Developing a magic number: the dynamic field theory reveals why visual working memory capacity estimates differ across tasks and development

Many daily activities require the temporary maintenance and manipulation of information in working memory. A hallmark of this system is its limited capacity— research suggests that adults can actively maintain only about 3-4 items at once. A central question in working memory research is the source of such capacity limits. One approach to this question is to study the developmental origins of working memory capacity. Developmental working memory research has shown a general increase in capacity throughout childhood and into adulthood. There have been few investigations, however, into the mechanisms behind these developmental changes, and proposals that have been put forth do not specify the processes underlying changes in capacity. One particularly puzzling, and unexplained, finding is an apparent decrease in visual working memory capacity over development, from an adult-like capacity of 3-4 items at 10 months to 1.5 items at 5 years. One probable source of this developmental discrepancy is that these two capacity estimates were derived from different tasks: preferential looking in infancy and change detection in later childhood. Although these tasks differ in many respects, existing theoretical explanations of the processes underlying performance in these tasks are underspecified, making it difficult to identify the origin of the apparent regression in capacity over development. To investigate the developmental discrepancy across tasks, I developed a unified model to capture how capacity limits arise in preferential looking and change detection. I used this model to generate three specific behavioral predictions: 1) in preferential looking, children and adults should show higher capacity estimates than infants; 2) capacity estimates should be higher in preferential looking than in change detection when tested in the same individuals; and 3) although capacity estimates differ, performance should be correlated across tasks because both rely on the same underlying working memory system. In addition, I proposed a fourth prediction regarding the unified model: