A large-scale study of infant intuitive physics

• In Event: Poster Session 04
  In Poster Session: PS 04 Topic - Cognitive Processes

Abstract

Even before their first birthday, infants already have expectations about how objects should move and interact, and are surprised if objects float in midair, pass through walls, or start moving without apparent cause (e.g. Baillargeon, 1987, Spelke et al., 1992). However, such findings are inferred by the pattern of successes and failures at a group level, and we still lack a quantitative understanding about the generalizability of these findings, the variability of individual differences, and the robustness of looking-time methods across repeated measurements.
Here, we used an online platform, Lookit!, to conduct 15-minute preferential looking experiments at unprecedented scope and scale with over 220 infants (aged 4-12 months). 50 of these infants additionally completed 12 total sessions within two months, providing an unusually dense dataset for within-subject analyses. Families participated from home and streamed session video via webcam for up to 24 trials per session. Trials began with an object introduction (~5 seconds) to orient children to the center of the screen. Next, two events involving that object are shown simultaneously on both sides of the screen, looping for 20 seconds. Critical trials contrasted typical physical events against those that violate principles of support relations (Baillargeon, 1987), gravity, or inertia (Kim & Spelke, 1999). We also incorporated calibration trials and control trials to detect individual differences in side bias or sensitivity to visual salience (see Table 1).
Our analytic strategy is to use generalized linear mixed effects regression to predict infants’ preference scores (i.e. proportion of time looking at either video). First, we aim to characterize infants’ physical knowledge on critical trials. We will report the percentage of participants with a significant preference (in either direction) on each event type. We will analyze developmental trends by testing if age significantly predicts infants’ preference scores for each physical concept (Figure 1). We will examine within-concept contrasts (e.g. support events with varying amounts of contact) as additional predictors.
Second, we aim to assess the stability of infant’s looking behaviors across repeated test sessions. We will ask if infants’ preference on critical trials change over the 12 sessions by including session number as a covariate. We will also report the variability and stability of responses to control trials and use clustering techniques to identify stable individual differences. If significant clusters are found, we will test if prior regression analyses of physics knowledge are robust to such individual differences.
Finally, we aim to map out the overall structure and co-development of physical beliefs. While researchers have investigated the fine-grained developmental trajectory of single concepts, such as support (Baillargeon 1987), it is unclear how knowledge about one concept may inform beliefs about another. Using hierarchical clustering methods, we will explore (a) which test events elicit similar looking time patterns across development and (2) if there are stable dependencies in the trajectories of physical beliefs.
With an existing analysis pipeline from pilot work and data collection complete, we anticipate 5 months to finish coding the remaining 80% videos and 1 month for analysis and poster preparation.

Authors

• Kailande Cassamajor, Howard University
  Presenting Author

• Junyi Chu, Massachusetts Institute of Technology (MIT)
  Non-Presenting Author

• Kimberly Scott, Massachusetts Institute of Technology (MIT)
  Non-Presenting Author

• Laura E Schulz, Massachusetts Institute of Technology (MIT)
  Non-Presenting Author