The Effects of Communication Strategies, Cognitive Traits, Group Composition and Social Information on Collective Decision-Making Performance - Simulation Experiments with Agent-Based Models

Abstract

Cognitive functions play a crucial role in managing complex socio-technical systems. Organizational bodies that manage those systems must continuously be aware of what is going on, make sense of what is going on, design purposeful responses, and adapt to new situations. System failure is often due to unreliable cognitive functions in teams that manage these systems. In case of disruption, management and emergency response teams must identify the most effective course of action to reestablish system functionality. Considerable research has investigated executive decision-making; however, we still do not completely understand what factors shape the collective cognitive processes that occur prior to making a choice. To address this issue, the present study aims to (1) characterize how different information aggregation mechanisms affect group decision-making, (2) investigate how individual social and cognitive traits affect group decision-making, and (3) identify how social signals affect decision-making and group norms. To increase our understanding of individual decision-making behavior under different framing conditions, research has primarily focused on laboratory experiments. Studying collective decision-making in a laboratory environment is difficult because it is hard to fully control the design variables. Therefore, we used a simulation experiment approach and developed an agent-based model that mimics group behavior. The treatment factors included information sharing strategy, information load, information diversity, agents' cognitive capabilities, and group composition, among others. A complementary laboratory experiment examined how social information about the behavior of others affects decision behavior and group norms.The study resulted in five principal findings: First, we contributed to the development of agents with human-inspired reasoning and information-processing behaviors. The proposed model improves on existing agent-based models by allowing agents to reason in a complex way about which action to select based on their available information. The model more closely resembles the mechanisms of behavioral change in the real world and demonstrates that system behaviors can emerge from the natural evolution of a system's information state. Second, we found that a group's communications strategy is the most important tool to shape decision performance. In particular, strategies that help groups surface all the information known by their members improve decision performance. This aligns with previous studies that found that communication interventions shape decision outcomes significantly. However, these studies could not control for other important factors that might shape decision outcomes. Adopting a computational approach allowed us to model and control for other important factors. Moreover, we found that random information sharing and instructed dissent were the best-performing information sharing strategies, implying that there is considerable room to improve group communication. Third, cognitive ability helps team members execute their communication strategy, which provides a mechanism to explain varying estimates about the effect of member-IQ on group performance. Cognitive ability is an amplifier rather than a main driver of group performance and should be studied in conjunction with other group processes. Fourth, we also found that individual-level factors cause less variation in decision performance than team-level factors, such as information load and diversity of the information pool. Experiments with group composition, which considered inequalities in information, social influence, and turn-taking among group members, found that group composition did not have a significant effect on group performance. Fifth, even weak social signals can frame decision behavior. We found that negative social signals had a greater effect than positive social signals, implying that it is easier to instigate anti-social behavior than pro-social behavior. These findings have implications for organizers and moderators of situational assessment, sense-making, and decision-making meetings. They should be trained in the design and implementation of effective group cognition tactics. Our results suggest team members should have diverse information sets and employ strategies that promote information coverage, for example, by assigning a team member the role of devil's advocate. We offer four suggestions to improve future computational models of collective cognition. First, at the agent level, modelers may expand agents' cognitive architecture, which could include dynamic goal setting mechanisms, a greater degree of learning and adaptation, and stronger coupling between cognitive processes and the environment of the agent. Second, at the group level, models could include more latent group processes, such as coalition formation, bargaining, and the emergence of informal group structures. Third, there is a need to improve the robustness of simulation experiments by systematically investigating the effects of scaling of the design variables. Fourth, the decision perspective has to be enlarged because it assumes that the decision alternatives are well defined and known by all the group members. This perspective does not fit to system states under disturbance and disruption. Precedent phases, in particular problem perception, sense-making, problem framing, and identification of courses of action, are more decisive (Simon, 1986}, requiring further research.