---

Statistics in motion: Does the infant motor system predict actions based on their transitional probability?

---

Abstract

Motor theories of action prediction propose that our neural motor system combines prior knowledge about likely action outcomes with current sensory input to predict other people's behavior. This knowledge can be acquired through observational experience, more specifically statistical learning. Recently, it has been shown that infants can detect in a stream of actions two actions that follow each other deterministically and that their motor system uses this knowledge to predict upcoming actions. However, real-life actions are more complex: whereas actions hardly ever follow one another with 100% probability, often certain actions are more likely to follow one another than others (e.g., grasping a mug to drink versus to pass it to someone). Here, we examined whether infants can learn the statistical structure of action sequences through observation and whether the activity of their motor system reflects the specific statistical likelihood of upcoming actions. We trained 18-month-old infants at home with videos of action sequences featuring different transitional probabilities. At test, motor activity was measured using EEG during perceptually identical time windows that linked actions with four probability levels (100%, 75%, 50%, 25%). We found that motor activity was parametrically modulated by the transitional probability of action pairs. Specifically, our results showed the strongest predictive motor activity for deterministic actions and least activity for actions with low levels of probability. These results show that infants’ predictive motor activity reflects the specific statistical likelihood of upcoming actions and thus underline the important role of statistical learning for infants’ developing action understanding.

Folder Structure

• EEG/: MATLAB/FieldTrip scripts for EEG preprocessing, frequency extraction, topographic plots, and trial-count summaries.
• StatisticalAnalysis/: R scripts for the analyses reported in the manuscript.
• StatisticalAnalysis/model_res/: fitted Bayesian model objects used by the active Bayesian R scripts when RUN = F (model.rds and modelOcc.rds).

Main Analysis Scripts

• StatisticalAnalysis/Mu_models.R: planned linear mixed-effects analyses over central motor channels (C3, Cz, C4) and exploratory/descriptive motor plots.
• StatisticalAnalysis/Mu_cluster_comparison.R: compares alternative central electrode clusters using AIC; source for manuscript Table 1.
• StatisticalAnalysis/Mu_Bayes_model.R: exploratory Bayesian motor model with probability, channel, and trial number; source for manuscript Figure 4, Supplementary Table 1 values, and the usable-trials-over-time plot.
• StatisticalAnalysis/Mu_models_Occipital.R: planned occipital/visual-control analysis over O1, Oz, and O2; source for manuscript Figure 5.
• StatisticalAnalysis/Mu_Bayes_model_Occipital.R: exploratory Bayesian occipital/visual-control model reported in the text.

EEG Pipeline Scripts

See EEG/README.md for the MATLAB preprocessing and data-export workflow.

Notes

• The Bayesian scripts read fitted model objects from StatisticalAnalysis/model_res/ when RUN = F; set RUN = T to refit models instead.
• Figure 1 and Figure 2 in the manuscript appear to be stimulus/manual figures rather than outputs of the statistical scripts.