Developing a Magic Number Four, Plus or Minus Two: The Dynamic Field Theory Reveals Why Visual Working Memory Capacity Estimates Differ Across Tasks and Development Vanessa R. Simmering 1 & John P. Spencer 1,2 (vanessa-simmering@uiowa.edu, john-spencer@uiowa.edu) 1 Department of Psychology, 2 Iowa Center for Developmental and Learning Sciences Iowa City, IA 52240 USA Abstract In the study of visual working memory, different tasks have been used to assess capacity across development: preferential looking in infancy, and change detection in children and adults. These different tasks have produced inconsistent results over development, showing higher capacity at 10 months than at 5 years. To reconcile these differences, we developed a new version of change detection to use with younger children (ages 3-5 years), and tested children in both types of tasks. Results indicate that preferential looking leads to higher estimates of capacity than change detection. In addition, capacity estimates from preferential looking were higher than the standard 3-4 items found in change detection with adults. This difference across tasks and development is explained by implementing both tasks in a single theoretical framework, the Dynamic Field Theory. Keywords: development; visual working memory; capacity; dynamic field model Visual Working Memory Capacity The past decade has seen an explosion of interest in the processes that underlie visual working memory (VWM) and the capacity of this memory system in adults (for historical precedent, see Miller, 1956). Beginning with Luck & Vogel’s (1997) seminal paper, change detection has been the method of choice for testing the capacity of VWM. Figure 1 shows a standard change detection trial. First, a memory array is shown briefly (100-500 ms) to the participant, followed by a brief delay (250-900 ms). Then, a test array appears in which either all of the objects match the memory array, or one object has changed to a new feature value. The participant reports whether there was a change in the display. Findings across a host of studies have shown that adults have a capacity for approximately 3-4 simple items (e.g., colored squares or oriented bars). Although these general findings have been replicated widely, they are not without controversy. For example, recent studies have shown that the details of the object features held in working memory can limit the capacity of this memory system, suggesting that working memory utilizes a pool of limited resources that can modify the number of “slots” available in a given task (Alvarez & Cavanagh, 2004). Although the exact nature of the cognitive representations that underlie VWM is under debate, researchers agree that VWM is a capacity-limited system. A relatively new question for VWM researchers is the developmental origin of such limitations. This question has been addressed in the domain of verbal working memory, where children show improvement over development in tasks like digit span. Interestingly, such limited capacity might have adaptive consequences early in development. For instance, in language development, the “less-is-more” hypothesis (e.g., Newport, 1990) suggests that children’s limited memory of variable input actually helps them learn a regularized grammar. Recently, researchers have begun to explore whether there is a comparable developmental change in VWM capacity. To explore developmental changes in VWM, researchers have taken two approaches. First, some studies have tested children in the standard change detection task, with some modification (Riggs, McTaggart, Simpson, & Freeman, 2006; Cowan, Naveh-Benjamin, Kilb, & Saults, 2006). These studies have shown that children have a lower memory capacity than adults. Moreover, capacity increases between 5 and 12 years of age. For instance, Riggs et al (2006) used the Pashler (1988) model and obtained capacity estimates of 1.52 items for 5-year-olds, 2.89 items for 7- year-olds, and 3.83 items for 10-year-olds. Cowan et al. (2006) also found improvements in performance in change detection between 8 and 10 years of age. 1 There have also been new procedures developed to study the development of VWM capacity. One notable line of experiments tested infants’ VWM capacity in a preferential looking task (e.g., Ross-Sheehy, Oakes, & Luck, 2003). A schematic of this task is shown in Figure 2. Infants are seated in a parent’s lap facing two computer monitors. Each monitor shows a stream of images, with arrays comparable to those used in the adult task separated by blank “delay” screens. The timing of these events is designed to approximate the timing of the adult task. On one screen (randomly determined across trials), one of the items in the array will change after each delay. The amount of time the 1 Cowan et al. (2006) did not compute capacity directly, but did report developmental improvements in children’s overall percent correct, which corresponds closely to capacity estimates. Sample array Delay Test array Response “change” Figure 1. Events in a single change-detection trial.