Dynamic Interactions Between Memory and Viewing Behaviors: Insights From Dyadic Modeling of Eye Movements

Humans use eye movements to build visual memories. We investigated how the contributions of specific viewing behaviors to memory formation evolve over individual study epochs. We used dyadic modeling to explain performance on a spatial reconstruction task based on interactions among two gaze measures: (a) the entropy of the scanpath and (b) the frequency of item-to-item gaze transitions. To measure these interactions, our hypothesized model included causal pathways by which early-trial viewing behaviors impacted subsequent memory via downstream effects on later viewing. We found that lower scanpath entropy throughout the trial predicted better memory performance. By contrast, the effect of item-toitem transition frequency changed from negative to positive as the trial progressed. The model also revealed multiple pathways by which early-trial viewing dynamically altered late-trial viewing, thereby impacting memory indirectly. Finally, individual differences in scores on an independent measure of memory ability were found to predict viewing effectiveness, and viewing behaviors partially mediated the relation between memory ability and reconstruction accuracy. In a second experiment, the model showed a good fit for an independent dataset. These results highlight the dynamic nature of memory formation and suggest that the order in which eye movements occur can critically determine their effectiveness.