How Teachers’ Assumptions About Learners’ Abilities Shape Inferences in Pedagogical Reasoning

• In Event: Diverse Knowledge, Diverse Cues: Implications for Learners’ Inferences and Actions

Abstract

Good teachers should teach at the appropriate level for a learner. In turn, learners must judge whether teachers are selecting appropriate material for them. Thus, learning from pedagogy entails a sophisticated recursive inference process: Teachers must accurately represent the learner’s abilities, and learners must consider whether the teacher has done so (Bonawitz & Shafto, 2016). Are children sensitive to whether teachers’ demonstrations are at the appropriate level for them? In past work, children’s exploration after observing a pedagogical demonstration given to another child was similar to their exploration after being taught themselves, but not when observing direct instruction to adults, suggesting children attend to whether an overheard teacher’s demonstration is likely to be relevant to them (Bonawitz & Shafto et al., 2011). These effects raise questions about how exactly children’s inferences are shaped by what they believe a teacher knows about them.

Method: First, children (age 6-8) played a picture matching game with an experimenter on Zoom, and were able to complete one of two matching sets (controlled). To create between-subjects conditions in which a teacher either overestimated, underestimated, or accurately represented the child’s performance on this task, a confederate (the “Teacher”; actually a pre-recorded video, unbeknownst to the participant) then saw participants’ results reported on screen. This display was either Accurate (showed one set completed), or it led the Teacher to Overestimate (erroneously showed both sets completed) or Underestimate (erroneously showed neither set completed) the child’s performance. The Teacher then presented three new matching games, and recommended one based on her understanding of the child’s prior competence; the other two games, the Teacher noted, would likely be too hard and too easy for the child respectively. Participants ranked their choices for which game they would prefer to play. (Figure 1 contains a summary of the method.)

Coding: Chosen games were assigned different point values (too easy=1, just right=2, too hard=3), and rankings were assigned different weights (first choice=2, second choice=1, third choice=0). We analyze the weighted average of this ranking.

Predictions: If children’s inferences are rationally informed by the Teacher’s purported understanding of their prior knowledge, children in the Accurate condition should be more likely to attempt the recommended “just right” task. In the Overestimate and Underestimate conditions, however, children should prefer the game that the Teacher believes will be too easy or too hard (respectively) for the child, thereby “compensating” for pedagogy that was generated using an inaccurate representation of their abilities.

Results: Although data collection has just begun (N=15 children, 5/condition; N=90 pre-registered), preliminary results support our predictions: Children prefer easier games when the Teacher overestimated their knowledge, and harder games when she underestimated their knowledge, while children in the Accurate condition are more likely to follow the Teacher’s recommendation (F(2,12)=5.2, p=.023; see Figure 2). This work begins to shed insight into yet another way in which young learners may capitalize on subtle information about others’ knowledge to make rational inferences during informal pedagogy.

Authors

• Ilona Bass, Harvard University
  Presenting Author

• Elise M Mahaffey, Arizona State University
  Non-Presenting Author

• Elizabeth Bonawitz, Rutgers University - Newark
  Non-Presenting Author