Abstract
Human observers can accurately report summary statistics of features varying across stimuli within an ensemble. However, the literature on ensemble coding has yet to resolved a crucial question: Is it necessary to encode each individual item in visual short-term memory (VSTM), or can summaries be extracted earlier, during initial processing and iconic memory? To evaluate this distinction, participants performed an orientation averaging task on a display that consisted of 15 bisected circles presented across two 10ms presentations, with a variable ISI between. Participants then recreated what they believed to be the average of the 15 circles by rotating a probe circle. Although the stimulus presentation times were within the parameters of iconic memory displays, and too brief for saccades to individual items, participants gave significantly precise reproductions of central tendency. We then compared the results of the averaging task to two item-specific tasks, temporal integration and segregation, conducted using the same stimuli and within the same experimental session. Results revealed that while the performance of the participants for the item-specific tasks was strongly dependent on the ISI, as previously reported, the averaging task was equally precise across the ISIs. These results suggest that central tendency information can be extracted rapidly in the absence of (and hence, prior to) the entry of items into STM. This ability to quickly process “the forest” rather than the individual “trees” may allow central tendency representations to be extracted regardless of whether individual item information is preserved in VSTM.