Properties of the JEID scale

The JEID scale was designed to measure the construct subjective evaluation of the objective earnings distribution. In order to (a) link justice evaluations of the earnings distribution to the actual distribution of earnings and (b) make evaluations comparable across individuals, the scale provides contextual information on average gross earnings and typical occupations. Respondents are asked whether they find low, middle, upper-middle, high, and top incomes unfairly low, fair, or unfairly high. This operationalization is characterized by two crucial choices: First, it acknowledges that respondents in general population surveys (i.e., laypersons) struggle to think in terms of distributions and attempts to capture complex perceptions of inequality are subject to substantive method effects (Eriksson & Simson, 2013; Heiserman & Simpson, 2021; Ziano et al., 2022). Instead, we argue that it is more intuitive for respondents to focus on specific points in a distribution (e.g., what the earnings of the worst-off look like). Accordingly, respondents are presented with information about the earnings distribution in the form of specific income groups with accompanying information of where in the distribution a specific income group is placed (e.g., what is the share of people who earn more or less than them). Second, the proposed JEID scale focuses on income from (dependent) work. We opted for this narrow income concept because earnings are the main source of household income both in the UK and in Germany and capture an important element of economic inequality.

This approach of providing context information on actual earnings and occupations when assessing respondents’ sense of justice was first introduced in the employee survey Legitimation of Inequality over the Life-Span (LINOS; Adriaans et al., 2019) and as part of the European Social Survey (ESS) Round 9 module “Justice and Fairness in Europe” (ESS, 2018). However, these first empirical forays were limited to isolated evaluations of earnings at the very bottom and the very top of the earnings distribution. Furthermore, in the case of the ESS, example occupations were not included, and respondents were thus provided with less context information. Providing example occupations is, however, highly relevant because individuals differentiate between legitimate and illegitimate inequalities, and occupations carry information on ability, training, and status, all of which are central in determining justice evaluations of earnings (Jasso et al., 2016; Sauer et al., 2016).

With the JEID scale, we developed this novel approach further to allow respondents to state their personal sense of justice based on the same objective facts, thus ensuring the comparability across individuals of the perceptions of the earnings distribution that underlie these justice evaluations. By extending measurement to a broad earnings spectrum, it is possible to identify the critical segments of the earnings distribution at which justice deficits/gaps (Verwiebe & Wegener, 2000) are perceived. Additionally, the JEID scale is based on the justice evaluation function, which is widely used in the empirical justice literature. As analogous measurements of reflexive justice evaluations are available, the JEID scale allows the comparison/contrast of the evaluation of the earnings across the income distribution (i.e., the rewards of others) with the evaluation of own rewards.

The aim of the present studies was threefold: (a) to develop and validate the German-language JEID scale based on three quota samples representing the heterogeneity of the adult population in Germany in terms of age, sex, and educational attainment; (b) to adapt the validated German-language scale to the English language; and (c) to validate the English-language version of the JEID scale and to compare the scale properties of both language versions.[1]

Item generation and selection

The JEID scale comprises five newly developed items that together depict a broad range of earnings—from the lower end, through the middle range, to the upper end of the earnings distribution. To define five earnings categories, we used information on the average earnings of five income groups (gross monthly earnings in euros/pounds), representing the 10th, 50th, 80th, 90th, and 99th percentile of the earnings distribution within the population of Germany/UK. We opted for these five percentiles to capture income groups across the distribution that are clearly distinguishable for respondents and at the same restrict the number of items to the JEID scale allowing for integration into survey programs that face strict time restrictions.

Each income group is described based on three typical example occupations within that group. The occupational examples are typical in the sense that they are most prevalent in the respective percentile. Moreover, in order to eliminate potential bias introduced through gender stereotypes, we chose the occupations such that one example occupation was male-dominated, one female-dominated, and one gender-mixed. All contextual information refers to full-time employees (at least 35 hours/week) and is based on the—at the time of conducting the studies—latest versions of the German Socio-Economic Panel (SOEP; Version 34; Goebel et al., 2019) from 2017 for the German-language version and on the UK Household Longitudinal Study “Understanding Society” (Wave 9; University of Essex, Institute for Social and Economic Research, 2019) from 2017/2018 for the English-language version. Respondents were asked to indicate whether they perceived the earnings of each specific group to be unfairly low, fair, or unfairly high.

We tested different response scale lengths across the first two studies to achieve appropriately differentiated responses while at the same time minimizing cognitive load. We tested a 7-point response scale in Study 1, and a 9-point response scale in Study 2. The leftmost (lowest) scale point was labeled unfairly low, the middle scale point was labeled fair, the rightmost (highest) scale point was labeled unfairly high. The other scale points had no verbal labels. In Study 1 and Study 2, half of the participants (approx. 50% per quota) received a response scale with numerical labels in addition to verbal labels, and the other half received a response scale without numerical labels. In the first condition, numerical labels ranging from –4/–5 to +4/+5 were shown below the verbal labels, with the scale midpoint labeled as 0. There were no statistically significant differences in average responses between the condition without numerical labels and the condition with numerical labels—either for the five items individually or for the overall scale mean. Hence, it was possible to pool responses from the two groups for the main analyses (see Figure 1). In our subsequent analyses, we coded the scale points such that the scale ranged from 1 (ungerecht niedrig/unfairly low), through 4/5 (gerecht/fair), to 7/9 (ungerecht hoch/unfairly high).

Figure 1

Response Scale Format by Condition and Study

a

b

Note. a: Study 1. b: Study 2.

Before the items and the response scale were administered in Study 1, a preliminary version underwent cognitive pretesting to optimize them and to assess the length of the response scale, the need for numerical labels, and the amount of and necessity for context-related information in the instruction and the items (for more detailed information, see Hadler et al., 2019). During the scale development process, we optimized and slightly adapted the instruction and items as well as the response scale. In Study 3, we used the final JEID scale with a 7-point response scale without numerical labels (see Figure 2).

Figure 2

Response Scale Format of the Final JEID Scale Used in Study 3

The German-language JEID scale was translated into English using the TRAPD (Translation, Review, Adjudication, Pretesting, and Documentation) approach (Harkness, 2003). First, one scale expert (English native speaker) and one professional translator (non-native English speaker) translated the instructions and item wordings and the response scale labels of the JEID scale into English independently of each other. Second, an adjudication meeting was held in which the two translators, a psychological expert, and an expert in questionnaire translation reviewed the translation proposals and developed the final translation. The translation procedure for the JEID scale differed slightly from the full TRAPD approach because one of the initial translations was not provided by a professional translator and the other was not provided by an English native speaker.

Samples

To develop the JEID scale and investigate its psychometric properties, we assessed JEID in three Web-based surveys (using computer-assisted self-administered interviewing [CASI]) that we conducted via the online access panel provider respondi AG. Data collection took place in Germany in January/February 2020 (Study 1) and in August 2020 (Study 2), and parallel in Germany and the UK in July 2021 (Study 3). We drew quota samples that represented the heterogeneity of the adult population in terms of age, sex, and educational attainment. Data from the last German Census (2011) were used as a reference for Studies 1 and 2 (https://ergebnisse.zensus2011.de/?locale=en). Data from the German Microcensus from 2018 were used as a reference for Study 3 (https://www-genesis.destatis.de/genesis/online?locale=en). To avoid bias introduced by a lack of reading/language proficiency, only native speakers were recruited. We explained our research goal (investigation of the quality of several questionnaires) to the participants. Respondents consented to their participation in an anonymous online survey and received a small financial incentive for their participation; no sensitive material was collected; and no special ethical review and approval was required under German law. We adhered to ethical standards comparable to the 1964 Declaration of Helsinki.

To assess test–retest reliability, we invited a subsample in all three studies to participate in a follow-up survey 2 weeks after the median day of the main survey. Only respondents who completed the full questionnaire (i.e., respondents who did not abort the survey prematurely) were included in the analyses. The gross sample sizes were N_Study 1 = 531, N_Study 2 = 618, N_{Study 3, Germany}= 463, and N_{Study 3, UK} = 483. After excluding invalid cases, the net sample sizes were N_Study 1 = 486 (retest: N_Study 1 = 189), N_Study 2 = 618 (retest: N_Study 2 = 299), and N_{Study 3, Germany} = 420 (retest: N_{Study 3, Germany} = 202), and N_{Study 3, UK} = 440 (retest: N_{Study 3, UK}  = 199). Table 2 shows in detail the sample characteristics and their distribution.

Table 2

Sample Characteristics by Study

Note. The German educational attainment levels were as follows: low = no educational qualification/basic school-leaving qualification (ohne Bildungsabschluss/Hauptschulabschluss); intermediate = intermediate school-leaving qualification (Mittlere Reife/Realschulabschluss); high = entrance qualification for a university of applied sciences/general higher education entrance qualification (Fachhochschulreife/Abitur). The UK educational attainment levels were as follows: low = never went to school, Skills for Life/1–4 GCSEs A*–C or equivalent; intermediate = 5 or more GCSEs A*–C/vocational GCSE/GNVQ intermediate or equivalent; high = 2 or more A-levels or equivalent.

Material

Value orientations

As proposed in the Theory section, just rewards in a given situation are guided by the principles of distributive justice, namely, equality, equity, need, and entitlement. Research suggests that people (and societies) differ with regard to which of these distributive rules they believe should guide the allocation of goods and burdens within a society (Arts & Gelissen, 2001; Hülle et al., 2018; Wegener & Liebig, 1995). Accordingly, we expected that such normative orientations coincide with the evaluation of existing inequalities. Preference for normative principles that suggest limiting inequality (such as a preference for the allocation of goods within a society based on the principle of need or equality) would be associated with perceptions of more severe injustice with regard to the extremes of the earnings distribution. By contrast, we expected normative orientations that provide justification for inequality based on the status quo or on individual differences in contributions (such as the distributive principle of entitlement or equity) to be associated with perceptions of less severe injustice. We used two measures of value orientations to test these assumptions.

As a first measure, we used the four subdimensions of the Basic Social Justice Orientation (BSJO) scale (both language versions: Hülle et al., 2018)—namely, need, equity, equality, and entitlement—to assess agreement with each of the five distributive principles. The BSJO scale comprises a total of 12 items (4 items for each subdimension) to be answered on a 5-point rating scale. For the analyses, we coded responses to range from 1 (strongly disagree) to 5 (strongly agree). The scale has satisfactory psychometric properties (Hülle et al., 2018).

In addition to the assessment of normative preferences (BSJO scale), we used the Left–Right Self-Placement scale (both language versions: Breyer, 2015) to operationalize political preferences. The scale assesses political orientation on a 10-point response scale ranging from 1 (left) to 10 (right); it is frequently used in social surveys such as the ESS and the German General Social Survey (ALLBUS). In line with the general stance on income inequality associated with the political left and right, we assumed that respondents who tended toward the left would evaluate the observed earnings distribution as more unjust than those who tended toward the right.

Sociodemographic characteristics

Justice theory suggests that non-reflexive justice evaluations are not strictly impartial¾that is, they are not completely independent of the individual’s own situation (Jasso et al., 2019; Lengfeld, 2007). For example, women and individuals with a high level of education have been shown to hold more egalitarian views (Hülle et al., 2018; Shorrocks & Grasso, 2020). There is mixed evidence with regard to age: Whereas Hülle et al. (2018) reported that younger respondents expressed more egalitarian views, Forsé and Parodi (2009) found the opposite tendency. Following the assumptions of relative deprivation theory (Merton & Merton, 1968; Runicman, 1966; Stouffer et al., 1949), individuals’ own positions in the earnings distribution may provide an important frame of reference (Hadler, 2005), with those at the bottom of the distribution holding more negative evaluations of the existing inequalities and those at the top of the distribution holding more positive evaluations.

Accordingly, we analyzed age, sex, educational attainment, and gross income as antecedents of the justice evaluation of the earnings distribution. Sex was operationalized as male (1) and female (2). Age was surveyed in years.

To measure educational attainment, we slightly adapted a question used in the ESS (2016) and asked for respondents’ highest level of education. After combining similar educational groups, the following categories remained: no educational qualification/basic school-leaving qualification (1), intermediate school-leaving qualification (2), and entrance qualification for a university of applied sciences/general higher education entrance qualification (3).

To assess gross income, we combined and adapted two questions, one of which was used in the ESS (2016) and the other in ALLBUS (2018). The adapted open-ended item read as follows: “How high is your own gross monthly income (employee: the amount before deductions for tax and social security; self-employed: average gross monthly income before deductions for overhead)?”

Social desirability

Income inequality features prominently in public discourse, and its growth and alleged detrimental effects are the subject of regular media coverage (Diermeier et al., 2017). Moreover, political parties and other institutions place the reduction of income inequalities prominently on their agendas (e.g., Goal 10 of the United Nations 2030 Agenda for Sustainable Development, “Reduce inequality within and among countries”). This may suggest that a negative attitude toward income inequality is seen as socially desirable. Therefore, we investigated a possible distortion of respondents’ item responses by social desirability.

We examined the susceptibility of JEID to two aspects of socially desirable responding (exaggerating positive qualities and minimizing negative qualities) with the Social Desirability–Gamma Short Scale (KSE-G; German-language version: Kemper et al., 2014; English-language version: Nießen et al., 2019). KSE-G comprises six items to be answered on a 5-point rating scale ranging from 1 (does not apply at all) to 5 (applies completely). To ensure that high values represented high levels of socially desirable responding, we recoded the items of the subdimension “minimizing negative qualities.” KSE-G has satisfactory psychometric properties (Nießen et al., 2019) and has been used, for example, in ALLBUS.

Item parameters

We took a look at the distribution of answers by study. Table 3 shows the descriptive statistics (i.e., means, standard deviations, skewness, and excess) for the five items as well as for the scale mean, separately for Studies 1, 2, and 3 (Germany and the UK). The distributions of justice evaluations for the five income groups from Study 3 (Germany and the UK) are additionally depicted in Figure 3.

As the results show, the distributions of answers to the five items and the scale means were comparable across all studies and countries: On average, evaluations tended most toward “unfairly low” for the incomes of low-income earners and “unfairly high” for the incomes of top-income earners. Evaluations generally differed substantially between income groups, with the exception that, in the German samples, the incomes of upper-middle- and high-income earners were rated as almost equally (un)fair. The variance within the evaluation of each item ranged between 1.15 < SD < 1.59 (Study 1 and Study 3 [Germany], 7-point scale), 1.19 < SD < 1.49 (Study 3 [UK], 7-point scale), and 1.59 < SD < 2.06 (Study 2, 9-point scale).

The average evaluation of the incomes of low- and middle-income earners was below the midscale point (i.e., tending toward “unfairly low”), and the average evaluation of the income of the other three income groups was above the midscale point (i.e., tending toward “unfairly high”).

Table 3

Descriptive Statistics by Study for the JEID Items

Note. P = percentile. The rating scale ranged from 1 (unfairly low) to 7 (unfairly high) in Study 1 (N = 486) and Study 3 (Germany: N = 420, UK: N = 440), and from 1 (unfairly low) to 9 (unfairly high) in Study 2 (N = 618).

We investigated psychometric properties of the JEID scale—namely, its objectivity, reliability, and validity. These properties can be used to assess the quality of a scale, and the documentation of an instrument should provide appropriate information about which quality criteria are met.

Objectivity

A scale is objective if three conditions are met: objectivity of application (the scale works independently of the administrator), objectivity of evaluation (the scale works independently of the evaluator of the instrument), and objectivity of interpretation (unambiguous and user-independent rules are provided). The JEID scale ensures (a) objectivity of application through the standardized questionnaire format and written instructions; (b) objectivity of evaluation through the fixed scoring rules and labeled response categories; and (c) objectivity of interpretation through the reference ranges presented in Table 11.

Reliability

As an estimate for the reliability of the JEID scale, we computed the test–retest stability, r_tt, over a period of about 2 weeks in all studies. Because the test–retest stability is sensitive not only to measurement error but also to state fluctuations in the attribute in question (Allen & Yen, 2002), test–retest stability is best understood as a lower-bound estimate of reliability. The test–retest correlations are displayed in Table 4.

The test–retest reliability for the five individual items (i.e., income groups) of the JEID scale was moderate for Germany and the UK. However, because the five items belong together, the reliability of the entire scale should also be considered. Therefore, we also analyzed the total reliability across all items using a within-person profile similarity (i.e., the correlation of each person’s response patterns across the two measurement occasions), and we aggregated this correlation across the entire samples. The results showed that the 7-point response scale (Studies 1 and 3: r = .73; UK: r = .71) proved to be more reliable over time than the 9-point response scale (Study 2: r = .66) and sufficient for research purposes (Aiken & Groth-Marnat, 2006). Overall, the profile similarity results indicated that the response patterns across all items were quite stable between test and retest.

Table 4

Reliability Estimates for the JEID Items by Study

Note. CI = confidence interval; P = percentile.

Validity

Validity evidence based on the relationship between scores on the JEID scale and on scales measuring other variables

We examined evidence based on the relationship between scores on the items of the JEID scale and on scales measuring other variables described in the Material section. As we computed this kind of evidence based on manifest (scale) scores, the reported correlations (see Table 5) are subject to attenuation and represent the lower bound of the true associations. The interpretation of the correlation coefficients is based on the meta-analytic effect size guidelines for individual differences proposed by Gignac and Szodorai (2016): relatively small effects (r ≥ .10), typical (medium) effects (r ≥ .20), and relatively large effects (r ≥ 30). According to these guidelines, a correlation of r = .19 corresponds to the 50th percentile of a meta-analytical distribution of correlations in individual differences research. For this reason, we highlight medium to large effects in bold in Table 5.

Although the JEID response scale is not linear, and it ranges from “unfairly low,” through “fair,” to “unfairly high,” we can still interpret the linear correlations in a meaningful way. This is because—as can be seen from the item distributions (see the Item parameters section)—for each item, almost all individuals fell either between “unfairly low” and the midpoint (“fair”) or between the midpoint and “unfairly high.” Thus, it is always clear what a positive/negative correlation means—namely, a shift toward the midpoint or away from the midpoint.

Table 5

Correlations of the JEID Items With Other Relevant Measures by Study

Note. P = percentile. Coefficients with r ≥ |.20| are in bold type. Study 1—N = 486 (N_{Basic social justice orientation} = 485; N_{Left–right placement} = 425; N_{Gross income} = 393; N_{Trust in government} = 461). Study 2—N = 618 (N_{Basic social justice orientation} = 414; N_{Left–right placement} = 544; N_{Gross income} = 470; N_{Trust in government} = 389; N_{Social desirability} = 417). Study 3: Germany—N = 420 (N_{Left–right placement} = 234; N_{Gross income} = 342); UK—N = 440 (N_{Left–right placement} = 211; N_{Gross income} = 320).

^a For these variables, only two thirds of the sample in Study 2 were surveyed due to the planned-missingness design.

^b For these variables, only two thirds of the sample in Study 3 (Germany and the UK) were surveyed due to the planned-missingness design.

^c For these variables, each respondent received only two thirds of items within one scale and subscale in the UK sample due to the planned-missingness design. Therefore, the reported correlations may underestimate or overestimate the true associations.

JEID correlated with basic social justice orientations and political orientations. We observed the strongest correlations for three of the four principles of distributive justice measured with the BSJO scale: Respondents who tended to prefer principles according to which benefits and burdens should be distributed equally (equality principle) or based on individual needs (need principle) were more likely to evaluate the income of low- and middle-income earners as “unfairly low” and the income of (high- and) top-income earners as “unfairly high.” In other words, they evaluated inequality as more unjust. By contrast, respondents who showed stronger support for the entitlement principle, according to which the distribution of benefits and burdens should be based on status, tended to evaluate the income of low- and middle-income earners more positively (i.e., as fairer). These respondents also perceived lower levels of overreward for the income of high- and top-income earners, thus showing an overall higher tolerance for inequality.

In addition, the further respondents moved from left to right on the political orientation scale, the less they tended to rate the extremes of the earnings distribution as unfair. This association, which was particularly strong in the UK, is also in line with evidence showing that liberals hold less system-justifying attitudes than conservatives (Jost et al., 2008).

Turning to sociodemographic characteristics, there were no differences in justice evaluations of the earnings distribution by sex for any of the five items of the JEID scale in the German samples. By contrast, in the UK sample, we found that women were more likely to evaluate low to upper-middle incomes more negatively, tending toward “unfairly low.” The same was true of older respondents—especially in the German samples—who tended to rate low (and middle) incomes as “unfairly low.” This is in line with studies that have found women and older people to hold more egalitarian views (Forsé & Parodi, 2009). There were no consistent associations between JEID and educational attainment and income across the studies. This might indicate that individuals’ justice evaluations of the earnings distribution do not reflect their own positions in the income hierarchy, but rather are shaped by normative preferences as to how goods and burdens should be allocated within a society.

Furthermore, JEID showed small-to-large correlations with both subdimensions of social desirability responding in Germany (in the UK, there were associations only with minimizing negative qualities). Exaggerating positive qualities depicts the self-deceptive enhancement component of communion-induced socially desirable responding, whereas minimizing negative qualities depicts the impression-management component (Nießen et al., 2019). Both subdimensions were related to evaluations of the incomes of low- and middle-income earners as “unfairly low” and evaluations of the incomes of top-income earners as “unfairly high,” indicating that evaluating inequality as unjust is seen as socially desirable in Germany and the UK.

Internal structure: Types of justice evaluations

Because the JEID scale comprises five items for the evaluation of five income groups that should not only be considered individually but also together, we examined the internal structure (i.e., different types of justice evaluations) of JEID. We used two methods to determine whether justice evaluations can be divided into different response types: (a) latent profile analysis (LPA) and (b) k-means clustering (MacQueen, 1967). In both methods, groups are formed based on the response patterns across the five items. LPA aims to identify latent subpopulations within a population at a categorical level, whereas k-means clustering aims to partition a sample into different clusters at a manifest level. LPA provides slightly clearer quantitative criteria for choosing the correct number of profiles. By contrast, k-means clustering requires the use of different techniques to extract the optimal number of clusters: the elbow method, the average silhouette method (Rousseeuw, 1987), and the gap statistic method (Tibshirani et al., 2001). The rationale for performing both LPA and the k-means clustering was twofold: first, to find clusters that ideally are independent of the chosen method; second, for user-friendliness, because some users will prefer LPA and others k-means clustering.

The evidence based on the internal structure of JEID suggest three distinct response profiles in Studies 1 and 2. The small groups (i.e., profile shares) identified in solutions with a higher number of profiles can be assigned to the three larger profiles because they differ only in the intercept and slope, not in the profile. The results of the LPAs can be found in Table 6, Table 7, Table 8, and Table 9, separately for the three studies. Given the similar pattern for Studies 1 and 2, we conducted LPAs in a confirmatory fashion in Study 3, testing only up to three profile solutions by means of multigroup LPA models with measurement equivalence across countries in all items and equal latent profile probabilities across countries. As the results show, the three-profile solution from Studies 1 and 2 could be replicated in Study 3 with similar response patterns.

The k-means clustering showed a similar pattern: Table 10 displays the results for these three techniques by study. Using the elbow method, the average silhouette method, and the gap statistic method, three groups belonged to the optimal number of clusters among all three approaches in Study 2. We extracted the same three groups for all three studies. Figure 4 depicts the average answers of each extracted justice evaluation group for the five income groups separately for Study 1 (Figure 4a), Study 2 (Figure 4b), and Study 3 (Germany: Figure 4c, UK: Figure 4d). The figures show that the extracted clusters are comparable across all studies.

Table 6

Results of the Latent Profile Analysis in Study 1

Note. BIC = Bayesian information criterion, LMR-LRT = Lo-Mendell-Rubin adjusted likelihood ratio test, B-LRT = Parametric bootstrap likelihood ratio test. In the case of BIC and entropy, the best (i.e., lowest [BIC] and highest [entropy]) values are in bold type. In the case of class proportions, the worst (i.e., < .05) values are in bold type. N = 486.

Table 7

Results of the Latent Profile Analysis in Study 2

Note. BIC = Bayesian information criterion, LMR-LRT = Lo-Mendell-Rubin adjusted likelihood ratio test, B-LRT = Parametric bootstrap likelihood ratio test. In the case of BIC and entropy, the best (i.e., lowest [BIC] and highest [entropy]) values are in bold type. In the case of class proportions, the worst (i.e., < .05) values are in bold type. N = 618.

Table 8

Results of the Multigroup Latent Profile Analysis in Study 3

Note. BIC = Bayesian information criterion, LMR-LRT = Lo-Mendell-Rubin adjusted likelihood ratio test, B-LRT = Parametric bootstrap likelihood ratio test. In the case of BIC and entropy, the best (i.e., lowest [BIC] and highest [entropy]) values are in bold type. N = 860 (N_Germany = 420, N_UK = 440).

Table 9

Sample Mean Distribution of the JEID Items Resulting From Latent Profile Analysis, by Profile Number and Study

Note. N_{Study 1} = 489; N_{Study 2} = 618; N_Study 3 = 860 (N_Germany  420, N_UK = 440).