Theory-Driven Versus Lexical Approaches to Value Structures

The recent article by De Raad and colleagues (2016), “Values in a Cross-Cultural Triangle: A Comparison of Value Taxonomies in the Netherlands, Austria, and Spain,” set itself an ambitious task. It sought to attain both optimal coverage of the culturally unique values in these three countries and to find the structuring of the value domain that they share. This is the latest step in the authors’ impressive program that began with the lexical study of value vocabularies in single countries. Here, the authors largely frame their lexical approach by contrast with the theory-driven approach of Schwartz (1992). They point to presumed problems with the latter approach that the lexical approach might overcome. I discuss these problems, point to links between the authors’ findings and the Schwartz theory “toward universals in values” (hereafter TUV), assess the authors’ success in reaching some of their goals, and pose a challenge for future steps that I hope they will undertake.

The authors correctly note that the lexical approach has the potential to identify values that are over- or underrepresented in a theory-driven approach. They bring the example of “security” as a value overrepresented in the Schwartz taxonomy. What is surprising to me, on the contrary, is the absence of a security value in the components the authors derive. Security does not appear in the labels of any of their value components or facets. The only word referring directly to security loads somewhat incongruously under “competence-self-realization” as the 43rd item in Table 6. Security values express the biologically grounded motivation to protect the self and the socially grounded motivation to preserve one’s social groups. Concern for security trumps other motivations when people feel even moderate physical or economic threat (Neuberg, Kenrick, & Schaller, 2011). It is prominent in the value rhetoric of right-wing politicians. These facts illustrate the centrality of security as a value. Its absence in the authors’ lexical analyses, rather than its importance in the Schwartz taxonomy, is the question worth addressing.

As an example of a value underrepresented in the Schwartz taxonomy, they bring “aesthetic.” The aesthetic domain of values is indeed missing from the Schwartz taxonomy, except in “world of beauty” in the Schwartz Value Survey (SVS). That item was part of the universalism value, which makes some sense conceptually. But should we consider this underrepresentation? Interestingly, when researchers in 20 countries were asked to add values that were missing from the SVS, before they gathered data with it, not one mentioned aesthetics. Moreover, even in the lexical studies of values I could find, a factor (actually a component) partly concerned with aesthetics emerged only in one (De Raad & Van Oudenhoven, 2008). In the current analyses, there was no component or facet that emphasized aesthetics in the structure of any of the three language groups. The word “aesthetics” did load on one facet, but not a facet focused on an aesthetics domain. It loaded, instead, on the “comfort” facet of a component labeled “status and prestige,” together with such words as comfort, respect, consumption, and sexuality (Table 6). Apparently, pursuit of aesthetics is not a salient motivating value for many people, regrettable as that may be.

As the authors note, theory-driven approaches omit many value terms that the lexical approach identifies. Indeed, the instruments developed to assess and apply the TUV have limited sets of items. But that is not their purpose. The goal of the TUV is not to produce a taxonomy of value terms. Rather, its contribution is to reveal the structure of the value domain—the postulate that values are organized on a circular motivational continuum that reflects the compatibility or conflict between the motivations that the values express. According to the theory, all the different human values that the lexical approach might identify as value terms across countries are located in the value circle. People who speak different languages and come from different cultures will label the points or segments in the circle differently and may even identify different points. They may also make more or less fine-tuned distinctions in various parts of the circle.

The authors describe the TUV as a fixed framework that cannot tap into the specifics of a culture. This confuses the theory with the choices about how to partition the circle into segments (4/10/19 segments depending on objectives), what points (items) to select, and how to name them. Partitioning is necessary to facilitate using values in prediction and explanation. The theoretical framework limits neither the set of values that may be identified nor their relations with one another. It only postulates that whatever values a group identifies, they will be organized in a circle by their motivational compatibilities and conflicts. The specific order of values that the TUV currently considers the best description of the circular value continuum emerged from empirical studies with different value lists and methods across nearly 100 cultural contexts. As Schwartz et al. (2012) discuss, this order continues to evolve. The choice of methods and items as well as historical change seem to affect the order, though only at the margins.

A two-dimensional circle does a reasonable job of capturing the continuum of values of the TUV. In past work, the current authors also arrayed their value structures on a circular plane. The authors note that the “circumplexes” they had found in their Dutch and Spanish value systems differed from the circle Schwartz typically finds. In contrast, Borg, Dobewall, and Aavik (2016) found the value circle even with data collected with the Estonian Value Inventory, an instrument based on a lexical approach that is unrelated to the value circle theory. The extent to which the order of motivations differs around the various circles is a matter of debate. Whatever the differences, their sources deserve in-depth study. Differences in methods of measurement and in item sampling may contribute. Also worth considering is the fact that it is probably more accurate to conceptualize the value space as a sphere. Viewing the sphere from different angles (above, one side, or the other, etc.) would reveal somewhat different two-dimensional arrays of the values. Different angles would produce some rearrangement of the distances among items and might produce some variation in the circular order.

The five-factor model of personality has shown that it is possible for the lexical approach to identify relatively unique and distinctive constructs that exhibit consistency across cultures. The five personality factors each include six coherent facets that load almost exclusively on their own factor. Interestingly, the authors’ lexical approach to values also suggests distinguishing five constructs. However, the authors used principal components analysis (PCA) rather than factor analysis and derived components rather than factors. As components, they are formative indexes consisting of value items that may measure somewhat different things. They are therefore more difficult to interpret clearly. The facets of each of the five components derived in their simultaneous components analysis (SCA) load to some extent on every one of the other four components. Thus, each labeled value component is a conceptually complex composite. Moreover, each one relates in complex ways to the others.

Consider relations of the SCA components to the value circle. The key proposition of the TUV, noted above, is that the universe of values forms a motivational continuum that can be represented by a circle (Schwartz, 1992, 2015). All value items fall somewhere in this circle in a location that reflects the congruence and conflict between the motivations they express and the motivations that all other value items express. This should apply to the value items in a lexically derived taxonomy no less than to items selected on theoretical grounds.

The emergence of the SCA components at subsequent levels of extraction (Figure 4) suggests that some components may be related in ways implied by the motivational circle of basic values. The seventh level distinguishes six values that tap contents quite similar to six of the basic TUV values: 7/2 similar to power, 7/3 to benevolence, 7/4 to tradition, 7/5 to achievement, 7/6 to self-direction, and 7/7 to stimulation. Tracing the emergence of these values up the levels reveals that the counterparts of self-direction, stimulation, and achievement all split off from 4/1 and those of benevolence and tradition split off from 5/3. The fact that the basic values in each of these two sets are adjacent in the value circle ¹ suggests that the components that resemble them are motivationally close.

Next, consider the facets of SCA Components I and III (Table 6). Component I consists of four facets: One is similar to the basic value of benevolence (the care and support facet), another is an interpersonal form of pleasure that also largely expresses benevolence, and two facets mainly express universalism (trust and understanding and communication). Benevolence and universalism, the counterparts of the facets of Component I, are adjacent in the motivational circle. The facets of Component III largely express benevolence (benevolence facet), a mix of universalism and benevolence (awareness), a mix of tradition and conformity (dedication), and conformity (virtue). The value counterparts of these facets are adjacent in the value circle, ordered from universalism to benevolence to tradition to conformity. These four values constitute the higher order self-transcendence and conservation values. What makes these values and the parallel components motivationally compatible is the shared goal of maintaining positive social relationships. Conspicuously missing from the SCA components is the security aspect of conservation values discussed earlier.

When the authors drop all the items necessary to find a shared set of items to use in assessing a common structure, most of the unique lexicon for each country is lost. Hence, this approach does not escape the problem that characterizes etic, or theory-driven approaches—imposing a set of constructs on the values universe of respondents. This is not a criticism of what the authors have done. It is just to note that in the end, the lexical approach does not free them from doing what they criticize in more etic approaches. We all must add etic to emic approaches when we want to find shared dimensions on which to compare across cultural groups.

The authors compare the three language groups (countries) on the five PCA components they measured with the same 139 items in each country (Table 5). They hesitate to compare the country differences they observe with those found by researchers using possibly related dimensions, noting that these other dimensions have different emphases. Several of these dimensions are, nonetheless, conceptually fairly close. Comparing the order the authors observe for the three countries with their order on cultural value dimensions that overlap somewhat with each component may shed some light on the validity of the components.

The order of countries on the authors’ component of self-direction (autonomy, dialogue, individualism) is uncorrelated with their order on Schwartz’s (2006) intellectual autonomy and correlates negatively with their order on Hofstede’s (2001) individualism and on Inglehart and Baker’s (2000) self-expression. The country order on virtue (sincerity, care, respect) correlates perfectly with the order on Schwartz’s embeddedness but is unrelated to the order on Hofstede’s collectivism. The order on pleasure (enjoyment, vitality, cheerfulness) correlates perfectly with the order on Schwartz’s affective autonomy. The order on spirituality (religion, chastity, tradition) correlates negatively with the order on Inglehart and Baker’s tradition. Finally, the order on pride and reputation (richness wealth, achievement) correlates negatively with the order on Schwartz’s hierarchy.

These comparisons increase our confidence in the validity of the pleasure and virtue PCA components, but not of the other three components. It is worrisome, however, that the pleasure and virtue PCA components did not emerge as distinctive constructs in the larger analysis. No single SCA component accounted for even 30% of the variance in the pleasure and virtue components (Table 7, top panel).

The authors suggest that their lexical approach is a way to ensure coverage of the value domain and to avoid blindness to the culture-specific characteristics of value systems. These are worthy goals. Does PCA that seeks the shared structure common to value taxonomies in different languages bring us closer to these goals? Regarding coverage, none of the single final components accounted for as much as 50% of the variance in five of the 15 unique values in the three languages (Table 7, country panels). The variance in these unique values was spread across several components and some was entirely lost, so their content was not discernible when looking at the five components. With the addition of more languages, this approach would yield a set of components whose labels and interpretation would make it even more difficult to discern the even greater number of culture specifics.

The goals of maintaining coverage of culture specifics and of deriving a set of coherent values from the lexicons of multiple languages may contradict one another. PCA can retain coverage and derive a shared kernel of values. However, the components of this shared kernel do not form well-defined, coherent factors. Rather, they combine disparate elements.

The reason most theorists, practitioners, and laypeople are interested in values is to help understand and explain individual and group attitudes, opinions, cognitions, and behaviors. Each component derived here could be used to build scores for a set of coherent values that might be used for this purpose (cf. Renner, 2003). The scores for such components would be based on the items that load substantially on each component, probably the items from its strongest facet. The conceptual underpinning of such scores would be closer to the latent variable thinking of factor analysis than to the formative variable thinking of principle components analysis. Moreover, by ignoring the other items in the component, this approach would severely curtail coverage of the original lexicons and lose much of what is culture specific. A challenge for the authors is to demonstrate that the value components they derive can preserve broad coverage of culture specifics while adding to the predictive and explanatory power of currently popular value approaches.