Human values refer to what is important to people in their lives and the goals they strive to attain. According to Schwartz (1992), values “(1) are concepts or beliefs, (2) pertain to desirable end states or behaviors, (3) transcend specific situations, (4) guide selection or evaluation of behavior and events, and (5) are ordered by relative importance” (p. 4). Values differ based on the specific goals they express and motivate people to pursue. Schwartz (1992) specified ten different types of values, which are recognized across cultures, which are operationalized in the Human Values Scale: conformity, tradition, benevolence, universalism, self-direction, stimulation, hedonism, achievement, power, and security. Table 3 presents the ten values, the motivational goal each expresses, and examples drawn from the items.

Table 3

Goals and examples of the 10 human values

Note. Retrieved from Schwartz, 2003a (shortened).

The ten human values relate to one another in the circular motivational structure depicted in Figure 1. This structure reflects the conflict or compatibility among the values experienced when one pursues them. Power and achievement are adjacent values in the circle because the goals whose pursuit they motivate (personal dominance and success) can usually be attained simultaneously through the same actions. Power and universalism are opposed in the circle because the goals whose pursuit they motivate (personal dominance and protecting others’ welfare) conflict and can rarely be attained through the same actions. The closer any two values in the circle, the more compatible their underlying motivations, the more distant, the more antagonistic. Two dimensions, shown in Figure 1, can be used to summarize the value relations: Self-enhancement values (achievement, power) oppose self-transcendence values (universalism, benevolence) and conservation values (security, conformity, tradition) oppose openness to change values (self-direction, stimulation). Hedonism shares elements of both openness and self-enhancement but is usually closer to the former. Figure 1 presents the order of values in the initial value theory and commonly observed with the Schwartz Value Scale (Schwartz, 1992). The Human Values Scale of the ESS and the Portrait Values Questionnaires on which it is based, more frequently yield an order in which tradition is adjacent to benevolence and conformity is adjacent to security (e.g. Bilsky, Janik, & Schwartz, 2011).

Figure 1. Theoretical model of relations among the ten human values, retrieved from Schwartz (1992, 2003a).

Item generation and item selection

In the ESS, the Human Values scale has been used in every round since its beginning in 2002. It was included in the ESS because values are supposed to predict many of the other variables assessed in the ESS concerning attitudes, opinions and behavior. They can also reflect fundamental social changes in societies and across countries. The items of the ESS Human Values Scale are based on the Schwartz Value Survey (SVS; Schwartz, 1992) and the Portrait Values Questionnaire (PVQ; Schwartz, 2003). The ESS board representing each country allocated 21 items to measure the ten values, three for universalism and two for the remaining values. The Human Values Scale is administered in separate male and female versions that differ only in the gender prepositions. To perfect the scale, a pilot study with two samples (UK and Netherlands) was conducted (N = 444). The items and the answer format in the ESS have remained the same over the years. For further information about the item selection process, see Schwartz (2003a).

Samples

The findings reported below are based on the samples in the sixth wave of the European Social Survey (ESS) in 2012. Participants were randomly selected from all persons aged 15 and over resident within private households, regardless of their nationality, citizenship, language or legal status, in the following countries: Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Hungary, Ireland, Italy, Lithuania, Netherlands, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, United Kingdom (European Union countries), Albania, Iceland, Israel, Kosovo, Norway, Switzerland, Russian Federation, Ukraine (non-European Union countries).

The sampling frames differed across countries. Some countries provided registers of residents or households that are available for social research. If no registers were available, multi-stage sample designs were usually applied, in which primary sampling units (PSUs) were selected at the first stage. At the second stage, households or addresses within the PSUs were selected. The final sample consists of N = 54,673 people from 29 countries, including N = 25,214 male and N = 29,442 female respondents. The average age is 47.08 years (SD = 18.43).

For the present analyses, the answers categories „Refusal“, „Don’t know“ or „No answer“ were classified as missing values, because these categories express neither approval nor disapproval and can thus be interpreted as missing statements. For the following analyses, design weights were applied to correct for different sampling probabilities in the countries. See Table 4 for sample size, age, gender and educational level for each country.

Table 4

Sample size (N), gender (%), age (M, SD) and educational level (% based on ISCED) for each country and across all countries (total)

Note. Observations were weighted based on design weights. The educational level data do not sum to 100% because the categories „Not possible to harmonise into ES-ISCED”, “other” and missing values are not reported. ^*1 = ISCED 1-2: lower secondary education or less, 2 = ISCED 3-4: upper secondary education, 3 = ISCED 5-7: advanced vocational or sub-degree and tertiary education, BA or MA level.

Dimensionality

We used a multi-group structural equation model to investigate the dimensionality of the scale. Parameters were estimated using the maximum likelihood robust estimation in Mplus in order to enable weighting by design weights.

First, we investigated a tau-congeneric model with different factor loadings which shows a good model fit (χ² (4,176) = 28,363.341, p < .001; RMSEA = .056; CFI = .892). Additionally, a more restrictive, tau-equivalent model with equal factor loadings was investigated, which, according to Cheung and Rensvold (2002), does not lead to a significant change of the model fit (χ² (4,495) = 30,497.972, p < .001; RMSEA = .056; CFI = .884). As the model fit is appropriate in both cases, we used the most economic, tau-equivalent model to estimate the reliability.[1] The model structure of the tau-equivalent model with standardized regression weights (for Germany) is presented in Figure 2. For reasons of better clarity, the related correlation coefficients are presented in Table 5.

Figure 2. t-equivalent measurement model for the Human Values Scale with standardized path coefficients for Germany. Observations (raw scores) were weighted based on design weights. There are correlations between all latent variables (see Table 5). CON = Conformity, TRA = Tradition, BEN = Benevolence, UNI = Universalism, SD = Self-Direction, STI = Stimulation, HED = Hedonism, ACH = Achievement, POW = Power, SEC = Security, N = 2,839 (Germany), RMSEA = 0.01, CFI = 0.89, c²(4176) = 36746.818, p = .000.

Table 5

Correlation coefficients of the measurement model for the Human Values Scale for Germany

Note. Observations (raw scores) were weighted based on design weights. SEC = Security, CON = Conformity, TRA = Tradition, BEN = Benevolence, UNI = Universalism, SD = Self-Direction, STI = Stimulation, HED = Hedonism, ACH = Achievement, POW = Power, N = 2,958 (Germany), all correlations are significant (p < .001), except SD-CON (p = .08), SD-TRA (p = .06), HED-CON (p = .12), HED-TRA (p = .63), POW-BEN (p =.01) and SEC-SD (p = .02).

Item parameters

The means and standard deviations (raw scores) of the items for the Human values scale for Germany and across all countries are shown in Table 6. Items have been recoded, so that higher scores represent greater agreement that the person in the item is like oneself. To compare value priorities across countries, centered values should be used. We provide raw and centered scores of all values for each country in the appendix.

Table 6

Means and standard deviations of the manifest items for Germany and across all countries (total)

Note. Observations (raw scores) were weighted based on design weights, recoded scale from 1 = “Not like me at all” to 6 = “Very much like me”, N = 2,828 (Germany), N = 50,560 (total).

Objectivity

For the Human Values Scale, objectivity of application is ensured by the standardized questionnaire format, written instructions and structured interview methods called CAPI (Computer Assisted Personal Interview) and PAPI (Paper And Pencil Interview). These interviews are personally conducted by specially trained and selected interviewers. The scale is objectively rated by using fixed categories. Furthermore, as a norming sample is available for the Human Values scale (see descriptive statistics), high objectivity of interpretation is achieved.

Reliability

The reliability for the Human Values Scale has been estimated for each of the ten values. As the underlying models have the same factor loadings, they are tau-equivalent. Therefore, Cronbach’s alpha provides an appropriate indicator for reliability (Cortina, 1993). Reliability scores for Germany and across all countries (total) are presented in Table 7. See the appendix for separate reliability scores for all countries. Many of the observed reliabilities are low by conventional standards. This reflects two aspects. First, each index is based only on two items (three for universalism). Second, items were selected to maximize coverage of the diverse conceptual complexity of each value construct rather than to maximize homogeneity and capture only a narrowly defined sub-part of the construct (cf. Saris, Knoppen, & Schwartz, 2012). Despite the low reliabilities, the value indexes have shown substantial validity (see below for examples). For further analyses, all ESS data sets are available at http://www.europeansocialsurvey.org.

Table 7

Cronbach’s Alpha for the Human Values Scale in Germany and across all countries (total)

Note. Observations (raw scores) were weighted based on design weights, N = 2,958 (Germany), N = 54,673 (Total). Reliabilities range from α = .15 (Tradition, Denmark) to α = .85 (Hedonism, Bulgaria) between all countries.

Validity

To ensure content validity, a construct should be defined in advance and all aspects of the construct should be considered. The grounding of the Human Values Scale (HVS) in a unifying theory of human motivation provides the basis for its content validity. The theory specifies ten cross-culturally recognized value orientations and defines each one in terms of the distinct motivational goals it expresses. The formulations of the items for each value derive directly from the definitions of the value they are intended to operationalize.

Construct validity is fulfilled when a scale measures the construct in a way that it is consistent with existing theories and definitions. The values theory provides a strong basis for assessing the construct validity of the scale. As noted in the theory section, the values theory proposes a specific circular structure of dynamic relations among the different values. This structure has implications both for the correlations among the values themselves and for the pattern of correlation between the set of 10 values and other variables.

Consider first the expected pattern of intercorrelations among the values (see Table 5). The theory implies that there should be especially strong intercorrelations between the values within each of the “higher order” values (the poles of the two descriptive dimensions) Consistent with this expectation, the intercorrelations within all four higher order values meet Cohen’s (1992) rule of thumb for strong correlations. Intercorrelations are >.50 for the self-enhancement values (achievement – power: r = .98), the self-transcendence values (benevolence – universalism: r = .83), the openness to change values (self-direction – stimulation: r = .70), and the conservation values (conformity – tradition: r = .92, conformity – security: r = .78 , security – tradition: r =  .78). Also, as expected, hedonism correlates strongly with both openness (r = .77) and self-enhancement (r = .58), more so with the former with which it is conceptually closer.

The theory further implies that correlations between theoretically opposed higher order values should be close to zero or negative. This is the case for the values of conservation vs. openness (r = -.36 for self-direction with conformity to r = -.07 for stimulation with tradition) and for the values of self-transcendence vs. self-enhancement (r = -.20 for power – universalism, r = .09 for power – benevolence, r = .10 for achievement – universalism, r = .28 for achievement – benevolence). All of these correlations between values are much weaker than the intercorrelations within the higher order values, as expected by theory. Schmidt, Bamberg, Davidov, Herrmann, and Schwartz (2007) reported a similar set of theory consistent value intercorrelations in the data from the first round of the ESS 2002.

The circular motivational structure of the theory implies that correlations of values with other variables should follow a sinusoid pattern. That is, starting with the most positively correlated value (e.g., stimulation with age), correlations with the other values should be sequentially weaker as one goes around the circle in both directions. The least positive or most negative correlation should be with a value on the opposite side of the circle (e.g., security, conformity, or tradition). For assessing the validity of the HVS, the expected sinusoid pattern makes it possible to assess the overall pattern of associations for all ten values. We investigated this aspect of the construct validity of the HVS by correlating its ten values with a sampling of variables available in round six of the ESS. We present each variable and discuss its expected and observed pattern of correlations with values in turn.

Years of education: Educational experiences presumably promote the intellectual openness, flexibility, and breadth of perspective essential for self-direction values (Kohn & Schooler 1983). These same experiences increase the openness to non-routine ideas and activity central to stimulation values. In contrast, these experiences challenge unquestioning acceptance of prevailing norms, expectations, and traditions, thereby undermining conformity and tradition values. The increasing competencies to cope with life that people acquire through education may also reduce the importance of security values. Row 1 of Table 8 reveals the expected positive correlations of years of formal education with self-direction and stimulation values and negative correlations with conformity, tradition, and security values. The more positive association with education as compared to stimulation and hedonism values is a small deviation from the sinusoid pattern. Because associations with values generally follow the order of the circle (Schwartz, 2006), deviations are especially interesting (Schwartz, 1996). This deviant correlation might reflect increased competiveness and concern for approval acquired through schooling because of the constant grading and comparing of performance in schools. It might also be due to self-selection into continuing higher education as a function of achievement values.

Age: As people grow older, they tend to become more embedded in social networks, more committed to habitual patterns, and less exposed to arousing and exciting changes and challenges (Glen, 1974). This implies that conservation values (tradition, conformity, security) should increase with age and openness to change values (self-direction, stimulation, hedonism) decrease. Once people enter families of procreation and attain stable positions in the occupational world, they tend to become less preoccupied with their own strivings and more concerned with the welfare of others (Veroff, Reuman, & Feld, 1984). This implies that self-transcendence values (benevolence, universalism) increase with age and self-enhancement values (power, achievement) decrease.[2] Row 2 of Table 8 shows just this pattern. The values that correlate most positively and most negatively with age are tradition and stimulation. The only deviation from a perfect sinusoid curve is the higher correlation of universalism vs. benevolence, a finding worth further investigation.

Table 8

Correlations of values with relevant variables

Note. Pearson correlation coefficients based on recoded and centered values for each country, observations were weighted based on design weights, SEC = Security, CON = Conformity, TRA = Tradition, BEN = Benevolence, UNI = Universalism, SD = Self-Direction, STI = Stimulation, HED = Hedonism, ACH = Achievement, POW = Power. All correlations greater than .02 are significant, p < .001.

Religiosity: Based on sociological, psychological, and theological analyses of Western religion, Schwartz and Huismans (1995) developed hypotheses regarding relations of religiosity to values. They noted that religion typically supports the prevailing social structure and norms; it thereby encourages accepting the social order and discourages questioning and innovation. Religion also serves to reduce uncertainty and it emphasizes feelings of respect and humility while opposing the pursuit of material goods and sensual pleasure. On these bases, they hypothesized that religiosity correlates positively with conservation values, especially with tradition, and negatively with openness to change values, especially with hedonism. They hypothesized moderately negative correlations with self-enhancement values because they serve self-interests but also help to maintain the social order. They hypothesized moderately positive correlations for benevolence because selflessness is compatible with religiosity, but near-zero relations for universalism because its emphasis on accepting diversity conflicts with the in-group focus of Western religions. Together, these hypotheses predict a sinusoid curve of values-religiosity correlations, with tradition most positive and hedonism most negative. Empirical research has consistently supported these hypotheses (summarized in Roccas & Elster, 2012 and in Saroglou, Delpierre, & Dernelle, 2004). Row 3 of Table 8 presents correlations between values and responses to the one item question „How religious are you, if at all?“ The correlations fit the expected sinusoid curve.

Work for a voluntary or charitable organization: We used the following item: “In the past 12 months, how often did you get involved in work for voluntary or charitable organisations?” Responses ranged from “at least once a week” to “less than once every six months” (reverse scored). Participation in voluntary or charitable activities expresses concern for others. These may be members of the in-group, implying a positive association with benevolence values, or beyond the in-group, implying a positive association with universalism values. In contrast, such activity entails sacrificing one’s own resources, implying a negative association with power values, and involvement outside the home where control over the situation is limited, implying a negative association with security values. Relations with the other values are less predictable without knowing the activities and settings the item brings to mind for respondents. Although the correlations in the fourth row of Table 8 are small, they largely fit the expected pattern, and all predicted correlations are significant at p<.001.

Acceptance of immigrants: We compiled a two item index of willingness to accept immigrants (α = .88). The items were: “Now, using this card, to what extent do you think [country] should allow people of a different race or ethnic group as most [country]’s people to come and live here?” and “How about people from the poorer countries outside Europe?” Responses ranged from 1 – “allow many” to 4 – “allow none” (reverse scored).

The value theory suggests that the trade-off between giving high priority to promoting the welfare of all others (universalism values) and avoiding personal, national, and interpersonal threat (security and conformity values) should influence readiness to accept immigrants who differ from one’s fellow citizens (Schwartz, 2007). The different life-style of “other" immigrants is also likely to threaten those for whom maintaining traditional ways of doing things is important. On the other hand, those high in stimulation values may be more favorable to “other” immigrants because they introduce variety into society. The pattern of correlations in the last row of Table 8 conforms to these expectations. Universalism has the most positive correlations and the correlations become weaker going around the circle in both directions, with the most negative correlation for security. The only deviation from the sinusoid pattern is the higher positive correlation for stimulation which was anticipated by theory.

The full set of findings presented above provides considerable support for the construct validity of the HVS. The intercorrelations among the values and the patterns of their association with demographic, attitudinal, and behavioral variables speak to the validity of the Human Values Scale.

Descriptive statistics (scaling)

Table 9 presents reference means and standard deviation norms for the Human Values Scale in Germany and for the total sample (across all countries). Reference norms for all other countries appear in the appendix. The values for skewness range from .00 (power) to .92 (benevolence). The values for kurtosis range between -.67 (stimulation) and 1.10 (benevolence). According to Miles & Shevlin (2001), deviations from normality ≤ 1 can be neglected. As all values for skewness are ≤ 1 and this is true for kurtosis except for the small deviation for benevolence, the assumption of a normal distribution can be accepted. This is confirmed by a comparison with the normal probability plots.

For further analyses, all ESS data files including descriptions of the selected variables can be downloaded in SPSS, Stata, SAS and other formats at http://www.europeansocialsurvey.org/.

Table 9

Reference norms for the Human Values Scale for Germany and across all countries (total) from ESS round 6

Note. Observations (raw and centered scores) were weighted based on design weights and items were recoded before the analysis, N = 2,958 (Germany), N = 54,673 (total).

Measurement invariance

We evaluated three levels of measurement invariance between countries: (1) configural invariance indicates that the factor structure (and hence the structure of the construct) is equivalent between countries, (2) metric invariance indicates that the items’ loadings (and thus their meanings) are equivalent across countries, and (3) scalar invariance indicates that the items’ intercepts (and thus items’ difficulties) are equivalent across countries.

Measurement invariance was investigated by multigroup structural equation modeling. To test configural invariance, all items were constrained to load on the same factors (see Figure 2) but the factor loadings and the items’ intercepts were allowed to differ between countries. To test metric invariance, all items were constrained to load on the same factors and the factor loadings were constrained to be invariant across countries, whereas the variances of the latent variables and the items’ intercepts were allowed to differ between countries. To test scalar invariance, all items were constrained to load on the same factors and the factor loadings as well as the items’ intercepts were constrained to be invariant across countries, whereas the means and variances of the latent variables were allowed to differ between countries. As suggested by Chen (2007) and Cheung and Rensvold (2002), we used the change in CFI and the change in RMSEA as decision criteria whereby a ΔCFI > .010 and a ΔRMSEA > .015 between two invariance levels suggest that the higher level of invariance should be rejected. Results are shown in Table 10 below.

Table 10

Measurement invariance for the Human Values across all 29 countries

Note. Observations (raw scores) were weighted based on design weights, N = 54,275, ***p ≤ .001.

As can be seen, the change in CFI as well as in the RMSEA suggests metric invariance. For scalar invariance, the changes in CFI as well as in RMSEA are significant (ΔCFI = .08, ΔRMSEA = .02). Therefore, only metric invariance can be accepted for the Human values scale. The model has the same structure and factor loadings in all countries. However, intercepts can differ between countries. Thus, there can be item specific difficulties between countries. Comparisons of means can therefore be made within a country, but not between several countries.

These results are in accordance with several studies that suggest violations of cross-country measurement invariance for the Portrait Values Questionnaire PVQ-21 (e.g., Davidov, 2008, 2010; Davidov, Schmidt, & Schwartz, 2008). Therefore, additional versions of the Human Values Scales have been developed during the last years. In 2012, Schwartz et al. proposed a revised human values scale, based on his refined value theory, with 19 values and 57 items (PVQ-5X). Some values were more narrowly defined (e.g. subtypes of security), whereas other values were completely new (e.g. face). A study by Cieciuch, Davidov, Vecchione, Beierlein, and Schwartz (2014), found that the cross-country invariance properties of the PVQ-5X are substantially better than those measured with the PVQ-21. This experimental scale has been further refined as the PVQ-RR (Schwartz et al., 2012). The PVQ-RR can be recommended for comparing values of different countries.

 [2] We base these predictions on the effects of aging and life stage, ignoring possible cohort and period effects, to simplify. Effects of the latter should also fit the circular pattern.