Inter-observer variance in perceptual performance and learning


Manfred Fahle and Sigrid Henke-Fahle


Abstract

Purpose. Normal observers and patients without apparent diseases are usually tacitly expected to yield quite homogeneous thresholds in clinical tests of visual perception. We tested this assumption.

Methods. Performance and improvement of performance through training of 70 observers were tested for different hyperacuity tasks using psychophysical tests.

Results. While the assumption of homogeneous results might be true for many tasks limited by the physical properties of the eye, such as two point resolution, we find a relatively wide variation of performance especially in untrained observers for tasks that require more elaborate processing in the visual cortex. Observers vary widely both in their baseline performance and in the extent and speed of learning for tasks such as vernier discrimination and stereoscopic depth perception. On average, speed of learning is inversely correlated to baseline performance, i.e., better initial performance is usually associated with slower improvement.

Conclusions. This finding indicates that re-testing of unusually high (pathological) thresholds in clinical tests of visual perception might improve discrimination between on one hand patients with poor performance due to lack of familiarity with the task and who improve with training and on the other hand patients who will improve far less if their poor performance is due to pathologic conditions.