Sampling Efficiency and Internal Noise for Summary Statistics J Solomon1, P Bex2, S C Dakin3 |
|---|
1Division of Optometry and Visual Science, City University London, United Kingdom |
| Psychophysically-derived estimates of the efficiency with which observers can estimate various image statistics are of intense interest to researchers working to describe attention. High estimates of efficiency for brief displays suggest pre-attentive, parallel processing. Sampling efficiency is typically inferred from the right-hand side of threshold-vs-(external)noise (TVN) curves. Allard and Cavanagh (2012) instead concentrated on the left-hand side of the TVN curve, where external noise is low. In an orientation averaging task, they argued that observers average only discriminably different elements, giving an effective sample size no greater than 1. We consider an alternative possibility: 'Late' (internal) noise dominates the left-hand side of the TVN curve. Late noise can be defined as random fluctuations in the effective representation of *all* items in a sample, whereas early noise is defined as random fluctuations in the effective orientation of *each* item. We replicated their experiment and find better fits to our data and theirs with a model containing late noise than with their proposed model in which sampling efficiency increases with external noise. |
| Up Home |
|---|