A Matter of Values: On the Link Between Economic Performance and Schwartz Human Values

1.1 Values

In a recent and widely cited paper, Hoff and Stiglitz (2016) argue in favour of viewing the economic actor as an enculturated actor, i.e. making decisions according to worldviews rather than preferences.^[1] We formalise the idea of enculturation using the Schwartz (Schwartz 1992, 1994, 2012) theory of individual human values, and considering these values as primary explanatory variables for economic activity in our model. Schwartz and his collaborators (Schwartz 1992, 1994, 2012) postulate that there are ten human values that are important for individual behaviour across all cultures. The Schwartz theory of values makes two central propositions, supported by empirical observations (Cieciuch et al. 2014; Schwartz and Boehnke 2004). Firstly, there is a structure of relations among values, which has the form of a circular continuum. This implies that the values that are close to each other are related to compatible motivations, whereas those values that lie on opposite sides of the circle are related to conflicting motivations. Secondly, individuals have a hierarchy of values. For each person, some values are more important than others. This determines how a person resolves potential conflicts between different Schwartz values. The individual values may be clustered into additional dimensions. One way to summarise the individual values is to define four higher order values (HVO) grouped by two kinds of two oppositional dimensions: self-transcendence opposed by self-enhancement and openness to change opposed by conservation, Schwartz (1992). By basing our approach on Schwartz’s core values, we also benefit from their measurable, innumerable and universal nature (Table 1 and Figure 1).

1.2 Group Activity Selection Problems

In the classical group activity selection problem, the goal is to assign a group of agents to different sets of activities, see (Darmann et al. 2018, 2017; Lee and Williams 2017). An agent can be assigned to one task at most. The preferences of the agents may depend on different factors. (Darmann et al. 2018) consider the number of agents participating in a given activity as an example of such a factor. Ganian, Ordyniak, and Rahul (2018) consider agent heterogeneity by including agents of different types in the model. A similar problem, though in less abstract terms, is also considered in the economic literature. Rosen (1978) considers the problem of optimal allocation of workers having different skills to different tasks. The output depends on the set of workers having different skills available to complete a given task. The group activity selection problem is seen as a combinatorial problem in this field of study. The global optimum is sought from the perspective of a central planner. By contrast, in our model, despite the similarities, we allow the agents to self-organise. Moreover, we do not assume perfect rationality on the part of agents.

1.3 Games

The complexity of economic activity and its resulting output is typically studied at a certain level of abstraction, for instance, by means of public goods games (Fehr and Gachter 2000) or trust games (Johnson and Mislin 2011). Public goods games feature the requirement of many economic outputs to be produced through team-work (joint production). While the group-level outcome of a public goods game depends on individuals’ efforts and skills, the difficulty of measuring individual contributions effectively is most commonly represented by an even split of the outcomes. Thus, individuals have an incentive to not corporate. Knack and Keefer (1997) show that trustworthy behaviour may lead to better economic outcomes. Besides the level of cooperation, the matching of heterogeneous agents is relative to determination of the productivity (see for example Angelovski et al. (2018)). This can be represented by adding a spatial dimension to the public goods games, allowing distance to determine interaction/matching probabilities (Nowak and May 1992), which allows the effects of different kind of networks to be tested (Choi, Kariv, and Gallo 2016).

An important piece of evidence from the literature on economic games is that human subjects in the lab do not follow the predictions of classical game theory (Henrich et al. 2001). Instead, there is a growing body of literature showing that value-based decisions may drive cooperation (Sagiv, Sverdlik, and Schwarz 2011; Tao and Au 2014), which can be implemented in formal (predictive) models by following psychological game theory (Azar 2019; Colman 2003; Geanakoplos, Pearce, and Stacchetti 1989), for instance. Notably, Mercuur, Dignum, and Jonker (2019) in a recent paper, using an agent-based approach (as we do), link Schwartz values to the actions taken of real human beings in an ultimatum game. Although the authors are not able to always reproduce the human subject behaviour with the theoretical simulation model based solely on values, they show that this approach is promising in that it explains behaviour quite good on the aggregate level and far better than an alternative model of economic behaviour, based on the traditional economic utility maximisation principles.

An important aspect of theoretical analysis of psychological game theory is that games are not ‘solved’ in the traditional way (backwards induction) but instead they are (forward) ‘computed’: The activities of the players reside on beliefs/experiences, which do not need to be coherent, although alternative analytical solution concepts exist (see, e.g. the JEBO special issue introduced in Dufwenberg and Patel 2019). In general, such an approach allows heterogeneity in behaviour and incomplete knowledge. An early and most prominent example of a simulation model with heterogeneous agents and self-coordination properties is the El Farol Bar model (Arthur 1994), in which agents have heterogeneous sets of prediction models and act according to their past experience of how these models performed. We follow this strand of literature, further motivating our decision in the next section.

1.4 Heuristics

In general, economic models are based on ideas of methodological individualism, putting the individual economic actor at the centre of any explanatory model. This Homo Economicus is described as someone who has full awareness of:

why they do what they do, described as complete preferences over outcomes (Vanberg 2008),

how they select a given action from all possible actions, often termed complete rationality but perhaps better referred to as full optimisation or substantive rationality (Simon 1976, 1978) and

what they finally do, determined by the conditions under which they act, i.e. the personal endowments, environmental conditions and the expectations about what other actors will do, assuming (implicitly) rational expectations (Muth 1961) and common knowledge and rationality (Brandenburger and Dekel 1993).

A lot of the criticism of this traditional approach focuses on the inconsistency of theoretical predictions with actual observed behaviours and decisions, see Kahneman (2011), for example, who discusses the role of heuristics in decision making. Moreover, the Homo Economicus approach is concentrated on the representation of individuals and fails to account for their social relations or the consequences of individual decisions based upon them (Arrow 1994; Gintis 2000). A common critique levied at traditional economic models is that the individual knows too much about the game and other people (Simon 1990) speaks of olympian rationalty, according to (Velupillai 2010).

In our theory, the individual is still an important cornerstone but is perceived as a constituent of the entire enveloping social network, acting on the basis of individual experiences.

Accordingly, we depart from the standard approach in all three dimensions:

why agents in our theory act in a specific way is determined by their Schwartz values and the inner goal to act in accordance with these values. This is in line with the idea of preferences over procedures (Vanberg 2008) and also loosely related to the idea of reduction of cognitive dissonance (Festinger 1954),

how agents act is captured through ideas of classical behavioural economics, where agents employ routines that make efficient use of the given knowledge and context, (Gersick and Hackman 1990; Kao and Velupillai 2015) and

what decisions agents make is determined by the relations that each agent has with other agents, the history of these relations and the personal capabilities of the agent in question. This combines evolutionary game theory in a spatial context (Nowak and May 1992) with ideas of heterogeneous strategies that are individually evaluated over time (e.g. Arthur 1994).

By departing from an optimisation strategy, we are able to actually model the decision-making process as a function of the past experiences of the specific individual. Thus, we are able to determine explicitly what each individual knows, based on the games they participated in at any specific point in time.

1.5 Integrative/Compensatory Decision Process

Often people are confronted with complex decisions in an uncertain environment. Such an environment renders optimisation practically impossible. In the literature, different models of learning are employed to model how people behave in such situations (Arifovic 1994; Arthur 1994; Brock and Hommes 1997; Dosi et al. 1999). Another approach assumes using explicit heuristics that do not belong to the class of reinforcement learning algorithms or constructivist rationality related ones but is instead based on ideas of ecological rationality (Smith 2003). Proponents of such an ecological rationality paradigm often study individual behaviour in very specific ways, i.e. using heuristics (Gigerenzer and Todd 1999; Gigerenzer and Selten 2001). The research on universal values shows that there is predictive power in those values which allow for the behaviour of individuals to be explained in fairly complex situations, i.e. in strategic games (Lönnqvist et al. 2013; Sagiv, Sverdlik, and Schwarz 2011). Accordingly, it should be possible to construct a set of competing rules for a given situation that corresponds to the dominance of certain values. The value portrait of an individual will then, in conjunction with the decision situation/environment, define the hierarchy and/or weight with which these rules are applied in order to reach a decision.

There are two basic ways in which such a strategic decision process can be modelled: procedural/non-compensatory and integrative/compensatory (Ayal and Hochman 2009).^[2] In a non-compensatory approach, the result depends on the order in which the individual decision criteria are applied (reducing the set of potential actions), whereas a compensatory decision process does not depend on the order in which the criteria are applied.

A typical example for the first type of non-compensatory strategies is the ‘Take-the-best, forget-the-rest’ heuristic, with discriminatory cues (Gigerenzer and Goldstein 1996).^[3] An agent defines a set of criteria that is suitable for excluding options and sorts them by priority (i.e. lexicographically). Then the agent iterates over this set of criteria, applying each criterion to the set of options until a single option remains. If a criterion removes all the remaining options, it is skipped. If all criteria have been processed and more than one option remains, a second heuristic is used. The successive application of ordered criteria was also proposed as an optimisation technique by Waltz (1967).

A typical example for the second type of compensatory heuristics is a traditional Bayes-tree. Consider the same set of discriminatory criteria as before. In a setting of risk, we can attribute unconditional probabilities to each criterion and, by aggregating the unconditional probabilities for each option through simple multiplication, define the total probability of success for each option (Anscombe and Aumann 1963; Savage 1954). The optimal option is then the one with the highest chance of success. In non-risky evaluations, the probabilities are exchanged with weights. For each option, the total weight is then the sum of those weights where the criterion has been fulfilled. Here, the optimal option is the one with the highest total weight. The application of decision weights for individual criteria was also proposed as an optimisation technique by Geoffrion, Dyer, and Feinberg (1972); Zionts and Wallenius (1976); Figueira et al. (2013).

In this work, we focus on the second type, integrative/compensatory strategies, allowing agents to decide on their activities based on all Schwartz higher order values at the same. Potentially different significance of four Schwartz higher order values is represented by decision weights.

In the next section, we present the results of the empirical analysis of the relations between Schwartz values and individual economic behaviour. These results form the basis for the mechanisms implemented in the agent-based model. Country level results are also presented in the same section. Then, in the following section, we detail the proposed agent-based model. Finally, we present and discuss the simulation results.

2.1 Data

We used the following data from ESS database:

1. 21-item human value scale, as developed by Schwartz (2003)

2. sociological variable – level of education (eisced)

3. economic variables: total hours normally worked per week in main job, overtime included (wkhtot), usual gross pay of the survey participant,^[4] (grspnum) and indirectly usual net pay of the survey participant, (netinum). The last two variables are only used for calibrating the model

4. post-stratification weight (pspwght).

For each Schwartz value and participant, centred scores were calculated based on the ESS9 documentation. Firstly, the raw scores for the ten Schwartz values were calculated as the means of the relevant items. Secondly, an individual’s mean score over all items was calculated. Thirdly, centred scores were obtained from the raw scores by subtracting the mean score. Finally, the Schwartz values are aggregated, using averages into higher order values (HOV): Self-Enhancement (Power, Achievement and Hedonism with a weight one half), Self-Transcendence (Universalism and Benevolence), Openness to Change (Stimulation Self-Direction, Achievement and Hedonism with a weight one half) and Conservation (Security, Conformity and Tradition).

We initially considered 47,086 observations. In the next step, we considered only participants of working age (greater than 17), with strictly positive number of total hours worked per week including overtime (based on the value of variable wkhtot) and having pay as a main source of income (the variable fvgac = 1). This further reduced the number of observations down to 22,555. Additionally, after rejecting the incomplete cases with respect to all but pay related variables that were not directly used in the model, this number was further reduced to 22,125.^[5] The data covered 26 European countries: Austria (AT), Belgium (BE), Bulgaria (BG), Switzerland (CH), Czech Republic (CZ), Germany (DE), Estonia (EE), Spain (ES), Finland (FI), France (FR), Great Britain (GB), Croatia (HR), Hungary (HU), Ireland (i.e. Italy (IT), Lithuania (LT), Latvia (LV), Montenegro (ME), Holland (NL), Norway (NO), Poland (PL), Portugal (PT), Russia (RS), Sweden (SE), Slovenia (SI) and Slovakia (SK).

We decided to consider the information on gross pay (instead of net pay) to allow a comparison of results for countries with different tax and social security systems. In rare cases (7.37 %) of missing information about gross pay, but when the information about net pay was available, the missing information was imputed. We used country specific linear models considering both of the relevant variables (the average R squared coefficient for different countries’ models was 67.7 %).

Additionally, in the final step, the original post-stratification weights are used to make the sample representative with respect to age group, gender and education. These weights correct for sampling and non-response errors.

2.2 Empirical Analysis

The results of the empirical analysis were used in the construction of the agent-based model, which is presented in the next section. The following factors/mechanisms: propensity for management, propensity to change occupation and skills (measured by the level of education) are agent-based model relevant. Therefore, we firstly present the empirically observed relations between Schwartz higher order values and these factors. Secondly, the correlation values between Schwartz higher order values and individual and collective output (measured by gross pay) are presented. The former are used to calibrate the model, and the latter to validate it.

2.2.4 Gross Pay

We also calculated the empirical Pearson correlations between Schwartz higher order values and usual gross income of the survey participant. Post-stratification weights were used as weights in the correlation analysis. The results are presented in Figure 2. Furthermore, Table 9 in Appendix A presents the raw values along with the means and standard deviations for each Schwartz higher order value. To conclude, we firstly observe that such Schwartz higher order values as Self-Enhancement and Openness to Change generally exhibit positive correlation coefficients for all the countries. In contrast, the remaining Schwartz higher order values: Self-Transcendence and Conservation, show negative correlation coefficients. Secondly, the specific values of the correlation coefficients may vary among countries, and in some cases, they may even indicate a positive value for one country and a negative value for another.

Figure 1:

Schwartz’s core values Schwartz (2012).

Figure 2:

Empirical correlations between net income and Schwartz higher order human values per country. V_SE – Self-Enhancement, V_ST – Self-Transcendence, V_OC – Openness to Change, V_CO – Conservation.

3.1 General Characteristics

The purpose of the model is to evaluate the link between the Schwartz higher order values and the aggregated economic performance. For this purpose, agents are created and located in geographical space (we use the notion of geographical distance in our model). A distinct agent population has been created for each of the 26 countries considered in the simulation, based on the results of the ESS9 survey. The primary agent characteristics include the following: 1) spatial coordinates, 2) the Schwartz values, 3) educational level and 4) the number of working hours. Each survey participant is represented by the given number of agents. We initialise the model by using the post-stratification weight, which gives the weight of a given agent in the sample used in ESS9. The exact number of agents corresponding to one participant in ESS9 is calculated by rounding the product of the post-stratification weight and the fixed number 3000. We choose 3000 as a heuristic trade-off balancing the accuracy of representation of the entire country population and the simulation run time. This number ensures sufficient representation of the whole population based on the weighted sample – rounding errors were negligible.^[11] Additionally, each agent is characterised by two spatial coordinates (x, y) corresponding to the geographic coordinates. These coordinates were sampled uniformly, disregarding any effects of the actual (potentially non-uniform) within-country Schwartz values’ distribution because of lack of the relevant data.

During each simulation step (we use 100 steps as this number is sufficient for the aggregate levels), an agent plays with its neighbours in a production game. There is no explicit time dimension for each step, and the results of the 100 steps are finally aggregated by taking the average. Firstly, agents decide whether they would like to organise the game (be a manager). If so, the agents invest their effort in their own game. Otherwise agents invest their total effort in the games of one of their neighbours. The neighbourhood is defined based on the Euclidean distance and the radius r, which is the simulation parameter. Secondly, the sum of all efforts collected by an agent is transformed into output, according to the agent-individual production function. Finally, the output is distributed among agents. The games are detailed in Section 3.2.

The consecutive phases of the simulation are enumerated below:

1. Game selection: Selection of one of the available games

2. Production: Produce the output

3. Distribution: Distribute the output among agents

In the first – game selection, the agents select a game they want to participate in, and invest all their efforts in such a game. In the second – production, the total effort of the agents is transformed into output according to the adopted production function. In the third – distribution, the output is distributed among all the agents proportionally to their effort.

The model is implemented in JAVA. The code is available at: https://www.comses.net/codebases/c0b005bc-e6a4-49bd-894d-38cfebc55284/releases/1.0.0/.

3.2 Games

4.1 Simulation Parameters

We executed the model separately for each country. The ranges for all the parameters determining agent behaviour are shown in Table 5. Furthermore, to systematically search the parameter space, we used 800 value sets determined by Sobol numbers (Bratley and Fox 1988; Christophe and Petr 2014). As the parameters are not directly observable (and thus unknown), we arbitrarily selected relatively wide ranges, additionally calibrated by the results of the preliminary simulation runs. We run the simulation for 26 country specific synthetic populations. In case of higher order values related parameters, we parametrised differences between parameters for two different higher order values rather than the absolute values. In such a way, we accounted for the empirical relationships between values and observable behaviours, shown in Tables 2 and 3.

We also considered four variants of rationality of integration (decision weights): simple, suppressed, ranked and non-linear; one variant of pay-off distribution: proportional to the invested effort and finally, one variant of the total effort determination: education based. We ran 2 simulations for each parameter set with different seeds, which led to 166,400 = 800 × 26 × 4 × 1 × 1 × 2 simulation runs in total. Each simulation run consisted of 100 simulation steps.

4.2 Calibration

The characteristics of the calibration and the subsequent sensitivity analysis are related to the explanatory nature of this model, with the objective being to show that the connection between Schwartz higher order values and macroeconomic variables can be made. We used empirical Pearson correlations between Schwartz higher order values and usual gross income for different countries for the calibration of the model, see Table 9 in Appendix A. Furthermore, based on the simulation data, we calculated the Pearson correlations between Schwartz values and the agent’s total income based on the entire agents’ population, for each country and each parameter set separately. We used an agent’s total income, which was calculated as the mean value for the results for two different seeds.^[13] The observed differences between empirical and simulation-based correlations were summarised using root mean squared error (RMSE) for each set of parameters.

Finally, calibration was conducted in two steps. In the first step, we calculated the average RMSE for each of four different decisions weights: simple (0.100), suppressed (0.127), ranked (0.0829) and non-linear (0.113), and chose the one with the smallest error. In the second step, we selected the one global parameter set (i.e. the same for each country) for which the squared difference between empirical and simulation-based correlations is minimal.

The values of the optimal parameter set (from the 3200) are as follows: rationality = hierarchical, total effort = education based, γ = 1.671, β ₁ = 4.199, β ₂ = −0.053, σ = 0.008, ν = 1.039, k = 1.654, p C O O = 0.056 , p S T O = 0.08 , p O C O = 0.106 , p SE O = 0.182 , p C O C = 0.051 , p S T C = 0.052 , p SE C = 0.108 , p O C C = 0.151 , r _CO  = 0.033, r _ST  = 0.04, r _SE = 0.067, r _OC  = 0.076, f _e  = 1.08.

The neoclassical production function, for the optimal set of parameters and the total effort values ranging from 0 to 500, is presented the left (a) panel of Figure 3. For the total effort value of 500, the output value is close to 0. The values of the output, but only for the simulation relevant range (the typical values of the total effort were included in this range during the simulation), are presented in the middle (b) panel. We can observe that the product is increasing and convex for this range. The normalised output, namely the output per unit of an effort, is presented in the right (c) panel. We use a different scale on each graph to improve readability.

Figure 3:

Production function for calibrated parameters.

The empirical Pearson correlations (Schwartz values and income) for different countries are presented as a heat map in Figure 4, and additionally in Table 10 in Appendix A.

Figure 4:

Simulated correlations between net income and Schwartz’s human values per country. V_SE – Self-Enhancement, V_ST – Self-Transcendence, V_OC – Openness to Change, V_CO – Conservation.

We have also calculated the correlations between collective income for each of the 26 populations and the average numerical value of the Schwartz higher order value in a given population, and the results are shown in Table 6. At the level of the population, Self-Enhancement is negatively and Self-Transcendence is positively correlated with economic outcome (GDP). Self-Enhancement and Openness to Change is negatively correlated with Gini coefficient (associated with lower income inequality). Opposite effect may be observed for Self-Transcendence and Conservation. Such results correspond to the empirical observations.

For the validation purposes, we calculated Pearson correlations between simulated and empirical gross domestic product (collective output) and Gini coefficients for different countries. Both correlations coefficients were positive 0.40 (n = 26, p value = 0.043) in case of collective output measures, and 0.38 (n = 23, p value = 0.072) in case of Gini coefficients.

The differences between simulated and empirical Pearson correlations (Schwartz values and income) for different countries are presented as a heat map in Figure 5, and additionally in Table 11 in the Appendix. The mean absolute difference for all countries and the Schwartz higher order values amounts to 0.052. It may partly result from the relatively high differentiation of the correlations by country and the abstract character of the model. Since we did not define individual parametrisations per country, structural differences that are not reflected in the different geospatial Schwartz values’ distributions (from ESS9 data) will contribute to these differences, too.

Figure 5:

Differences between empirical and simulated correlations between net income and Schwartz human values per country. V_SE – Self-Enhancement, V_ST – Self-Transcendence, V_OC – Openness to Change, V_CO – Conservation.

4.3 Sensitivity Analysis

We conducted two types of sensitivity analysis for the obtained results. Firstly, based on different simulation parameter values, and secondly, considering the distribution of Schwartz higher order values in the population.

The regression analysis was conducted in two steps. Firstly, we fitted the generalised additive model, Wood (2004, 2011 to identify potential non-linear relations between explanatory variables (gross output and Gini’s coefficient) and different parameter values. Such a non-linear relationship was identified for the gross output and the following parameters: β ₁ and β ₂ – production function parameters and p C O C  – probability of game change for Conservation. Secondly, we fitted linear regression models with additional squared terms for these three parameters. The results are presented in Table 7 for the mean output and Table 8 for the Gini coefficient.

Generally speaking, the following parameters: β ₁, β ₂, f _e , and all radius-related parameters increase the gross output. The probability of game change p C O C rather reduces the gross output. The impact of probabilities of game ownership for different higher order values is ambiguous (Table 8).

The following parameters: β ₁, β ₂ and f _e increase the Gini coefficients, indicating higher inequality. As a tendency, probability of game ownership, probability of game change and radius-related parameters decrease the Gini coefficients.

We also conducted a sensitivity analysis with respect to the distribution of Schwartz higher order values in the observed populations. Firstly, we constructed a set of synthetic populations by changing (decreasing and increasing) the importance of a given Schwartz higher order value in a population, and additionally selected pairs of higher order values simultaneously, for each agent separately. The Schwartz higher order values were changed each time by adding a constant:

The marginal values of −2 (2) make all individual agents have the least (respectively the most) important higher order value equal the stressed value. Finally, the new values are adjusted in such a way that the corresponding decision weights remain in the range [0,1]. We then calculated the economic performance parameters – total economic performance as measured by gross output and income inequality – for the scenarios under analysis and compared the results with the base scenario. The results for total economic performance (as measured by aggregate income) and higher order values, averaged for all 26 countries, are presented in Figure 6. The baseline results are represented by the horizontal dashed red line in the respective figures.

Figure 6:

Total economic gross income of the population and the Schwartz higher order human values. (a) Self-Enhancement, (b) Self-Transcendence, (c) Openness to Change and (d) Conservation.

We can observe non-linear relations between gross output and higher order values. The effect with the greatest magnitude is observed for Self-Enhancement and is negative. The fact that Self-Enhancement reduces cooperation between agents may explain this observation. The effect for Openness to Change is non-monotonic first increasing and then, for the significantly positive stress factors, decreasing. This non-monotonic relation may suggest the potential conflict between self-interest and common interests. Furthermore, we can also observe that a higher level of Conservation (ceteris paribus) corresponds with an increased collective output. However, the empirical data show the opposite relationship, see Table 4. One potential reason could be the negative correlation of the Conservation with education level, which is indirectly accounted for in the empirical data, but not in the simulation.

The results for total economic performance and the pairs of higher order values, averaged for all 26 countries, are presented in Figure 7.

Figure 7:

Total economic gross income of the population and the pairs of Schwartz higher order human values. (a) Self-Enhancement & Openness to Change, (b) Self-Transcendence & Conservation, (c) Self-Transcendence & Openness to Change and (d) Self-Enhancement & Conservation.

Similarly, we can observe non-linear relations between gross output and higher order values. The effect with the greatest magnitude is observed for the pairs Self-Enhancement and Openness to Change (negative, but irrelevant for significantly negative stress factors) and Self-Transcendence and Conservation (positive, but irrelevant for significantly positive stress factors).

The results for income inequality, averaged for all 26 countries, are presented in Figure 8. We measure income inequality by means of Gini’s coefficient.

Figure 8:

Gini’s coefficients for the economic gross income of the population and the Schwartz higher order human values. (a) Self-Enhancement, (b) Self-Transcendence, (c) Openness to Change and (d) Conservation.

We can observe non-linear relations between economic inequality and higher order values. Self-Enhancement and Openness to Change decrease Gini coefficient, while Self-Transcendence and Conservation increase it. This may be explained by the higher total number of games available in the local neighbourhood (higher probability of game ownership and larger radius), more frequent game changes by the agents in the former case and the opposite effect in the latter.

The results for economic inequality and the pairs of higher order values, averaged for all 26 countries, are presented in Figure 9.

Figure 9:

Gini’s coefficients for the economic gross income of the population and the pairs of Schwartz higher order human values. (a) Self-Enhancement & Openness to Change, (b) Self-Transcendence & Conservation, (c) Self-Transcendence & Openness to Change and (d) Self–Enhancement & Conservation.

Similarly, we can observe non-linear relations between Gini’s coefficients and higher order values. The effect with the greatest magnitude is observed for the pairs: Self-Enhancement and Openness to Change (negative) and Self-Transcendence and Conservation (positive) as the similar effects of both higher order values in a given pair accumulate. The non-monotonic relation observed for the two remaining pairs: Self-Transcendence and Openness to Change and Self-Enhancement and Conservation may be explained by the fact opposite individual effects of both higher order values in a given pair cancel out.