Social Network Dynamics and Sustained Grading Discrimination: An Agent-Based Model
Dorottya Kisfalusi, Karoly TakacsSchool grades are important determinants of educational attainment. Several empirical studies found evidence for a gap in grades for different groups of students after controlling for blind and unbiased test scores. Girls, for instance, were found to receive better non-blind assessments than boys and ethnic differences in grading were observed in several countries. This study tries to explore mechanisms that could explain this persistent difference.
A possible mechanism has been described as “Acting White”: high performing minority students are pulled back and are threatened by exclusion from friendship due to a pressure arising from anti-achievement norms and oppositional culture. As a different mechanism, their social network ties could create a filter bubble and prevent minority students from improving their situation. Endogenous network dynamics may amplify the negative effects of biased grading as selection into homophilous friendship groups and peer influence on efforts might self-reinforce the grading gap. Moreover, teachers might simply discriminate based on their previous experiences. Even if direct experiences do not exist, differential treatment could be driven by status generalization processes such that members of groups with lower status value in society are judged by stricter standards than others.
These mechanisms are hard to disentangle in empirical studies. Using agent-based simulation, we can examine the impact of various mechanisms on the persistence of grading bias. Moreover, we investigate if homophilous friend selection and peer influence amplify the grading bias over time.
Our model represents one grade of a school with “red” and “blue” students. Students are arranged in a friendship network and attributed with stable abilities that follow a normal (or empirically informed) distribution. Students receive grades based on their abilities, effort, and an innate stereotype of the teacher. Students update their academic effort based on their direct experience of how their effort was translated into grades. If they feel insufficiently rewarded, they lose motivation and study less. Moreover, their effort levels are influenced by their friends. In every round that could be considered as one grading period, they also update their friendship ties with a certain rate. Structural updates follow mechanisms such as reciprocity, preferential attachment, and transitivity; but ties are also selected based on similarity concerning group membership and academic achievement.
From an initial relatively fair distribution of abilities, we do not see the emergence of large within group differences. Our preliminary results do not indicate that clear-cut oppositional cultures could evolve in the lack of statistical differences between the groups and given equal opportunities. As we hardwire in individual development, it is not surprising to see average effort and grades increasing. This tendency towards high effort seems to be strengthened in case of homophilous selection. Low density and tie decay work against overall positive outcomes as they cause segments of low achievers from the same group to emerge.
We conclude that just as for opinion dynamics models, it is difficult to explain the emergence of polarized outcomes such as oppositional cultures and persistent grading differences from equal abilities and opportunities.