The relative importance of local contingencies and global biases for statistical learning.

Abstract

Effective behavior requires adapting to the changing regularities evident in the world. Analogous to the global and local processing distinction for perception, these statistical regularities may be evident in global biases (i.e., some events are more likely) or local contingencies (i.e., subsequent events depend on preceding events). To explore whether mental model updating unfolds in distinct ways according to global and local statistical properties, we had healthy individuals perform two variations of an updating task in which both global and local statistical properties changed over time. Participants predicted whether the next triangle in a sequence of triangles would point up or down. The probability of pointing up or down was fixed for epochs of trials (i.e., global probability) and correlated with the colors of elements in the display. In addition, we made the triangle's apex direction on trial n+1 depend on the pointing direction of the prior trial (i.e., local probability). For both experiments, it was the local contingencies that dominated participant choices. When global and local statistical cues of equal magnitude are available, we conclude that healthy individuals are biased towards using the local statistical properties.