Interobserver variance in perceptual performance and learning

Abstract

Purpose: Normal observers and patients with apparent disease usually are tacitly expected to yield homogeneous thresholds in clinical tests of visual perception. The authors tested this assumption.

Methods: Through training of 70 observers, performance and improvement of performance were tested for different hyperacuity tasks using psychophysical tests.

Results: Although the assumption of homogeneous results might be true for many tasks limited by the physical properties of the eye, such as two-point resolution, the authors 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, that is better initial performance usually is associated with slower improvement.

Conclusions: This finding indicates that retesting of unusually high (pathologic) thresholds in clinical tests of visual perception might improve discrimination between patients whose performance is poor because of lack of familiarity with the task and who improve with training and patients who improve far less if their poor performance results from pathologic conditions.