Dissociating variability from error-based processes in observational learning

Observational learning benefits are reported after watching "learning" models. Reasons for benefits are mostly speculative and the reliability of effects compared to watching correct models is relatively poor. Variation in what an observer sees when watching a learner is one proposed condition for a beneficial effect. Another is related to seeing errors, which engage error detection and correction processes. To date, variable versus errorful conditions have not been distinguished. In a pre-registered design, three groups (n = 34/group) practiced putting a golf ball to the centre of a grid and returned the next day to test for retention and transfer. Practice involved alternating blocks of watching putting videos (8 blocks of 20 trials), with physical practice, without and with outcome feedback (8 blocks, 10 trials each). Self-predictions were captured in no-feedback trials. Two "variable" groups watched a model putt to different corners of the grid, with instructions conveying intended "hits" (variable-correct) or unintentional "misses" (variable-error). A third group watched constant correct putts. Watching variable demonstrations caused more putting variability in practice and accuracy benefits in transfer, compared to constant. However, there were no group differences in retention, nor any error-group related differences. Although watching errors did not impact directional biases or facilitate putting accuracy as expected, it improved self-estimations of outcomes (i.e., error detection related to prediction of sensory effects). These data suggest that variability and errors in demonstrations have small, yet different effects on motor learning, with variability improving generalizability and errors contributing to improved feedforward predictions.