Thesis defense

Myside bias and the spread of information within groups

Gafari Lukumon (UM6P/ENS-Ulm/IJN)
Practical information
25 June 2024

University of Mohammed VI Polytechnic, Amphi Rabat ABS Building, Bloc C, 1st floor, Rabat Campus & online



  • Mr. Brent STRICKLAND École Normale Supérieure Supervisor
  • Ms Rebecca LITTMAN University of Illinois, Chicago Rapporteur
  • Ms Valerie SHALIN Wright State University Rapporteur
  • Ms. SIHAM LEKCHIRI Western Carolina University Examiner
  • Mr. Dirk Michael BOEHE University of Mohammed VI Polytechnic, Morocco Examiner
  • Mr. Salvador MASCARENHAS ENS-PSL Examiner
  • Mr. Mark KLEIN University of Mohammed VI Polytechnic, Morocco Guest


This thesis examines “myside bias” (i.e. a disposition to evaluate evidence, test hypotheses, search for information in a way that disproportionately favors prior attitudes and beliefs) and its influence on the spread of information within groups. The perspective is that the dynamics determining information spread in social groups emerge from aggregated individual processes, preferences and decisions. Interventions to improve the quantity and quality of information flow can happen at individual or aggregate levels. The thesis is structured into four axes. 

In the first axis, spanning seven studies, we investigated why group members do not always express their opinions in a group and found that observers judge group members who express pessimism about the group's future as being less loyal to the organization (compared to members who express optimism). Inferences of disloyalty persisted even when observers believed that the pessimistic individual was more accurate and when the individual aligned with the (pessimistic) majority. Individuals who wish to be perceived as loyal face pressure to exhibit blind optimism. Our findings revealed that people are more likely to promote and extend invitations to non-work social functions to optimistic group members than to pessimistic ones. These results were replicated across three types of organizations, including companies, political organizations and sports leagues.

The second axis of the project focused on individuals' attitudes toward fact-checking groups that exhibit ideological homogeneity or heterogeneity. Across two pre-registered studies, we investigated preferences (and motivations) for the conclusions of these types of fact-checking groups and trust in them. Our results revealed a strong preference for bipartisanship, emphasizing the need for a balanced mix of heterogeneous groups in the fact-checking process. Participants (left and right-leaning Americans) preferred and favored heterogeneous - diverse fact-checking groups, with left-leaning individuals displaying a greater interest. Notably, both Democrats and Republicans trusted heterogeneous fact-checking groups as much as their own but expressed lower trust in opposing ideological groups. Pilot studies conducted in Nigeria replicated some of these findings, highlighting the cross-cultural relevance of the research.

In the third axis, through a series of agent-based model analyses, we investigated the impact of confirmation bias (myside bias) on multiple steps in accumulating scientific knowledge. We compared cases where confirmation bias exists in processes such as the belief updating of a single scientist, a population of information receivers (such as those who read research articles), a population of experimenters, a population of peer reviewers, and all stages mentioned above. We found (unsurprisingly) that groups are more likely to converge on a false consensus the more biased their individuals are. More surprisingly, however, we also found that biased groups converge on conclusions, true and false alike, faster than non-biased groups. Our findings thus suggest a speed-accuracy trade-off associated with myside bias, and groups seeking to modulate bias at the community level should consider the likely consequences on accuracy as well as decision speed.

The last axis focused on the types of information people find easy to spot (good vs. bad) and the effect of token budget sizes in information harvesting.