agreement with each item by placing the appropriate number on the
line preceding that item. Please be open and honest in your responding.

7 - Strongly agree
6 - Agree
5 - Slightly agree
4 - Neither agree nor disagree
3 - Slightly disagree
2 - Disagree
1 - Strongly disagree

1. In most ways my life is close to my ideal.

2. The conditions of my life are excellent.

3. I am satisfied with my life.

4. So far I have gotten the important things I want in life.

5. If I could live my life over, I would change almost nothing.

The limitations to the inferences that one can make from the total scale score are the
central focus of validity research (see the volume by Zumbo (1998) for a discussion of the
theory and practice of validation in quality of life research). Statistical validation research
should be focused on supporting the interpretations made from the total scale score (see,
for example, Kane 2006; Zumbo 2007).
As part of validation research, we are interested in the following question: Given that
the items are combined to create one scale score, do they measure one latent variable? A
methodological matter of question, however, is what one means by the statement ‘‘… do
they measure one latent variable.’’ Is one to read this to mean that the items should measure
one, and only one, latent variable (henceforth referred to as ‘‘unidimensionality’’)? Or is
one to read this to mean that the items should measure predominantly one latent variable
with some possible secondary minor latent variables (henceforth referred to as ‘‘essential
unidimensionality’’)? A question that follows naturally is whether unidimensionality (i.e.,
one latent variable) is an unnecessarily strict standard for quality of life and other psy-
chological measures. Or, is essential unidimensionality sufficient for the inferences made
with quality of life measures?
Note that most quality of life measures are used for epidemiological, evaluation, and
psychological or sociological studies wherein group-level summary statistics are reported
and/or the total scale scores are used in correlation and predictive statistical analyses.
However, the essential feature in these analyses involves ordering or ranking individuals
based on their quality of life scores. Individual scores are not interpreted on their own, but
rather one is interested in using the magnitude of the scores in data analyses.
The current measurement methodology literature gives us two answers to these ques-
tions. First, the tradition that has evolved in the methodology of confirmatory factor
analysis is that an item should load on one, and only one, latent variable. For example,
Byrne’s (1998) text, a widely used text that has shaped a great deal of practice of validation
research, exemplifies the spirit of the confirmatory factor analysis tradition. For a sub-
jective well being scale that is summed to one total score, such as Diener’s SWLS, the
a priori specification of the confirmatory factor analysis model would allow all the items to
load on only one latent variable. See Vittersø et al. (2002), and Vautier et al. (2004) for
recent examples where the strict standard of unidimensionality is applied and empirically
tested by using confirmatory factor analysis of the SWLS.

123
Psychometric Properties of Life Satisfaction Scales 491

Second, it has been recognized since the early 1950s that psychological measures
usually will not, by definition and design, be unidimensional because the constructs and/or
the measurement process will elicit secondary minor latent variables. Humphreys has
argued for over half a century that the construction of valid psychometric measures
requires that these measures consist of a dominant latent variable and numerous minor
latent variables (Humphreys 1952, 1962, 1970, 1982, 1985). Humphreys’ point is that to
measure any psychological latent variable of interest (e.g., subjective well being), the
inclusion of numerous minor latent variables is not merely desirable, but unavoidable. It is
important to note that these unavoidable minor latent variables may represent ‘‘… nonerror
noise intrinsic to behavioral measurement of psychologically important traits …’’
(Humphreys 1982, p. 2). This particular tradition stems from exploratory factor analysis
and has culminated in the introduction of the psychometric concept of essential unidi-
mensionality. The qualifier ‘‘essential’’ is used widely in the psychometric literature but
‘‘largely,’’ ‘‘primarily,’’ or ‘‘mainly’’ unidimensional may be more accurate and under-
standable for researchers.
On a related note, a variation on Humphreys’ thesis can be seen in the personality
psychology literature where its argued by some (e.g., Church and Burke 1994) that the
confirmatory (statistical model testing) factor analysis appears to perform optimally only
when the factor pattern solution exhibits simple structure. Typically, simple structure is
observed when each variable measures one and only one factor in the domain. Of course,
one should be reminded that simple structure was introduced by Thurstone as a way of
side-stepping the rotational problem in exploratory factor analysis: the problem is that
when one has more than one factor, there is no way of choosing among the many possible
rotations that perform equally well in describing the data. Thurstone’s solution relied on a
form of parsimony and the extent to which it is a reasonable solution to the rotational
problems, and whether it is a reasonable criterion for deciding on the number of factors, is
still debatable.
What does essential unidimensionality mean for quality of life and other such psy-
chological measures? What this means is that if one holds quality of life measures to the
standard of unidimensionality (one and only one latent variable) some of the most widely
used measures would fail—we will illustrate that below with our example. Instead, we
would agree with many of our psychometric colleagues that unidimensionality is simply
too strict a requirement and that essential unidimensionality is the sufficient condition one
needs to meet to interpret and use the total scale score of psychological measures with
confidence.
It is important that readers do not leave with an impression that essential unidimen-
sionality is an inferior methodological standard. Rather, one should note that essential
unidimensionality is the sufficient condition in order to sum up item responses to one total
score for use in epidemiological, psychological, and sociological research because con-
structs in these domains will naturally have secondary minor latent variables. In short,
unidimensionality is an ideal that is not realistic for epidemiological, psychological, and
sociological research.
The purpose of this article is to illustrate how one should approach the question of the
number of latent variables measured by a set of items, which are summed to a single total
score and hence purported to measure one latent variable. The take-home message from
this article is that quality of life measures will not be unidimensional; so the methodology
needs to reflect essential unidimensionality.

123
492 S. L. Slocum-Gori et al.

1 Example from Quality of Life Research

1.1 Sample

The respondents were 410 adults from northern British Columbia, Canada. The age ranged
from 18 to 90 years (M = 46.9, SD = 13.42). There were 239 males and 166 females,
with the remainder not reporting their gender. The male respondents were older (males
M = 49, females M = 44). The median household income was $65,000.

1.2 Analyses and Results

An examination of the bivariate scatterplots among the items indicates that there are no
aberrant data points and that the relationships tend to be linear. Because we have a 7-point
scale we can use a Pearson covariance matrix (Byrne 1998; Muthén and Kaplan 1985,
1992). Therefore, we are going to use Pearson covariance matrix and ML estimation—
Normal Maximum Likelihood Factor Analysis. A component analysis was used only in
helping decide on the number of factors to extract and hence to create the Scree plot but the
loadings are from the ML estimation. The criterion for model fit will be the RMSEA and
the Chi-squared test. According to Browne and Cudeck (1993) the RMSEA should be less
than 0.05 to indicate a good fit, values ranging from 0.08 to 0.10 indicate mediocre fit, and
greater than 0.10 indicate poor fit.

1.2.1 Unidimensionality

The factor model posited one latent variable with all the items loading on that one latent
variable. The model parameters (i.e., loadings and uniquenesses) were estimated via
normal maximum likelihood factor analysis. The minimum fit function Chi-Square with 5
degrees of freedom was 32.93, p = 0.0001, and the RMSEA was 0.117, indicating that the
unidimensional model is not supported by the data. Although typically the factor loadings
would not be interpreted for a model that is not supported by the data, they are reported in
Table 1 because they were useful in the interpretation of the essential unidimensionality
results. Even though the model is not supported by the data the loadings for the unidi-
mensional solution are substantial, except for the last item ‘‘If I could live my life over, I
would change almost nothing’’ which had a loading of middling magnitude.

Table 1 Factor loadings for the unidimensional and essentially unidimensional factor models
Items Unidimensionality Essential
Unidimensionality

Factor 1 Factor 2

In most ways my life is close to my ideal. .889 .841 .056

The conditions of my life are excellent. .911 1.017 -.101
I am satisfied with my life. .907 .747 .185
So far I have gotten the important things I want in life. .745 -.003 .917
If I could live my life over, I would change almost nothing. .588 .288 .355
Note: The loadings for the essential unidimensionality solution are based on the pattern matrix

123
Psychometric Properties of Life Satisfaction Scales 493

1.2.2 Essential Unidimensionality

The essential unidimensionality model posited two latent variables with all items free to
load on both latent variables. Again, normal maximum likelihood factor analysis was
conducted except that the eigenvalues were computed and plotted for the Scree plot in
Fig. 2. It is clear from the Scree plot and inspection of the eigenvalues that there is one
dominant latent variable that accounts for the 73% of the multivariate variation among the
items. In addition, the second latent variable, which due to the mathematics of eigenvalue
decomposition will always be smaller in magnitude, accounts for 12% of the multivariate
variation in the item responses. This is an essentially unidimensional model because
although the factor model has two latent variables, the first latent variable is, from the
Scree plot, obviously dominant over the remaining latent variables.
The essentially unidimensional model was clearly supported by the data: minimum fit
function Chi-Squared with 1 degree of freedom is 0.70, p = 0.402, and an RMSEA of
0.0001. Table 1 lists the factor loadings for the essentially unidimensional model. To aid in
the labeling of the factors, a Direct Oblimin with Kaiser normalization factor rotation was
conducted. The inter-factor correlation was 0.788. From Table 1 it is evident that the first
three items define the first (and dominant) latent variable whereas the fourth item defines
the secondary dimension. The last item is cross-loading, and weakly on both (a related
finding regarding the last item can be seen in Vittersø et al. (2002)). A similar factor
solution, with the first three items loading on one factor and the last two items loading on a
separate factor, was also found by McDonald (1999), Chapter 9 and Vautier et al. (2004)
using confirmatory factor analysis.

1.3 Interpreting the Total Score in the Context of an Essentially Unidimensional Scale

Does meeting the standard of essential unidimensionality allows one to interpret the total
scale score as a quantity that varies along one dimension? That is, can the variation in the
total score be mostly explained by one source, the dominant latent variable? To answer this
question, the factor scores for the unidimensionality and essential unidimensionality
models were saved in the database. The correlation results in Table 2 show that the total
SWLS scale score is highly correlated with the unidimensional factor and with the

Fig. 2 Scree plot Scree Plot

4

3
Eigenvalue

0
1 2 3 4 5
Component Number

123
494 S. L. Slocum-Gori et al.

Table 2 Correlations among the factor scores and the scale total score
1 2 3 4

1. Unidimensional 1.00
2. Dominant factor, of the essentially unidimensional model .990 1.00
3. Secondary minor factor of the essentially unidimensional model .783 .701 1.00
4. SWB total scale score .973 .944 .853 1.00

dominant factor resulting from the essential unidimensionality model. Of course, because
the dominant and secondary factors are highly correlated the secondary factor is correlated
to both the unidimensional factor and the SWLS total score.1

2 Conclusions

Three observations are noteworthy at this point. First, because the inter-factor correlation
for the essential unidimensional results is so high, the latent variables, although separated,
are manifestations of a dominant latent variable that Diener labels as ‘‘subjective well-
being’’—which explains why it is appropriate for the SWLS to be scored as a single
composite variable. Second, when conducting a typical factor analysis the number of
eigenvalues greater than one is an indicator of the dimensionality of a scale. Our example,
however, is a good illustration of how the eigenvalue-greater-than-one rule can sometimes
be misleading for essentially unidimensional scales.
Third, it is important to note that other studies point to the finding that Diener’s SWLS
does not meet the strict unidimensionality criterion. Where these studies differ, however, is
how the secondary factors are interpreted. For example, Vautier et al. (2004) consider the
secondary factors as measurement artifact and hence problematic. Judge (1990), on the
other hand, also found that a single dimension does not capture the construct of life
satisfaction but he postulates that the secondary dimensions reflect affective well-being. As
Pavot and Diener (1993b, p. 169) state, Judge’s findings suggest that, although related, life
satisfaction and affective well-being are separable. Our perspective on this matter is that
secondary dimensions are inherent in our measures, and that these secondary dimensions
are not necessarily error or artifact, but rather they are inherent in the measurement
process.
It is important to note that factor analysis alone cannot determine whether these addi-
tional factors (dimensions) are measuring a ‘‘secondary minor latent variable’’ versus just
measuring noise that has no long-term substance. In fact, one needs to be reminded at this
point of Messick’s conceptualization of measurement validity (see, Hubley and Zumbo
1996; Zumbo 1998, 2007), which places construct validity at the core of validity and is the
key organizing principle of the empirical validity data. In short, one builds a nomological
network of ideas and data to buttress the inferences one makes from measures. Therefore,
to separate substance from noise we advocate a data-driven exploratory strategy that does

1
Note that because there is not a mathematically optimal and unique factor score estimation method, the
inter-factor correlation will not equal the correlation among the two-factor scores; in our case, the interfactor
correlation matrix from the Direct Oblimin rotation was .788 whereas the factor scores are correlated .701
for those same two factors.

123
Psychometric Properties of Life Satisfaction Scales 495

not consider these minor extra dimensions as nuisance, but rather admits that they come
with the mental measurement territory.
An example of our view can be seen when measuring the domain of mathematical
problem solving ability. When testing mathematical ability with written problems verbal
abilities will be inherently tested as well—it is difficult to imagine how one could convey a
mathematical problem to an examinee without somehow tapping into verbal ability. Yet,
we score and interpret these test results as mathematical ability and not verbal ability. The
presence of verbal ability is expected and natural. One can try and conceive of the verbal
ability component as ‘‘error’’ or ‘‘artifact’’ but the reality is that it is going to be there
whenever one measures the mathematical ability to solve problems. From the essential
unidimensionality perspective, mathematical problem solving abilities cannot be measured
without also tapping into verbal abilities, and that mathematical problem solving has an
inherent feature of reading and verbal comprehension.
In conclusion, we should not lose sight of the purpose of factor analysis in psycho-
metric work: One factor analyses a set of items to help one interpret the total score. In
this context, essential unidimensionality means that even though there is a secondary
minor factor, one can still interpret the total scale score as a quantity largely varying
along one dimension.
The take-home messages in this article are:
• Strict unidimensionality is an unnecessarily strict desideratum because psychological
variables will naturally have secondary minor factors. In our example, the SWLS fails
the statistical test of strict unidimensionality.
• Although a scale may fail strict unidimensionality it may meet the standard of essential
unidimensionality, which is more realistic for psychological constructs. Again, this is
seen in our example with the SWLS.
• Meeting the standard of essential unidimensionality allows one to interpret the total
scale score as a quantity that varies along one dimension, as seen in Table 2.
Our example illustrates a multi-step approach to operationally test for essential unidi-
mensionality that involves an investigation of the:
(1) statistical fit of a unidimensional model,
(2) multi-factor models: adding secondary factors to the model until one attains a model
that statistically fits the data; with all the items allowed to load freely on all the
factors,
(3) eigenvalues and Scree plot, and
(4) correlation of the total scale score with the various factor scores from the
unidimensional and multi-dimensional solutions above.
An essentially unidimensional model is supported when one finds the one-factor solution
being rejected (step 1) but it is evident (from steps 2 through 4 above) that the construct
itself is comprised of predominantly one latent variable. That is, the multi-factor model in
step 2 fits and the results in step 3 suggest that the secondary factors are minor relative to
the first factor and therefore the scale is ‘‘essentially’’ unidimensional. Step 4 allows one to
investigate whether the total scale score is highly correlated with the dominant factor from
step 2 and hence supports the interpretation of the total score as the dominant variable.
Recent research (e.g., Zumbo et al. 2003) is extending this methodology in line with the
research questions posed by Vittersø et al. (2002) comparing life satisfaction across cul-
tures and nations.

123
496 S. L. Slocum-Gori et al.

References

Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen & J. S. Long
(Eds.), Testing structural equation models (pp. 136–162). Newbury Park, CA: Sage.
Byrne, B. M. (1998). Structural equation modeling with LISREL, PRELIS, and SIMPLIS: Basic concepts,
applications, and programming. Mahwah, NJ: Lawrence Erlbaum Associates.
Church, A. T., & Burke, P. J. (1994). Exploratory and confirmatory tests of the Big Five and Tellegen’s
three- and four-dimensional models. Journal of Personality and Social Psychology, 66, 93–114. doi:
10.1037/0022-3514.66.1.93.
Diener, E., Emmons, R. A., Larsen, R. J., & Griffin, S. (1985). The satisfaction with life scale. Journal of
Personality Assessment, 49, 71–75. doi:10.1207/s15327752jpa4901_13.
Hubley, A. M., & Zumbo, B. D. (1996). A dialectic on validity: Where we have been and where we are
going. The Journal of General Psychology, 123, 207–215.
Humphreys, L. G. (1952). Individual differences. In C. P. Stone & D. Taylor (Eds.), Annual review of
psychology (Vol. 3, pp. 133–147). Palo Alto, CA: Annual Reviews, Inc.
Humphreys, L. G. (1962). Organization of human abilities. The American Psychologist, 17, 475–483. doi:
10.1037/h0041550.
Humphreys, L. G. (1970). A skeptical look at the pure-factor test. In C. E. Lunneborg (Ed.), Current
problems and techniques in multivariate psychology: Proceedings of a conference honoring Professor
Paul Horst. Seattle: The University of Washington.
Humphreys, L. G. (1982). Systematic heterogeneity of items in tests of meaningful psychological attributes:
A rejection of unidimensionality. University of Illinois (Unpublished).
Humphreys, L. G. (1985). General intelligence: An integration of factor, test, and simplex theory. In B. B.
Wolman (Ed.), Handbook of intelligence: Theories, measurement, and applications. New York: Wiley.
Judge, T. (1990). Job satisfaction as a reflection of disposition: Investigating the relationship and its effects
on employee adaptive behaviors. Doctoral Dissertation, University of Illinois.
Kane, M. (2006). Validation. In R. Brennan (Ed.), Educational measurement (4th ed., pp. 17–64). Westport,
CT: American Council on Education and Praeger.
McDonald, R. P. (1999). Test theory: A unified treatment. Mahwah, NJ: Lawrence Erlbaum Associates
Press.
Muthén, B., & Kaplan, D. (1985). A comparison of some methodologies for the factor analysis of nonnormal
Likert variables. The British Journal of Mathematical and Statistical Psychology, 38, 171–189.
Muthén, B., & Kaplan, D. (1992). A comparison of some methodologies for the factor analysis for non-
normal Likert variables: A note on the size of the model. The British Journal of Mathematical and
Statistical Psychology, 45, 19–30.
Pavot, W., & Diener, E. (1993a). The affective and cognitive context of self-reported measures of subjective
well being. Social Indicators Research, 28, 1–20. doi:10.1007/BF01086714.
Pavot, W., & Diener, E. (1993b). Review of the satisfaction with life scale. Psychological Assessment, 5,
164–172. doi:10.1037/1040-3590.5.2.164.
Vautier, S., Mullet, E., & Jmel, S. (2004). Assessing the structural robustness of self-rated satisfaction with
life: A SEM analysis. Social Indicators Research: An International and Interdisciplinary Journal for
Quality-of-Life Measurement, 68, 235–249.
Vittersø, J., Røysamb, E., & Diener, E. (2002). The concept of life satisfaction across cultures: Exploring its
diverse meaning and relation to economic wealth. In E. Gullone & R. A. Cummins (Eds.), The
universality of subjective wellbeing indicators (pp. 81–103). The Netherlands: Kluwer Academic
Publishers.
Zumbo, B. D. (Ed.). (1998). Validity theory and the methods used in validation: Perspectives from the social
and behavioral sciences. Social Indicators Research: An International and Interdisciplinary Journal
for Quality-of-Life Measurement, 45(1–3), 1–509.
Zumbo, B. D. (2007). Validity: Foundational issues and statistical methodology. In C. R. Rao & S. Sinharay
(Eds.), Handbook of statistics: Psychometrics (Vol. 26, pp. 45–79). The Netherlands: Elsevier Science
B.V.
Zumbo, B. D., Sireci, S. G., & Hambleton, R. K. (2003). Re-visiting exploratory methods for construct
comparability and measurement invariance: Is there something to be gained from the ways of old? In
Annual Meeting of the National Council for Measurement in Education (NCME), Chicago, Illinois.

123