Poster #69 - Under Which Circumstances Do Infants Predict the Goal of Others’ Action: a Review

• In Event: Poster Session 07

Abstract

Background
Infants show the ability to predict simple human actions in the first year of life. This ability has been postulated as fundamental to the development of social cognition in childhood. Previous research has applied online predictive gaze measures to understand whether infants predict the goal of others’ action under different action types (e.g. reaching, feeding), physical properties of the interacting objects (e.g. saliency), action end effect (e.g. sound), and continuity of the action. However, there is no systematic review accessing the overall statistical evidence and how predictive gaze is influenced by task methodology. Considering high variability and small sample sizes in infant research are common, the current review employed the p-curve analysis to evaluate the evidential value while taking selection bias into account.

Methods
The current review included 21 studies. Fifty-six independent samples were identified, consisting of 933 infants (age range: 4 to 25 months old). Eligibility criteria were (1) typical developing participants under three years old, (2) online predictive gaze behavior as the dependent variable, (3) manual actions performed by human agents, and (4) articles published in peer review journals. To access the evidential value, mean action prediction latency across samples was extracted. In the developmental literature, one sample t-test using the latency value against 0 ms is often used to classify whether an infant shows predictive gaze behavior. Thus, we pooled t-values computed from the mean latencies for the p-curve analysis. P-curve refers to the distribution of statistically significant p-values from a set of independent samples. A right-skewed distribution suggests that a true effect exists (Simonsohn et al., 2014). Furthermore, to access the influence of task methodology, each t-value was converted to Hedges’ g. Using Hedges’ g as a dependent variable, four methodological factors were analyzed using multiple regressions. The first three factors were action type, action end effect, and continuity of action. The last factor, task difficulty was coded based on the hypothesis of each study. Taking the saliency of the interacting object as an example, a highly salient object has a lower degree of difficulty compared to a low salient one.

Results
The result of the p-curve shows that 47 out of 56 p-values were significant and the curve was significantly right-skewed (estimated power 99%). For the influence of task methodology, the initial model of multiple regression model showed that both continuity of action and task difficulty were related to the results of action prediction after controlling age (F(6, 49) = 2.681, p = 0.02 adjusted R2 = 0.1549). After backward stepwise elimination, the model (F(3, 52) = 5.304, p = 0.003 adjusted R2 = 0.190) was consistent with the initial model.

Conclusions
Our results indicate the evidential value of action prediction in infancy in the existing literature is adequate. Task difficulty and whether the observed action is continuous or discontinuous significantly influence predictive gaze behaviors. We discuss the implications of task difficulty and continuity of action that influence infants’ and children’s social understanding and interaction in real life.

Authors

• Hsing-Fen Tu, Uppsala universitet
  Presenting Author

• Terje Falck-Ytter

• Gustaf Gredebäck, Uppsala universitet