3 A psychologically sound middle‐range agent‐based social simulation of collective
3.3 Dynamical Analysis
3.3.2 Homogenous populations with social influence
Vice versa egoistic agents stress egoistic tendencies (egoisticTendency=0.3, altruisticTendency=0.1). For this type, average contribution levels do not exceed 0.07
yielding negligible PG levels between 0.0 and 0.1.
A third social orientation preference may be found for neutral agents with a balanced weighting between altruistic and egoistic tendencies (egoisticTendency=0.2, altruisticTendency=0.2). Contribution levels of this agent type in average can go up to 0.21 while the average level of the PG does not exceed 0.6.
The three basic agent types are to be understood as discrete representatives of ranges of parameter combinations in HAPPenInGS‐A which lead to qualitatively similar macro‐level outcomes. Altruistic and egoistic types are located at the two extreme ends of the spectrum of social orientations that are covered by HAPPenInGS‐A. The neutral type represents the transition range of parameters between the extremes.
The sensitivity analysis of this section has disregarded social influences by setting the importance of the social conformity preference to 0 for all agents. The next section explicitly analyses the effect of increasing the weight put on the social conformity preference in populations that are homogeneously composed of one of the three basic agent types.
Basic agent type
Preference set publicGood‐
Importance
egoisticTendency altruisticTendency socialConformity‐
Importance
Altruist 1.0 0.1 0.3
0.0 to 0.5, resolution 0.1
Neutral 1.0 0.2 0.2
Egoist 1.0 0.3 0.1
Table 4. Agent types and respective preference sets used in the sensitivity analysis. Three basic types of agents regarding social orientation are investigated. For each basic type six different settings for the social conformity preference are considered.
Simulation results are reported with respect to three different indicators:
Agents’ behaviours in terms of their mean contribution (black).
Agents’ achieved social conformity as the mean value of their perceived social conformity (red). For each agent the social conformity perception is calculated according to the criterion formula for preference #4 in Table 2. It reflects an agent’s
absolute attainment of its social conformity preference.
The success of the collective action as the mean level of the PG of all groups (green with error bars of one standard deviation).
The first set of diagrams compares the performance of homogeneous populations of the three agent types with increasing importance of the social conformity preference. To assess performance we calculate means of the indicators during the last 200 steps for each run. In the diagrams we show the mean values of these aggregations and standard deviations of the 20 initialisations used.
Figure 10. Aggregated results of the sensitivity analysis for altruistic agents. For each run performed we calculate the mean values of three different performance indicators during the last 200 simulation steps. In the diagrams the error bars show the standard deviation of the mean values of the 20 runs performed for a distinct parameter setting. See text for further explanation.
Figure 11. Aggregated results of the sensitivity analysis for neutral agents. For each run performed we calculate the mean values of three different performance indicators during
the mean values of the 20 runs performed for a distinct parameter setting. See text for further explanation.
Figure 12. Aggregated results of the sensitivity analysis for egoistic agents. For each run performed we calculate the mean values of three different performance indicators during the last 200 simulation steps. In the diagrams the error bars show the standard deviation of the mean values of the 20 runs performed for a distinct parameter setting. See text for further explanation.
As a general pattern for all three types of social orientations we observe that with increasing importance of social conformity indeed the social conformity perceived by the agents rises and settles close to the maximum level of 1. Clearly, this demonstrates that agents manage to satisfy the social conformity preference in addition to their other preferences. However, when relating this observation to behaviours of the agents we see clear differences between the three types.
For egoists low contributions are reinforced as a social norm when the importance of social conformity during agent decision‐making increases (see Figure 12). This is mainly due to the homogeneous settings of agent behaviours during model initialisation which sets investment levels to 0 for all agents (see section 3.2.5). These settings are well in line with egoistic social orientations and in addition provide initially high local conformity perceptions. Therefore,
the passive social norm of low contributions persists throughout the simulation in the egoistic population.
For the altruistic population (Figure 10) we observe more diverse dynamics: Altruistic agents manage to achieve high contribution levels and provide high levels of the PG along with increasing social conformity perceptions until the importance of social conformity exceeds 0.3. Obviously, when comparing to the egoistic case, in the altruistic population a different social norm emerges until the importance of social conformity goes beyond this threshold.
For higher preferences for social conformity even altruists show the same lock‐in of passive behaviour as egoists – the high weight of the social conformity preference dominates altruistic preferences during decision‐making and therefore inhibits initial mobilisation. For neutrals we observe a similar behaviour intermediate to that observed for the other two types (see Figure 11).
The following analysis investigates the case of the altruistic population in more depth. We examine the temporal dynamics underlying the described aggregated observations and
introduce two additional performance indicators:
The standard deviation of agent behaviours within a population (vertical dashed, black).
The temporal stability (convergence) of agent behaviours as the total decrease or increase of contribution relative to the previous time step on the individual level (delta contribution, blue).
Results are reported in terms of mean values of the respective 20 runs performed per simulation tick. Furthermore, we pick three distinct settings for the weight of the social conformity preference during decision‐making: In Figure 13 we display simulation results for socialConformityImportance=0 which reflects the extreme case where social networks have no influence in the decision process. In the aggregated simulation results discussed above (see Figure 10) we find that for a medium setting for the importance of social conformity (socialConformityImportance=0.2) success of the PG provision and the achieved social conformity have their maxima. Therefore, we pick this setting for a second in‐depth analysis that is shown in Figure 14. Finally, Figure 15 displays results for the extreme case of socialConformityImportance=0.5.
Figure 13.Temporal dynamics for altruistic agents and preference for social conformity 0. We show mean values of the respective 20 runs performed per simulation tick. See text for further explanation.
Figure 14. Temporal dynamics for altruistic agents and preference for social conformity 0.2.
We show mean values of the respective 20 runs performed per simulation tick. See text for further explanation.
Figure 15. Temporal dynamics for altruistic agents and preference for social conformity 0.5.
We show mean values of the respective 20 runs performed per simulation tick. See text for further explanation.
In all three figures above, due to the homogeneous initialisation of the agents with zero contributions, the social conformity perceived by the agents starts on the maximum level during the initial simulation steps. If the preference for social conformity is high (Figure 15) this passive social norm is sustained throughout the simulation and social mobilisation does not take place. For more moderate preferences for social conformity we observe successful social mobilisation (Figure 13, Figure 14). In both cases the achieved social conformity drops during the initial phase of social mobilisation. For the first case (preference for social conformity 0) the maximum success of PG provision is reached after 150 ticks and sustained until the end of the simulation. In contrast, for the moderate social conformity preference (Figure 14) mobilisation proceeds slower but continuously improves towards the end of the simulation. This slow‐down of social mobilisation may be attributed to an on‐going process of social adjustment: Clearly, compared to Figure 13, in Figure 14 the achieved social conformity does not drop as steeply and instead continues to increase during the simulation after tick 100. Likewise, for the medium social conformity setting in Figure 14 we observe a decrease of the standard deviation of the contributions (dashed black) which does not occur
for the low social conformity setting in Figure 13. An additional difference between simulations with low and medium settings for the social conformity preference lies in the development of the behavioural stability indicator (delta contribution in blue) that reflects the individual change in behaviour between two consecutive simulation ticks: While under both settings contributions settle around a mean of 0.3, in the absence of social conformity preferences individual behaviours appear to fluctuate significantly (around 0.1) from one tick to the next throughout the simulation, i.e. agents continuously adjust their behaviour. With medium social conformity preference this effect is dampened and delta contribution approaches zero over time. Apparently, the striving for social conformity not only decreases the spread of behaviours within the population; it also decreases the spread of behaviours selected by one agent over time.
The following section goes a step further and reports on simulation results for agent populations with heterogeneous preference profiles.