Evaluating the structure of human values with confirmatory factor analysis Shalom H. Schwartz a, * and Klaus Boehnke b a Department of Psychology, The Hebrew University of Jerusalem, Jerusalem 91905, Israel b International University Bremen, Bremen, Germany Abstract This is the first statistical test of a theory of the structure of human values (Schwartz, 1992). The theory postulates that 10 basic values are discriminated in all societies and that these val- ues form a quasi-circumplex structure based on the inherent conflict or compatibility between their motivational goals. Past support for the theory came from subjective judgments of visual plots of the relations among value items in samples from over 60 countries. We formally test the postulated structure and several potential refinements. We employ a specially designed confirmatory factor analysis (CFA) approach with new data from two sets of 23 samples from 27 countries (N ¼ 10,857). In both data sets, CFAs confirm the 10 basic values, a modified quasi-circumplex rather than a simple circumplex structure, and the claim that values form a motivational continuum. Ó 2003 Elsevier Inc. All rights reserved. Keywords: Values; Confirmatory factor analysis; Value structure; Cross-cultural analyses 1. Introduction The past 45 years have seen a steady stream of papers that propose that the best way to represent personality and affect is a circumplex structure (Fabrigar, Visser, & Browne, 1997; Tracey, 2000). That is, they postulate that personality or affect vari- ables lie on the circumference of a circle, and the strength of association between variables decreases as the distance between variables on the circle increases. A recent book on models of personality and emotions (Plutchik & Conte, 1997a) has given * Corresponding author. E-mail address: [email protected](S.H. Schwartz). 0092-6566/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0092-6566(03)00069-2 Journal of Research in Personality 38 (2004) 230–255 www.elsevier.com/locate/jrp JOURNAL OF RESEARCH IN PERSONALITY
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
JOURNAL OF
ESEARCH IN
R
Journal of Research in Personality 38 (2004) 230–255
www.elsevier.com/locate/jrp
PERSONALITY
Evaluating the structure of human valueswith confirmatory factor analysis
Shalom H. Schwartza,* and Klaus Boehnkeb
a Department of Psychology, The Hebrew University of Jerusalem, Jerusalem 91905, Israelb International University Bremen, Bremen, Germany
Abstract
This is the first statistical test of a theory of the structure of human values (Schwartz, 1992).
The theory postulates that 10 basic values are discriminated in all societies and that these val-
ues form a quasi-circumplex structure based on the inherent conflict or compatibility between
their motivational goals. Past support for the theory came from subjective judgments of visual
plots of the relations among value items in samples from over 60 countries. We formally test
the postulated structure and several potential refinements. We employ a specially designed
confirmatory factor analysis (CFA) approach with new data from two sets of 23 samples from
27 countries (N¼ 10,857). In both data sets, CFAs confirm the 10 basic values, a modified
quasi-circumplex rather than a simple circumplex structure, and the claim that values form
a motivational continuum.
� 2003 Elsevier Inc. All rights reserved.
Keywords: Values; Confirmatory factor analysis; Value structure; Cross-cultural analyses
1. Introduction
The past 45 years have seen a steady stream of papers that propose that the best
way to represent personality and affect is a circumplex structure (Fabrigar, Visser, &Browne, 1997; Tracey, 2000). That is, they postulate that personality or affect vari-
ables lie on the circumference of a circle, and the strength of association between
variables decreases as the distance between variables on the circle increases. A recent
book on models of personality and emotions (Plutchik & Conte, 1997a) has given
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 231
this approach even greater impetus. The current paper concerns the domain of
individual values in which the most widely recognized current model also proposes
a circumplex structure (Rohan, 2000; Schwartz, 1992).
Factor/principal components analysis and multidimensional scaling (MDS) are the
methods commonly used to test circumplex structures, but they are usually inade-quate to provide statistical tests. Neither offers a simple, quantifiable method to for-
mally assess the extent to which the observed data possess a circumplex structure (for
detailed critiques of these methods, see Fabrigar et al., 1997 or Tracey, 2000). Instead,
researchers typically reach conclusions by making two subjective judgments of the ob-
served plot of relations among variables. First, they assess how well this plot appears
to conform to a circular pattern. Then they assess the extent to which the order of the
variables around the circle appears to correspond to the order in the theory. All past
assessments of the structure of basic human values have relied on such subjectivejudgments of plots produced by an MDS approach. This paper is the first direct,
quantitative evaluation of the postulated circumplex structure of values.
2. Value theory and past assessments
The theory of basic human values (Schwartz, 1992) identifies 10 motivationally
distinct types of values that are likely to be recognized within and across cultures:power, achievement, hedonism, stimulation, self-direction, universalism, benevo-
lence, tradition, conformity, and security. Table 1, below, defines the 10 values.
The most important feature of the theory is the structure of dynamic relations
among the 10 values that it explicates. It postulates that actions expressive of any
value have practical, psychological, and social consequences that may conflict or
be compatible with the pursuit of other values. For example, actions that express he-
donism values are likely to conflict with those that express tradition values and vice
versa, and acting on self-direction values is likely to conflict with maintaining con-formity values and vice versa. On the other hand, hedonism values are compatible
with self-direction values, and tradition values are compatible with conformity val-
ues. Studies in 19 countries reveal systematic associations of many behavior, atti-
tude, and personality variables with priorities for these values (see citations in
Schwartz & Bardi, 2001). The structure of value relations explains the patterns of
these associations. When the constructs in a set vary in the degree of their similarity
and dissimilarity or conflict, as the values do, they are likely to yield a circumplex
structure (Plutchik & Conte, 1997b).The circular structure in Fig. 1 portrays the total pattern of relations among val-
ues postulated by the theory. The circular arrangement of the values represents a
motivational continuum. The closer any two values in either direction around the
circle, the more similar their underlying motivations; and the more distant any
two values, the more antagonistic their underlying motivations. The motivational
significance of items that operationalize adjacent values overlaps in part but differs
sharply from that of items that operationalize distant values. This structure is a cir-
cumplex, except for the placement of tradition outside conformity. The claim of a
Table 1
Samples included in the analyses
Set I Set II
N N
Belgium University Students 249 Australia University Students 111
Denmark School Teachers 682 Austria School Teachers 196
East Germany Adults 233 Brazil University Students 151
East Germany University Students 441 Bulgaria University Students 241
France (Nancy) University Students 214 England School Teachers 209
Ghana School Teachers 214 France (Paris) University Students 390
Hong Kong University Students 222 Ghana University Students 210
Hungary University Students 160 Hong Kong School Teachers 126
Israel School Teachers 188 Hungary School Teachers 130
Japan School Teachers 173 Israel Adults 181
Macedonia University Students 245 Israel University Students 427
Netherlands University Students 217 Japan University Students 327
Peru University Students 145 Macedonia School Teachers 206
Philippines University Students 289 Mexico School Teachers 361
Poland University Students 141 Philippines School Teachers 157
Russia School Teachers 194 Russia Adults 189
Russia in Israel Adult Immigrants 202 Russia in Ber-
lin
Adult Immigrants 181
Slovakia School Teachers 186 Slovakia University Students 233
Turkey University Students 242 Switzerland University Students 264
USA School Teachers 108 Uganda University Students 188
USA (Seattle) University Students 270 USA (Illinois) University Students 374
West Germany University Students 388 USA Nurses 259
West Germany School Teachers 148 West Germany University Students 195
5551 5306
232 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
motivational continuum is especially important for relating value priorities to other
variables. It implies that these relations take the shape of a sinusoidal curve that fol-
lows the order of the values around the circle (Schwartz, 1992). Note that the theory
postulates a circular arrangement of the 10 values, not of the items. For items, it pos-
tulates that each item correlates more highly with the set of items that measure the
same value than with the set of items that measure a different value. Thus, in tech-
nical terms, the theory assumes that the items in the value survey form 10 latent fac-
tors and only the factors relate to one another in a circular manner.Extensive research has assessed the theory in over 200 samples in more than 60
countries from every inhabited continent (representative national samples, school
teachers, university students, adolescents, samples of workers in specific occupa-
tions). Researchers examined two-dimensional projections of the relations among
value items, using MDS or Similarity Structure Analysis (SSA; Borg & Shye,
1993; Guttman, 1968). They concluded that the data largely support (a) the distinc-
tiveness of the 10 values, (b) the idea that these values are comprehensive of the ma-
jor, motivationally distinctive types of values, and (c) the ordering of valuespostulated by the circumplex structure (Fontaine, 1999; Schwartz, 1992, 1994;
Fig. 1. Theoretical model of relations among 10 motivational types of values.
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 233
These conclusions were based on visual inspection of the spatial plots, guided by apriori criteria (Schwartz, 1992; Schwartz & Sagiv, 1995). They revealed that, in the
vast majority of samples, the items that operationalize each value occupied a distinct
region in the space, with no substantial empty spaces between regions. Moreover, the
order of these distinct regions around the circle generally approximated the theorized
order shown in Fig. 1. As noted, however, visual inspections cannot formally test
whether the observed data possess a circumplex structure. Data that appear to fit
a circumplex structure may or may not be found to fit such a structure when tested
directly, as Fabrigar et al. (1997) demonstrated. Therefore, a first objective of thispaper is directly to test the circumplex structure of the theory of human values using
an application of structural equation modeling designed for this purpose. We ask:
Does confirmatory factor analysis (CFA) corroborate the theoretical structure of va-
lue relations that was supported by interpretations of past SSA and MDS analyses?
Equally important is our second objective—to test possible refinements to the va-
lue theory. These refinements were suggested in the published literature but never
tested. Each refinement was inspired by examining the separate plots of relations
among value items in large numbers of samples. The refinements fit the majorityof samples. The theory formally incorporated one refinement: It modified the strict
circumplex structure of 10 values into the structure shown in Fig. 1. This structure
places tradition and conformity values at the same polar angle around the circle,
but tradition is outside conformity. It provided a better visual fit than the strict cir-
cumplex to the average plot of the first 40 samples studied and to the separate plots
in 29 of these samples, and no worse a fit in the remaining 11 samples (Schwartz,
1992, p. 35). Although this presumed refinement has been justified conceptually, it
has never been tested formally.
234 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
3. Models tested
Below, we spell out each refinement proposed to the theory, note the conceptual
arguments for it, and then test it. One could perform these tests with the data from
past research. However, the refinements drew upon post hoc examination of thosedata. Hence, this would not test the power of the theory to account for new data.
Therefore, for totally independent, stringent tests of the theory and tentative refine-
ments, we analyze data that have not figured in any previous assessments of the
structure of values. We wish to assess which theoretical model best fits the basic pat-
tern of value relations common to cultures, rather than to identify unique variants in
particular cultures. We therefore combine the individual data from many cultures. In
order to assess the robustness of the findings, we replicate the analyses in two inde-
pendent sets of samples.
3.1. Model 1: 10 orthogonal factors
The theory of basic human values (Schwartz, 1992, 1994) derived the 10 motiva-
tionally distinct types of values from three universal requirements of the human con-
dition: needs of individuals as biological organisms, requisites of coordinated social
interaction, and survival and welfare needs of groups. The theory holds that groups
and individuals represent these requirements cognitively, as specific values aboutwhich they communicate in order to coordinate with others in pursuing the goals
that are important to them. The 10 values are the content aspect of the theory.
The structural aspect of the theory specifies relations of similarity and dissimilarity
among these values. Perhaps, however, each value is independent. This would be
in line with Rokeach�s (1973) view of values, on which Schwartz built, and with tra-
ditional exploratory factor analysis. We therefore first test how well a model of 10
orthogonal factors fits the data.
3.2. Model 2: A quasi-circumplex model of 10 factors1
At the beginning of the chapter that first presented the value theory, Schwartz
(1992, pp. 13–15) speculated that the 10 values might form a perfect circular struc-
ture. That is, the 10 values would form a circle as in Fig. 1, but with tradition located
between benevolence and conformity. The theory made no assumption as to whether
1 A comment on the terminology used with circumplex models is necessary to clarify the usage that
follows. Guttman (1954) first used the term ‘circumplex’ to include any model that postulates a circular
arrangement of relations among variables. He subdivided such models into those that assume the variables
are equally spaced on the circumference of the circle (circulant) and those that do not assume equal spacing
(quasi-circumplex). We use these terms as he did. Guttman provided no label for models that assume a
circular arrangement of variables but do not locate all the variables on the circumference of the circle.
Hence, there is no term for the ‘‘definitive’’ model proposed by the value theory. It postulates that
conformity values are more central and tradition values more peripheral (see Fig. 1). The remaining values
are arrayed in order around the circle, though not necessarily equidistant. We label this model ‘‘modified
quasi-circumplex.’’
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 235
values are spaced equally around the circle (a circulant model) or unequally (a quasi-
circumplex model). The chapter rejected this circumplex, but provided no direct,
quantitative test of the appropriate circumplex model to justify this rejection. We
therefore next test the fit of a circumplex model to the data. Because the theory does
not assume that the values are distributed at equal intervals around the circle (circ-ulant model), we test a quasi-circumplex model. We compare the fit of this model to
the fit of the model of ten orthogonal factors.
3.3. Model 3: The 1992 theory—A modified quasi-circumplex model
The ‘‘definitive’’ version of the theory of basic values postulates the modified qua-
si-circumplex structure presented in Fig. 1. As noted, this change derived from find-
ing that a theory that locates tradition outside of conformity at the same polar anglein the circle was a better visual fit to the plots of data in the available samples. Scru-
tinizing plots from many later samples reaffirmed this model (Schwartz, 1994; Sch-
wartz & Sagiv, 1995). The theoretical explanation for locating conformity and
tradition at the same polar angle in the circle is that they share the same broad mo-
tivational goal—subordinating self in favor of socially imposed expectations. What
distinguishes them is that ‘‘conformity values entail subordination to persons with
whom one is in frequent interaction—parents, teachers, and bosses. Tradition values
entail subordination to more abstract objects—religious and cultural customs andideas’’ (Schwartz, 1992, p. 40). Conformity values emphasize restraint of actions, in-
clinations, and impulses that might upset or harm others and violate their expecta-
tions. Tradition values emphasize respect, commitment, and acceptance of the
customs and ideas of one�s culture or religion.
Central versus peripheral locations in a circle typically reflect differences between
constructs in the degree of their abstractness, closeness to the self, or prevalence in
everyday interaction (Levy, 1985). On all three counts, tradition values are more
likely to be located peripherally than conformity values. The more peripheral the lo-cation of a value, the less positive its correlations with the values on the opposite side
of the circle. Hence, the peripheral location of tradition would signify that it is less
compatible than conformity with hedonism and stimulation values. We test the fit of
the ‘‘definitive,’’ modified quasi-circumplex model of values to the data and compare
its fit to that of the quasi-circumplex model. We label this model ‘‘modified quasi-cir-
cumplex’’ because it orders the values around the circle but includes central vs. pe-
ripheral positioning.
3.4. Model 4: Combining tradition and conformity to yield nine values
If tradition and conformity values share the same broad motivational goal, per-
haps a simpler, more parsimonious model would fit the data better. Tradition and
conformity may constitute a single value. This would be a quasi-circumplex model
of nine values, with the combined tradition/conformity value located between benev-
olence and security. We formally test this quasi-circumplex model and compare it
with the modified quasi-circumplex model.
236 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
3.5. Model 5: Power values peripheral to achievement values
In discussing the plots of the first 40 samples studied, Schwartz (1992, p. 40) noted
that, in 26 samples, a distinct region of power values could be drawn toward the periph-
ery of the circle behind achievement values. In most samples, however, it was also pos-sible to connect the region of power values to the center of the circle. Hence, the
‘‘definitive’’ version of the theory located power values between achievement and secu-
rity. Was this decision correct? Might a model locating achievement and power values
at the same polar angle in the circle, with power peripheral to achievement, fit the data
better? Theoretical arguments for this alternative model appear in Schwartz (1992, pp.
40–41):
[Both values] focus on social esteem. However, achievement values refer more to striving to
demonstrate competence in everyday interaction. . . whereas power values refer more to the
abstract outcomes of action in the form of status in the social structure. . . .achievement val-
ues refer to the striving of the individual. . . whereas power values also refer to the hierarchi-
cal organization of relations in society.
Thus, like tradition and conformity, power and achievement differ on character-istics that might lead to a central versus peripheral order: degree of abstractness,
closeness to the self, and prevalence in everyday interaction. We therefore test
whether this model fits the data better than the previous models. We also test
whether a more parsimonious model that treats power and achievement as a single
value (model 6) is an even better fit.
3.6. Model 7: Higher-order types of value
Based on the SSA analyses in the first 40 samples, Schwartz (1992, p. 43) sug-
gested a simpler way to view value structures. Relationships among the values can
be summarized in terms of a two-dimensional structure composed of four higher-or-
der value types. One higher-order type, called openness to change, combines stimu-
lation and self-direction values. It forms a bipolar dimension with the higher-ordertype called conservation that combines security, conformity, and tradition values.
This dimension arrays values in terms of the extent to which they motivate people
to follow their own emotional and intellectual interests in unpredictable and uncer-
tain directions (openness) versus to preserve the status quo and the certainty it pro-
vides (conservation).
A third higher-order type, called self-enhancement, combines power and achieve-
ment values. It forms a bipolar dimension with the higher-order type called self-tran-
scendence that combines universalism and benevolence values. This dimensionarrays values in terms of the extent to which they motivate people to enhance their
own personal interests even at the expense of others (self-enhancement) versus to
transcend selfish concerns and promote the welfare of others, close and distant,
and of nature (self-transcendence). Hedonism values share some elements of both
openness and self-enhancement. Consequently, hedonism is located between these
two higher-order types.
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 237
Schwartz treats the higher-order types merely as a way to describe the value struc-
ture more simply. Nevertheless, he and others sometimes use these four types, rather
than the 10 values, to predict behavior and attitudes (e.g., Bilsky, 1998; Hrubes, Aj-
zen, & Daigle, 2001; Ros, Schwartz, & Surkiss, 1999; Schwartz, 1994). If one wants
to simplify the value structure, is this particular set of higher-order types more mean-ingful than other possible groupings of values? Or do the values form a continuum,
so any combination of adjacent values (e.g., power with security) would be equally
legitimate? We address these questions by evaluating their implications for associa-
tions among the values.
Let us assume that this particular set of four higher-order types is more meaning-
ful conceptually than alternative combinations. We would then expect adjacent val-
ues within these higher-order types to be more strongly associated with one another
substantively and empirically than they are with the adjacent values from other high-er-order types. For example, the two self-transcendence values, universalism and
benevolence, should be more highly intercorrelated than universalism is with self-di-
rection (also adjacent, but in the openness higher-order type) or than benevolence is
with conformity (adjacent, but a conservation value). We put this idea to a formal
test by specifying a model in which correlations between values within higher-order
types are higher than the correlations of these values with adjacent values from other
higher-order types.
3.7. Model 8: Freely estimating hedonism correlations
The theory of value structure locates hedonism values between stimulation and
achievement values because they share elements of both openness and self-enhance-
ment. The theory does not specify whether hedonism is related more to one or to the
other higher-order type, so the above models treated it as equally linked to both.
The CFA approach enables us to ask whether hedonism is closer to openness, to
self-enhancement, or equally close to both. Model 8 addresses this question.
3.8. Models 9 and 10: Subtypes of values
Partitioning the domain of value items, like partitioning any circular structure, en-
tails an arbitrary selection of class-intervals (Plutchik & Conte, 1997b). The domain
could as easily be partitioned into more or fewer intervals, since the values constitute
a motivational continuum. Schwartz (1992) judged that 10 distinct values are suffi-
cient to capture the major motivational differences in value meanings and to createuseful indexes for measuring individuals� value priorities. These 10 values are psy-
chologically and socially meaningful and each has links to important constructs in
the literature. However, for some purposes, finer discriminations may be desirable
within particular values.
3.8.1. (9) Nature and social concern subtypes within universalism
Two potentially useful subtypes of universalism have been proposed (Sagiv, 1994;
238 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
concern,� includes the value items equality, world at peace, and social justice. This
subtype is more strongly associated than the other universalism items with concern
for and action to promote the welfare of people outside one�s in-group. The second
proposed subtype, labeled �nature,� includes the value items unity with nature, pro-
tecting the environment, and world of beauty. This subtype is more strongly associ-ated than the other universalism items with environmental attitudes and actions.
Are these merely subsets of universalism items worth distinguishing only when
one studies particular topics to which they are relevant? Or are they real subtypes
in the sense that the value items within each are more closely related to one another
than to the other universalism items? We test a model that represents them as real
subtypes that constitute components of universalism.
3.8.2. (10) Personal and group subtypes within security
In presenting the value theory, Schwartz (1992, p. 9) raised the possibility of split-
ting security into two separate types of values, one concerned with personal security
and the other with the security of larger groups. The motivational goal of the general
security value is safety, harmony, and stability of society, of relationships, and of
self. Visual examination of the plots in the first 40 samples supported the existence
of a single security value. Perhaps, security of one�s groups is a precondition for se-
curity for self. Nonetheless, the plots did suggest ‘‘that people may distinguish some-
what between the security of self and that of the collectivities of which they aremembers’’ (Schwartz, 1992, p. 41). We therefore test a model that represents per-
sonal and group subtypes as components of security.
Various additional refinements to the structure of values postulated in the value
theory are possible. However, we limit ourselves to the above models because they
cover all of the refinements suggested in the literature until now.
4. Methods
4.1. Samples
Respondents from 46 samples in 27 countries (N¼ 10,857) provided the values
data. None of these respondents had been included in assessments of the structure
of values in past reports. Table 1 describes the sample characteristics. Samples were
volunteers recruited at their place of work or study or at home. We randomly as-
signed each sample to one of the two sets, in order to replicate the analyses in eachset of samples. Set I included 5551 individuals and set II 5306 individuals.
4.2. Instrument
Respondents completed a slightly expanded, 57-item version, of the Schwartz Va-
lue Survey (SVS) in their native language. Each item is followed in parentheses by a
short explanatory phrase (e.g., EQUALITY [equal opportunity for all]). Respondents
rate the importance of each value item as a guiding principle in their life on a 9-point
Table 2
Definitions of the motivational types of values and items used as markers
Power: Social status and prestige, control or dominance over people and resources (authority, social
power, wealth, preserving my public image)
Achievement: Personal success through demonstrating competence according to social standards
(ambitious, successful, capable, influential)
Hedonism: Pleasure or sensuous gratification for oneself (pleasure, enjoying life, self-indulgent)
Stimulation: Excitement, novelty, and challenge in life (daring, a varied life, an exciting life)
Self-direction: Independent thought and action—choosing, creating, exploring (creativity, freedom,
independent, choosing own goals, curious)
Universalism: Understanding, appreciation, tolerance, and protection for the welfare of all people and
for nature (equality, social justice, wisdom, broadminded, protecting the environment, unity with
nature, a world of beauty)
Benevolence: Preservation and enhancement of the welfare of people with whom one is in frequent
personal contact (helpful, honest, forgiving, loyal, responsible)
Tradition: Respect, commitment, and acceptance of the customs and ideas that traditional culture or
religion provide (devout, respect for tradition, humble, moderate)
Conformity: Restraint of actions, inclinations, and impulses likely to upset or harm others and violate
social expectations or norms (self-discipline, politeness, honoring parents and elders, obedience)
Security: Safety, harmony, and stability of society, of relationships, and of self (family security, national
security, social order, clean, reciprocation of favors)
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 239
scale from ‘‘opposed to my principles’’ ()1), through ‘‘not important’’ (0), to ‘‘of su-
preme importance’’ (7). As markers for each of the 10 values, we used the items rec-
ommended as standard indexes for cross-cultural research, based on the consistency
of their meanings across cultures (Schwartz, 1992, 1994). Table 2 presents brief def-
initions of each value followed by its marker items. We included ‘‘self-indulgence,’’
from the expanded SVS, to provide a third item to measure hedonism.
4.3. Analyses
We designed an approach for the CFA analyses suited for testing all of the differ-
ent structural models. This approach must test quasi-circumplex models as well as
modified quasi-circumplex models. It must also test a model that specifies different
distances among particular values, to represent relations within and between high-
er-order value types. Additionally, it must test the existence of subtypes within values
(universalism and security). Finally, the approach must treat the values as latentvariables and the items as observed variables, allowing a test of the appropriateness
of the marker items for measuring the 10 values in a unified analysis.
Tracey (2000) described computer programs specifically designed to test circum-
plex conceptual models. Both of the most available programs, CIRCUM (Browne,
1992) and RANDALL (Tracey, 1997), do not fit the requirements of testing all as-
pects of the current theory and its proposed refinements. Neither program permits
specifying two-level models as required here. We cannot test the circumplex assump-
tion for the latent variables (the values) together with a simple factorial structure forthe manifest variables. This limitation, that precludes simultaneously modeling rela-
tionships of items to factors and relations among factors, is the primary reason we
do not use these programs. In addition, neither program provides an appropriate
240 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
way to handle more central as opposed to peripheral locations of factors in a circle.
Hence, they are not suited to testing the ‘‘definitive’’ model of the value theory, a
modified quasi-circumplex. We therefore developed a strategy tailored to the require-
ments of testing all aspects of the models of interest, using a structural equation
modeling program (LISREL 8.14; J€oreskog & S€orbom, 1993).We tested the model of 10 orthogonal factors with an ordinary CFA on 10 latent
variables. To test the theory-driven models of the value structure, we specified a ref-
erence matrix of expected correlations among the values for each model. This matrix
represents exactly a pattern of expected correlations among the values that repro-
duces the model. We then tested the fit of the reference matrix to the observed data.
In order to create a reference matrix for a model, it is necessary to specify actual
correlation coefficients that fit the expected pattern of associations of the model to be
tested. Any set of arbitrarily chosen coefficients that fits the pattern of expected as-sociations might be used, because the model must reproduce the pattern. Once refer-
ence values are chosen for any two correlation coefficients, all the remaining
coefficients follow from the pattern needed to reproduce the model.
We adopted a data-driven approach to determine reasonable reference coefficients.
In a preliminary analysis, we estimated a model of 10 freely intercorrelated latent fac-
tors in both samples, with LISREL. We then calculated the average intercorrelation
between all of the pairs of factors that the theory postulates to be adjacent in the circle
(e.g., self-direction/universalism; see Fig. 1). This yielded a reference correlation of .68for adjacent values. We applied the same procedure to all pairs of values that the the-
ory postulates to have opposing substantive contents (e.g., self-direction/security) and
obtained a reference correlation of .08 for values on opposite sides of the circle.
Based on these analyses, we fixed the correlation coefficient between all pairs of
adjacent values at .68 for the circumplex model of 10 equally spaced values. Thus,
we fixed the correlations for power/achievement, power/security, security/conformi-
ty, and so on around the circle, at .68 in the reference matrix. Similarly, we fixed the
correlation between all pairs of values on opposite sides of the circle (universalism/power, benevolence/achievement, etc.) at .08. Distances between pairs of values
around a circle of 10 values can range from adjacent to four �steps� away (opposite).
We therefore computed the amount to reduce the correlation for each step by divid-
ing the difference between the maximum correlation (.68) and the minimum (.08) by
four. This yielded a reduction of .15 for each step. Table 3, below the diagonal, pre-
sents the resulting reference matrix. Table 3 arranges the values according to their
postulated order around the circle from universalism to self-direction. The greater
the distance between any pair of values, the less positive the correlation betweenthem. The coefficients that reproduce the five expected distances among 10 values
around the circle are .68, .53, .38, .23, and .08.2
2 A reviewer of an earlier version of this paper remarked that a constant difference between correlation
coefficients does not produce equidistance of items along a circle. This is not a problem for the current
approach because the theory does not posit equidistance among the 10 values. Rather, it posits relations of
q1 > q2 > q3 > q4 > q5 < q6 < q7 < q8 < q9, where q1 to q9 stand for the correlations of one value with
the other nine values in the order presented in Fig. 1. The reference matrix represents this set of relations.
Table 3
Reference matrixes of expected factor intercorrelations for confirmatory factor analyses: 10 factor quasi-
circumplex (model 2) below the diagonal and basic theory modified quasi-circumplex (model 3) above the
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 241
For each subsequent model (3–8), we modified the reference matrix of expected
correlations to represent the relations among the values expected according to that
model. In the results, we explain how we constructed the reference matrix of ex-
pected correlations for each of these models. The models that postulate the existence
of subtypes (9 and 10) imply that each subtype consists of items that correlate more
highly with one another than expected from their relation to the same latent value.
That is, the correlations with the latent value of the items expected to constitute each
subtype do not account fully for the intercorrelations among the items themselves.To model this pattern, we permitted correlated errors between the items in each sub-
type. We then tested whether models with these correlated errors fit the data better
than models with no correlated errors.
Our criterion for the goodness of fit of the models is a combination of two in-
dexes, RMSEA and SRMR, recommended by Hu and Bentler (1999; cf. MacCallum
& Austin, 2000). RMSEA is especially sensitive to models with misspecified factor
covariances or latent structures, SRMR to models with misspecified factor loadings.
Hu and Bentler suggest that, for samples of 5000, the approximate size of our sam-ples, the probabilities of rejecting a valid model or of accepting an invalid model are
extremely small when RMSEA is close to .06 and SRMR to .11. We report v2 sta-
tistics to test differences in fit between nested models. Many of the models are not
nested, however. To test differences between non-nested models, we report the Ak-
aike information criterion, AIC (Akaike, 1973; Sakamoto, Ishiguro, & Kitagawa,
1986). Among a set of hypothesized, non-nested models, the best model is the one
that minimizes the value of AIC. In all cases, our aim is to select the best model
for describing the data among the theory-based alternatives.3
3 The decision whether one model fits the data better than another is sometimes difficult in this study.
With the large samples, trivial differences in v2 are statistically significant. We therefore set a conservative
probability level of p < :001 for considering Dv2 between nested models to be significant. Moreover,
differences in the fit coefficients of the models are often small. When this is so, we use the differences in the
RMSEA, SRMR, and AIC as aids to choice, picking the model that is most convincing conceptually and
that is also best according to the set of statistics.
242 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
As noted, we conducted all analyses on two independent sets of data. For both
data sets, we first calculated ordinary covariance matrixes.4 All analyses used max-
imum-likelihood estimation.
5. Results
5.1. Model 1: 10 orthogonal factors
We first tested amodel that treats each value as independent. Table 4 reports results
for this model of 10 orthogonal factors. An RMSEA index greater than .06 and an
SRMR index greater than .11 indicate that this model fit the empirical data poorly.
5.2. Model 2: A quasi-circumplex of 10 factors
The preliminary model of the value theory, suggested but then rejected by Sch-
wartz (1992), was a perfect circumplex structure of 10 factors. We tested this model
by using the reference correlation matrix below the diagonal in Table 3. The pattern
of correlations among all the values in this matrix represents exactly the pattern of
expected correlations for a circumplex structure. The second row in Table 4 presents
the fit statistics for this model. The RMSEA index is slightly higher than the recom-mended cutoff criterion in set I and just at that level in set II. The SRMR criterion is
met in both sets of samples. Thus, the 10-factor quasi-circumplex model shows a
moderate fit to the data. It clearly fits the data better than the model of 10 orthog-
onal factors. Both the v2 value and the AIC value decreased substantially in both sets
of samples. In this and all subsequent theory-driven models, all items loaded signif-
icantly (p < :0001) on the appropriate latent value.
5.3. Model 3: The 1992 theory—A modified quasi-circumplex model
The ‘‘definitive’’ representation of the value theory is the quasi-circumplex struc-
ture presented in Fig. 1 (Schwartz, 1992, 1994). It locates tradition outside of confor-
mity, but at the same polar angle in the circle. This reduces the number of locations
around the circle from 10 to 9. The pattern of expected correlations above the diag-
onal in Table 3 represents this structure. We generated this pattern by modifying the
correlations for the quasi-circumplex model (below the diagonal in Table 3) as fol-
lows. With only nine locations of values, the number of �steps� from any value tothe opposite location in either direction around the circle is three. We therefore de-
4 The samples that make up the data sets vary substantially in size. Hence, we also did the analyses on a
correlation matrix for each set that weighted each sample equally. We constructed the average matrixes as
follows. We transformed the correlations in each sample-specific correlation matrix into a Fisher-Z score,
averaged these Z-scores across the samples of the data set, and transformed the mean Z-scores back into
correlation coefficients. Analyses of these average correlation matrixes yielded results very similar to those
reported below. Because the mathematical properties of a matrix of averaged correlation coefficients are
uncertain, the text reports analyses on the unweighted covariance matrix.
Table 4
Confirmatory analyses of models of the structure of values in two sets of samples
Model df Sample set I, N ¼ 5551 Sample set II, N ¼ 5306
244 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
termined the size of a step by subtracting the minimum correlation between pairs of
values (.08) from the maximum (.68) and dividing by three, yielding steps of .20.
Thus, in the reference matrix, adjacent values are expected to correlate .68, values
separated by one other value around the circle (e.g., universalism and stimulation)
.48, those separated by two other values (e.g., universalism and hedonism) .28,and those separated by three other values (e.g., universalism and achievement) .08.
In order to represent the fact that tradition and conformity are located at the
same polar angle, we increased the expected correlation between them by one step
to .88. To represent the location of tradition toward the periphery of the circle,
we reduced the expected correlations of tradition with the values that were one,
two, and three steps away from it around the circle. The reduction had to increase
with each step without reaching the equivalent of a full step, .20. We therefore re-
duced the expected correlations of tradition by .05 for each step. The expected cor-relations of tradition with its adjacent values (benevolence and security) were
unchanged at .68. Its expected correlations with the values one step away (universal-
ism and power) were reduced by .05, compared with the expected correlations of
conformity, and fixed at .43. The expected correlations of tradition with the values
two steps away (self-direction and achievement) were reduced by .10 and fixed at
.18. Finally, its expected correlations with the values three steps away, on the oppo-
site side of the circle (hedonism and stimulation), were reduced by .15 and fixed at
).07.Row three in Table 4 presents the fit statistics for this model. The RMSEA index
is slightly higher than the recommended cutoff criterion in set I and the criterion is
met in set II. The SRMR criterion is met in both sets of samples. As compared with
the quasi-circumplex model (2), the modified quasi-circumplex model fits the empir-
ical data somewhat better. The goodness of fit indexes improved slightly in both sets
of samples, and both the AIC value and the v2 value decreased substantially both in
set I and in set II.
5.4. Model 4: Combining tradition and conformity to yield nine values
The modified quasi-circumplex model locates tradition and conformity values at
the same polar angle on the circle because they share the same broad motivational
goal. Model 4 asks whether tradition and conformity might better be treated as a sin-
gle value rather than as separate values. This model is a quasi-circumplex with nine
rather than 10 values. It combines the tradition and conformity items to form a sin-
gle value located between benevolence and security. The reference matrix of expectedcorrelations to reproduce this model is similar to that for model 3. However, tradi-
tion values are dropped from the matrix. The joint tradition/conformity value has
the same expected correlations as conformity did in model 3.
Row four in Table 4 presents the fit statistics for this model. RMSEA shows no
improvement in either set of samples compared with model 3. SRMR increases
slightly in set I and decreases slightly in set II. Both models fit the data reasonably
well. The increase in the AIC value in set I (176.5) suggests that model 3 is better, but
the decrease in the AIC value in set II (67.1) favors model 4. Thus, treating tradition
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 245
and conformity as a single value reduces the goodness of fit a little in set I and
improves it even less in set II. On empirical grounds, it is difficult to choose between
the two models. To remain consistent with the value theory and with the research
evidence that conformity and tradition have meaningfully different associations
with other variables, we retain model 3. We treat conformity and tradition as distinctvalues, as the basis for further analyses.
5.5. Model 5: Power values peripheral to achievement values
Schwartz (1992) raised the possibility that power and achievement values might be
located at the same polar angle in the circle, with power peripheral to achievement.
The proposed relationship between these two values parallels the relationship of tra-
dition and conformity. We therefore followed the same procedures used to generatethe matrix of expected correlations for model 3 in order to generate the matrix for
model 5. That is, we fixed the correlation between power and achievement at .88
and reduced the expected correlations of power with the values that are one, two,
and three steps away from it around the circle by .05, .10, and .15, respectively, as
compared with achievement. The expected correlations of power with its adjacent
values are fixed at .68 for security and hedonism. Its expected correlations with
the values one step away are fixed at .43 (stimulation and conformity) and .38 (tra-
dition). Its expected correlations with the values two steps away are fixed at .18 (be-nevolence and self-direction). Finally, its expected correlation with universalism, the
only value three steps away, is fixed at ).07.Row five in Table 4 presents the fit statistics for this model. Compared to model 3,
the RMSEA and the SRMR indexes show a poorer fit in both sets of samples. The
substantially higher AIC values in both sets indicate that model 5 describes the data
less well than model 3. We therefore reject this model and continue to use model 3,
the modified quasi-circumplex, as the basis for comparing subsequent models.
5.6. Model 6: Combining power and achievement to yield nine values
Model 6 asks whether power and achievement might better be treated as a single
value rather than as separate values. The reference matrix of expected correlations to
reproduce this model is similar to that for model 3 in Table 3, with necessary changes
to reflect reducing the number of distinct values by one.5 Row five in Table 4 indi-
cates that this model fits the observed data less well than model 3 in both sets of sam-
ples, judged by all indexes. We therefore reject this model as an improvement on themodified quasi-circumplex model.
5 The combined power/achievement value has the same expected correlations with security, tradition,
and conformity that power did in model 3, and the same expected correlations with hedonism, stimulation,
and self-direction that achievement did in model 3. In addition, the expected correlations of hedonism are
.2 higher with conformity and security and .1 higher with tradition, and the expected correlation of
stimulation with security is .2 higher.
246 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
5.7. Model 7: Higher-order types of value
Schwartz (1992, 1994) proposed that four sets of values form higher-order value
types: openness to change—joining stimulation and self-direction values; conserva-
tion—joining security, conformity, and tradition values; self-enhancement—joiningpower and achievement values; and self-transcendence—joining universalism and be-
nevolence values. The 1992 chapter treated the higher-order types as descriptive.
Some later publications, however, treated them as though the values within each
are substantively more related to one another than they are to the adjacent values
from other higher-order types. If this is so, the correlations between the values within
each higher-order type should be greater than their correlations with the adjacent
values in another higher-order type.
To generate a reference matrix of expected correlations to reproduce this model,we increased the correlations among values within the same higher-order type by .05,
from .68 to .73 (e.g., self-direction/stimulation).6 Correlations between adjacent pairs
of values from different higher-order types were left at .68 (e.g., power/security). Ac-
cording to the theory, hedonism shares elements with both openness and self-en-
hancement. To model this, we increased correlations of hedonism with the values
in these higher-order types by .025, half the increase between the values within each
higher-order type. This yielded expected correlations of hedonism with achievement
and stimulation of .705, and with self-direction and power of .505.Row seven in Table 4 presents the fit statistics for this model. The RMSEA index
is slightly higher than the recommended cutoff criterion in set I and the criterion is
met in set II. The SRMR criterion is met in both sets of samples. Thus, a quasi-cir-
cumplex structure that treats the higher-order types as more than merely descriptive
provides a reasonable fit to the empirical data. Comparing this model (7) to the mod-
ified quasi-circumplex model (3) that assumes no greater association among values
within the same higher-order type reveals no improvement of fit. RMSEA is un-
changed and SRMR decreases slightly in sample set I but increases slightly in sampleset II. However, the AIC values are larger for model 7 than for model 3 in both sam-
ple sets (set I, 170.2; set II, 138.4). This suggests that model 3 fits the observed data
better. The greater simplicity and conceptual elegance of model 3, together with the
statistics, support retaining the theory of values as forming a motivational contin-
uum that it represents rather than accepting this refinement.
5.8. Model 8: Freely estimating hedonism correlations
The modified quasi-circumplex model (3) constrained hedonism to correlate
equally with its adjacent values, achievement and stimulation. This assumed that
the motivation underlying hedonism is as much self-enhancement as openness. By
freeing the correlations of hedonism with achievement and stimulation, model 8
6 The correlation between tradition and conformity was not increased further because it had already
been increased to .88 in model 3.
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 247
evaluated this assumption. This and the subsequent models are nested, so we use Dv2
to estimate the significance of changes.
The fit statistics in row eight of Table 4 all show some improvement over model 3.
Dv2 values decreased significantly in both sample sets (Dv2 ¼ 625:2 and 396.3, 2df,
p < :001). The freely estimated correlations of hedonism in sets I and II, respectively,were .39 and .42 with achievement and .63 and .64 with stimulation. This indicates
that hedonism is closer to openness, though it also relates to self-enhancement.
Given the significant improvement in fit, we compare the remaining models with
model 8.
5.9. Model 9: Nature and social concern subtypes within universalism
This model tests the existence of two potential subtypes of universalism proposedin the literature: social concern—including the items equality, world at peace, and
social justice; and nature—including the items unity with nature, protecting the en-
vironment, and world of beauty. The existence of these subtypes implies that each
consists of items that correlate more highly with one another than expected from
their relation to the latent universalism factor. To model this pattern, we permitted
correlated errors between the items in each subtype. We then tested whether this
model fits the data better than the model with no correlated errors.
Row nine in Table 4 presents the fit statistics for this model. Compared to model8, RMSEA improves in both sets of samples and SRMR improves in set I. More-
over, v2 values are significantly lower in both sample sets (Dv2 ¼ 1577 and 1292.4,
7df, p < :001). This test supports the existence of social concern and nature subtypes
within universalism values.
5.10. Model 10: Personal and group subtypes within security
This model tests the existence of two potential subtypes of security: personal se-curity—including the items family security, reciprocation of favors, and clean; and
group security or security of the wider collectivity—including the items social order
and national security. Following the same procedure as in model 9, we modeled this
pattern by permitting correlated errors between the items in each subtype.
Row 10 in Table 4 presents the fit statistics for this model. Compared to model 9,
RMSEA is unchanged in both sets of samples, while SRMR improves slightly only
in set I. However, v2 values are significantly lower in both sample sets (Dv2 ¼ 313:7and 298.9, 4df, p < :001), indicating improved fit. These findings support, albeitweakly, discriminating the personal and group subtypes within security values.
5.11. Loadings of items on factors in the final model
Table 5 provides the loading patterns of items on the latent factors in the final
model (10) for both data sets. Not only are all loadings significant (p < :001), butall are substantial (>.40), with only one exception. �Accepting one�s portion in life�loads .26 on the tradition value factor in set I, though it too is significant
le 5
ding pattern of items on latent factors for final model (10)
em Universal-
ism
Benevo-
lence
Tradition Confor-
mity
Security Power Achieve-
ment
Hedonism Stimula-
tion
Self-
direction
SI SII SI S II SI SII SI SII SI SII SI SII SI SII SI SII SI SII SI SII
quality .43 .46
orld at peace .49 .51
nity with
nature
.40 .47
isdom .45 .41
orld of beauty .48 .54
cial justice .58 .59
roadminded .59 .43
otecting the
environment
.54 .54
oyal .57 .49
onest .66 .58
elpful .62 .63
esponsible .59 .52
orgiving .55 .49
espectful .49 .53
oderate .51 .44
umble .61 .50
ccepting one�sportion in life
.26 .30
evout .53 .48
liteness .66 .64
lf-discipline .57 .57
onoring
parents and
elders
.73 .66
248
S.H
.Schwartz,
K.Boehnke/JournalofResea
rchin
Perso
nality
38(2004)230–255
Tab
Loa
It
E
W
U
W
W
So
B
Pr
L
H
H
R
F
R
M
H
A
D
Po
Se
H
Obedient .61 .57
Social order .53 .52
National
security
.62 .59
Reciprocation of
favors
.53 .45
Family security .48 .45
Clean .67 .64
Social power
Wealth
Authority
Preserving
public image
Ambitious .62 .56
Influential .57 .57
Capable .55 .50
Successful .69 .63
Pleasure .69 .66
Enjoying life .71 .66
Self-indulgent .60 .57
Exciting life .66 .69
Varied life .62 .69
Daring .52 .53
Freedom .54 .53
Creativity .47 .53
Independent .52 .34
Choosing own
goals
.57 .45
Curious .46 .50
Note. SI is sample set I; SII is sample set II.
S.H
.Schwartz,
K.Boehnke/JournalofResea
rchin
Perso
nality
38(2004)230–255
249
.54 .54
.59 .58
.62 .65
.54 .57
250 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
p < :001. Furthermore, loadings are quite similar in the two data sets, differing by .08
or less for 43 items. Only in three cases, �broadminded�, �independent,� and �humble,�is there a difference in loadings of more than .10 between the two data sets.
6. Discussion
The current analyses are the first formal, quantitative assessment of the theory of
the structure of basic human values (Schwartz, 1992). Analyses in two independent
sets of 23 samples from 27 countries yielded very similar results. We can therefore
have considerable confidence in the conclusions these results suggest. The analyses
support the ‘‘definitive,’’ modified quasi-circumplex version of the theory and some
proposed refinements, but reject others.
6.1. The basic value structure
The discrimination of items into 10 distinct values, each defined by its motiva-
tional content, is confirmed. Each of the 46 items correlates significantly
(p < :001) with its a priori latent value factor. The SRMR statistic, which is espe-
cially sensitive to misspecified factor loadings, shows an excellent fit. Moreover,
had correlations with multiple factors been permitted, no item would have correlatedas highly with any other factor.7
The poor fit of the model of 10 orthogonal factors confirms that the 10 values are
not independent. The quasi-circumplex structure of relations among the 10 values
(model 2), proposed as a preliminary theory (Schwartz, 1992), provides a reasonable
fit to the data. Subjective judgments of the SSA plots of relations among values in 40
samples had led Schwartz (1992) to reject the simple quasi-circumplex structure. In-
stead, he proposed a modified quasi-circumplex, with tradition values peripheral to
conformity values (model 3). The current analyses support this change. Substantialimprovements in AIC and in the RMSEA and SRMR indexes in both sets of samples
provide a statistical justification for adopting the modified quasi-circumplex as the
‘‘definitive’’ model for the theory of value structure.
6.2. Proposed model refinements
6.2.1. Tradition and conformity
Because tradition and conformity values share the same broad motivationalgoal, we evaluated a simpler model that combined them into one (model 4). Com-
pared with the modified quasi-circumplex, the fit indexes are virtually the same for
this model. The AIC values indicate a better fit for model 3 in sample set I but a
better fit for model 4 in sample set II. Thus, the CFA results give no definitive an-
swer whether to retain the distinction between tradition and conformity values. A
decision therefore depends on the fruitfulness of this distinction. Because tradition
7 The relevant modification indexes are available from the authors.
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 251
and conformity values are adjacent in the value structure, their associations with
other variables should be quite similar. This is indeed the most common finding.
However, numerous differences in the associations of these two values indicate that
it is fruitful to treat them as distinct. We cite a subset of the varied findings that
support the distinction.Tradition and conformity values have significantly different correlations with
three of the ‘‘Big 5’’ traits—agreeableness, extraversion, and conscientiousness
(Roccas, Sagiv, Schwartz, & Knafo, 2002), with views of how much the govern-
ment is doing about human rights (Spini & Doise, 1998), and with indicators of
religiosity, across religions and nations (e.g., Schwartz & Huismans, 1995). Tradi-
tion values discriminate strongly among voters for different political parties in
many nations, whereas conformity values do not (Barnea, 2003; Barnea & Sch-
wartz, 1998). Parent–child value similarity is high for tradition values, among bothadolescents and young adults, but low for conformity values (Boehnke, 2001; Kna-
fo & Schwartz, 2002). Finally, in samples from 61 nations, conformity values tend
to be moderately important (typically 5th or 6th in the hierarchy of 10 values),
whereas tradition values are significantly less important (typically 8th or 9th; Sch-
wartz & Bardi, 2001).
6.3. Power and achievement
Like tradition and conformity, power and achievement differ on characteristics
that might lead to a central versus peripheral order. The peripheral location of power
(model 5) is rejected by the analyses. It yields poorer fit indexes than the modified
quasi-circumplex in both sets of samples. The simpler model that combines power
and achievement into a single value (model 6) is also rejected.
6.4. Higher-order value types
The most significant, applied modification of the value structure has been to use
only the four higher-order value types—openness to change, conservation, self-en-
hancement, and self-transcendence types—as predictors. Are these particular high-
er-order types more justified than other possible groupings of the 10 values? Use
of these particular types implies that the values that constitute each are substantively
closer to one another than to the other values to which they are adjacent in the value
circle (model 7). Schwartz (1992, 1994) attributed no substantive significance to the
particular higher-order types. He grouped the 10 values only to describe the valuestructure more simply. He maintained that the values form a motivational contin-
uum. If the 10 values do form a continuum, other groupings of adjacent values
(e.g., universalism with self-direction) would be equally legitimate for purposes of
simplifying.
The CFA analyses reveal that increasing intercorrelations among values within
higher-order types provides no improvement in fit. This supports the assumption
that the values do indeed form a motivational continuum. The continuum idea
implies that the array of value items can be partitioned into as many or as few
252 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
categories as is optimal for a researcher�s purposes. Higher-order types can be
formed, but alternative combinations of adjacent values into higher-order types
are as legitimate as the previously designated higher-order types.
Universalism and self-direction, for example, can form a higher order ‘‘intellec-
tual openness’’ type, and power and security an ‘‘uncertainty control’’ type. Thesehigher-order types predict a classic liberal vs. conservative orientation in politics,
whereas the previously specified higher-order types do not (Barnea, 2003). Benev-
olence, tradition, and conformity, for example, can form a ‘‘conventional proso-
cial’’ type that predicts the agreeableness trait of the Five Factor Model more
strongly than either the previously specified self-transcendence or conservation
higher-order types (data from Roccas et al., 2002). In sum, the support for a mo-
tivational continuum of values gives researchers the freedom and flexibility to
choose higher-order combinations of adjacent values particularly suited to the top-ics they study.
6.5. Hedonism
Hedonism values share with power and achievement values their emphasis on the
interests of self. They share with stimulation and self-direction their emphasis on
openness to change. The value theory does not specify which emphasis is stronger.
Freeing the correlations of hedonism with its adjacent values (model 8) addressedthis issue. A significant improvement in fit indexes in both sets of samples indicates
that hedonism is not equally close to self-enhancement and openness. Although it
correlates significantly with both, it is clearly closer to openness. This suggests that,
for most people, hedonism values focus more on freely experiencing pleasure and less
on pursuing pleasure competitively.
6.6. Universalism and security subtypes
The motivational goal of universalism values is to understand, appreciate, and
protect the welfare of all people and nature. The literature suggested that this broad
value includes two separable subtypes, social concern and nature. Allowing corre-
lated errors among the items in each subtype (model 9) significantly improves fit in-
dexes in both sample sets. This supports the division of universalism into subtypes.
The motivational goal of security values is safety, harmony, and stability of society,
relationships, and self. Schwartz (1992) suggested that this value too might include
separable subtypes, one focused more on self and the other on groups and society.Allowing correlated errors among the items in these subtypes (model 10) also im-
proves the fit indexes in both sample sets.
We did not test the possibility of treating the subtypes of universalism and of se-
curity as distinct values, because such a model has not been proposed in the litera-
ture. For studying particularly relevant issues, however, separate indexes for the
subtypes might be useful. For example, a nature index might be used for environ-
mental issues and a personal security index for personal safety issues (e.g., installing
burglar alarms and other home protection devices).
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 253
6.7. Reflections on the theory of value structure
One crucial assumption underlies the theory of value structure: Actions that ex-
press any value have consequences that conflict or are compatible with the pursuit
of other values. The motivational goals of some values (e.g., power and benevolence)are postulated to be inherently antagonistic, the goals of others (e.g., power and
achievement) to be inherently congruent. The total set of antagonistic and congruent
relations among the 10 values gives rise to the modified quasi-circumplex value struc-
ture (Fig. 1). Conflict or antagonism implies negative correlations between values on
opposite sides of the structure. Yet, the correlations between opposing values in the
reference matrixes for the various models ranged only from +.08 to ).07. Does this
contradict the assumption of opposition between conflicting values?
As noted above, some people tend to rate all values relatively high or low, regard-less of content. This biases observed intercorrelations among values upward. Stan-
dardizing each participant�s responses eliminates this bias, though it introduces
other problems.8 Freely estimated correlations among the latent factors for the 10
values, based on the standardized data, provide a rough assessment of the true de-
gree of opposition. The correlations between pairs of values that the theory describes
as antagonistic ranged from ).49 to ).81, averaging ).72. Thus, these data support
the assumption of opposition between conflicting values that is central to the value
theory.Future theorizing about the value structure might consider and test two issues
that our analyses indirectly raised. First, do the 10 values differ in the breadth of
their content? Schwartz (1992) suggested that universalism and security are concep-
tually broad, including subtypes that were confirmed here. Are other values espe-
cially narrow conceptually (e.g., hedonism and stimulation)? The method
employed here could be used to model and test theory-based specifications of differ-
ences in the conceptual breadth of the values. For that purpose, however, it would be
desirable to index each value with an equal number of items.The second issue concerns the central-peripheral distinction found with confor-
mity and tradition values. Do values vary on this general dimension in addition to
the dimension of motivational content? This dimension was interpreted as indicating
variation in the closeness of a value to the self, its involvement in everyday interac-
tion, and its specificity vs. abstractness. Centrality may also signify greater normative
importance (Melech, 2001). The centrality of value items in the space correlates pos-
itively with their importance ratings. Systematic theorizing that considers the loca-
tions of value items on this potential dimension might enrich the value theory andpoint to hypotheses testable with the methods used here.
In conclusion, it is worth noting the overlap between the inferences supported by
different methods. The theory of the structure of values was refined by examining
8 We did not standardize in the main analyses for two reasons. First, this would impose the same
variance on everyone�s responses. But people differ greatly in the degree to which they discriminate among
their values. Hence standardizing would distort responses. Second, standardizing exacerbates problems
with missing data.
254 S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255
subjective judgments of SSA plots in many samples. Such judgments can be mislead-
ing. In this case, however, the key judgments were supported by the statistical tests.
This was so even though the theoretical model makes fine distinctions and the tests
were performed on new data. Studies using multidimensional scaling methods led to
the conclusion that the value structure is robust to respondents� gender, age, or levelof education (Schwartz, 1992, 1994; Struch, Schwartz, & van der Kloot, 2002). The
fact that the CFAs confirm the value structure inferred from earlier visual inspec-
tions increases our confidence in this conclusion. The success of these subjective
judgments demonstrates that visual inspection of spatial representations can reveal
reliable, theoretically meaningful relations among variables. It is most successful
when guided by a clear theory and when replicated in many samples.
Acknowledgments
This research was supported by Grant No. 94-00063 from the United States–Israel
Binational Science Foundation (BSF) and by a grant from theNational Science Foun-
dation (Israel Academy of Sciences) to the first author, and by Grant I-242-065.04/92
from theGerman-Israeli Foundation for Scientific Research andDevelopment to both
authors, and was facilitated by the Leon and Clara Sznajderman Chair of Psychology.
References
Akaike, H. (1973). Information theory and an extension of the maximum likelihood principle. In B. N.
Petrov, & N. Csaki (Eds.), Second international symposium on information theory (pp. 267–281).
Budapest: Akademiai Kiado.
Barnea, M. (2003). Relations of personal values to political orientations across cultures. Doctoral
Dissertation. The Hebrew University of Jerusalem, Israel.
Barnea, M., & Schwartz, S. H. (1998). Values and voting. Political Psychology, 19, 17–40.
Bilsky, W. (1998). Values and motives. Paper presented at the International Research Workshop ‘‘Values:
Psychological Structure, Behavioral Outcomes, and Inter-Generational Transmission.’’ Maale-
Hachamisha, Israel.
Boehnke, K. (2001). Parent-offspring value transmission in a societal context: Suggestions for a utopian
research design with empirical underpinnings. Journal of Cross-Cultural Psychology, 32, 241–255.
Borg, I., & Shye, S. (1993). Facet theory: The method and its application. Newbury Park, CA: Sage.
Browne, M. W. (1992). Circumplex models for correlation matrices. Psychometrika, 57, 469–497.
Fabrigar, L. R., Visser, P. S., & Browne, M. W. (1997). Conceptual and methodological issues in testing
the circumplex structure of data in personality and social psychology. Personality and Social
Psychology Review, 1, 184–203.
Fontaine, J. R. J. (1999). Cultural bias in Schwartz’ value instrument: An exemplary study on cultural bias in
social psychological and personality questionnaires. Unpublished doctoral dissertation, Katholieke
Universiteit Leuven, Belgium.
Guttman, L. (1954). A new approach to factor analysis: The radex. In P. F. Lazarsfeld (Ed.),
Mathematical thinking in the social sciences (pp. 258–348). New York: Columbia University Press.
Guttman, L. (1968). A general nonmetric technique for finding the smallest coordinate space for a
configuration of points. Psychometrika, 33, 469–506.
Hrubes, D., Ajzen, I., & Daigle, J. (2001). Predicting hunting intentions and behavior: An application of
the theory of planned behavior. Leisure Sciences, 23, 165–178.
S.H. Schwartz, K. Boehnke / Journal of Research in Personality 38 (2004) 230–255 255
Hu, L.-t., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis:
Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55.
J€oreskog, K., & S€orbom, D. (1993). LISREL 8: A guide to the program and applications. Chicago: SPSS.
Knafo, A., & Schwartz, S. H. (2002). Accounting for parent–child value congruence: perception, acceptance,
and beyond. The Hebrew University of Jerusalem, Israel, submitted manuscript.
Levy, S. (1985). Lawful roles of facets in social theories. In D. Canter (Ed.), The facet approach to social
research (pp. 59–96). New York: Springer-Verlag.
MacCallum, R. C., & Austin, J. T. (2000). Applications of structural equation modeling in psychological
research. Annual Review of Psychology, 51, 201–226.
Melech, G. (2001). Value development in adolescence. Unpublished doctoral dissertation, The Hebrew
University of Jerusalem, Israel.
Plutchik, R., & Conte, H. R. (Eds.). (1997a). Circumplex models of personality and emotions. Washington,
DC: American Psychological Association.
Plutchik, R., & Conte, H. R. (1997b). Epilogue: The future of the circumplex. In R. Plutchik, & H. R.
Conte (Eds.), Circumplex models of personality and emotions. Washington, DC: American Psycholog-
ical Association.
Roccas, S., Sagiv, L., Schwartz, S. H., & Knafo, A. (2002). The Big Five personality factors and personal
values. Personality and Social Psychology Bulletin, 28, 789–801.
Rohan, M. J. (2000). A rose by any name? The values construct. Personality and Social Psychology Review,
4, 255–277.
Rokeach, M. (1973). The nature of human values. New York: Free Press.
Ros, M., Schwartz, S. H., & Surkiss, S. (1999). Basic Individual values, work values, and the meaning of
work. Applied Psychology: An International Review, 48, 49–71.
Sagiv, L. (1994, July). Cross-cultural differences in the components and meaning of universalism values.
Paper presented at the XII International Conference of the International Association for Cross-
Cultural Psychology, Pamplona, Spain.
Sakamoto, Y., Ishiguro, M., & Kitagawa, G. (1986). Akaike information criterion statistics. Dordrecht,
Netherlands: Reidel.
Schwartz, S. H. (1992). Universals in the content and structure of values: Theory and empirical tests in 20
countries. In M. Zanna (Ed.), Advances in experimental social psychology (Vol. 25, pp. 1–65). New
York: Academic Press.
Schwartz, S. H. (1994). Are there universal aspects in the content and structure of values? Journal of Social
Issues, 50, 19–45.
Schwartz, S. H. (1997). When are universalism values particularistic? Paper presented at the International
Colloquium: Universalism vs. Particularism Toward the 21st Century, Prague, Czech Republic.
Schwartz, S. H. (in press). Basic human values: Their content and structure across countries. In A. Tamayo
& J. Porto (Eds.), Valores e trabalho [Values and work]. Brasilia: Editora Universidade de Brasilia.
Schwartz, S. H., & Bardi, A. (2001). Value hierarchies across cultures: Taking a similarities perspective.
Journal of Cross-Cultural Psychology., 32, 268–290.
Schwartz, S. H., & Huismans, S. (1995). Value priorities and religiosity in four Western religions. Social
Psychology Quarterly, 58, 88–107.
Schwartz, S. H., & Sagiv, L. (1995). Identifying culture specifics in the content and structure of values.
Journal of Cross-Cultural Psychology, 26, 92–116.
Schwartz, S. H., Sagiv, L., & Boehnke, K. (2000). Worries and values. Journal of Personality, 68, 309–346.
Spini, D., & Doise, W. (1998). Organizing principles of involvement in human rights and their social
anchoring in value priorities. European Journal of Social Psychology, 28, 603–622.
Struch, N., Schwartz, S. H., & van der Kloot, W. A. (2002). Meanings of basic values for women and men:
A cross-cultural analysis. Personality and Social Psychology Bulletin, 28, 16–28.
Tracey, T. J. G. (1997). RANDALL: A Microsoft FORTRAN program for the randomization test of
hypothesized order relations. Educational and Psychological Measurement, 57, 164–168.
Tracey, T. J. G. (2000). Analysis of circumplex models. In H. E. A. Tinsley, & S. D. Brown (Eds.),
Handbook of applied multivariate statistics and mathematical modeling (pp. 641–664). New York: