Acquisition of Gender STEM Stereotypes through Reinforcement Learning in School-Age Children

• In Event: Poster Session 11
  In Poster Session: PS 11 Topic - Sex, Gender

Abstract

Both adults and infants have the ability to use reinforcement learning to spontaneously learn rules for behavior and to generalize them to novel situations (Werchan et al., 2016; Collins et al., 2014; Collins & Frank, 2013). While vital for cognitive development, we hypothesize that this learning mechanism may also foster the acquisition of stereotypes and prejudicial attitudes. In this study, we focused on the STEM gender stereotype: the idea that women perform worse than men in STEM fields (Smeding, 2012; Spencer et al., 1999; Eccles, Jacobs, & Harold, 1990). Children, ages 6-15 (N=38), were told that they would play a game in which cartoon monster characters participated in academic classes (reading, writing, math, and science). Children were offered the opportunity to implicitly learn rules of class achievement by assigning grades to masculine and feminine monster characters and receiving feedback on their selections (Figure 1). In the learning phase of the task, children assigned grades to the different types of monsters for each of their classes. After assigning a grade to the monster, participants received feedback on whether their assigned grade was correct. In a generalization phase of the task, children were asked to use their previous knowledge to grade the monsters in two new classes that were related to the original classes. A total of 20 participants completed the “stereotype matching” condition of the task (masculine monsters paired with math/science achievement) and the remaining 18 participants completed the “stereotype non-matching” condition (feminine monsters paired with math/science achievement). First we found a developmental effect on learning: older children exhibited superior learning through feedback in comparison to younger children, F(3, 105) = 3.76, p = .01, η2 = 0.10. Second, while those in both the stereotype matching and non-matching conditions learned similarly, only those in the stereotype matching condition (where masculine monsters were paired with science and math achievement) showed generalization of the previous learned rule, F(1, 29) = 9.8, p = .004, η2 = 0.25. Third, we found that for children with more years of formal schooling, there was no evidence for an association between the rate of learning and generalization ability whereas for children with less years of formal schooling, the rate of learning was associated with generalization ability, F(1, 29) = 14.22, p = .001, η2 = 0.33. The findings of this study suggest a flexibility of learning at a young age, and potentially opens the doors to beneficial interventions that can occur on both a systematic and individual level to eliminate harmful stereotypes that can impact future development and achievement.

Authors

• Rachel Martino, University of Delaware
  Presenting Author

• Denise Werchan, Brown University
  Non-Presenting Author

• Dima Amso, Columbia University
  Non-Presenting Author