Abstract
The efficiency with which observers discriminate faces declines with age. For example, using the classification image (CI) method with sub-sampled faces, Creighton et al (VSS 2014) showed that older adults sampled diagnostic face information less efficiently compared to younger adults. However, many older observers lacked obvious structure in their CIs, perhaps reflecting increased trial-to-trial variability in response strategy. Response consistency can be used to estimate the ratio of internal-to-external (i/e) noise affecting observers' decision. Here, we use the double-pass response consistency technique to estimate age-related changes in i/e ratios. Contrast thresholds were measured in 7 younger and 6 older observers performing a 2-AFC task for faces embedded in high external noise. In the first half of the experiment, noise fields were randomly generated on each trial, and stimulus contrast varied according to 2 interleaved staircases. In the second half of the experiment, this same sequence of face identities, contrast, and noise fields were repeated exactly, and percent correct and percent agreement (across the two halves) were calculated. The slope of the accuracy-vs.-consistency function was then used to estimate the magnitude of each observer's i/e ratio. To facilitate comparison with previous work (e.g., Gold et al, 2004; Creighton et al.), thresholds were measured with full and sub-sampled faces. Preliminary findings show higher thresholds in older than younger adults for both full- and sub-sampled faces, and this difference was greater for sub-sampled faces. The slopes of the consistency functions were slightly shallower for older than younger observers in the sub-sampled condition, suggesting increased thresholds partly reflect an age-related increase in older adults' i/e ratio. We currently are testing older observers from the original CI study to see if individual differences in response inconsistency are associated with the degree of structure observed in their CIs.
Meeting abstract presented at VSS 2018