The role of bias in a letter acuity identification task: a noisy template model
Hatem Barhoom 1, Gunnar Schmidtmann 1, Mahesh R. Joshi 1, Paul H. Artes 1, Mark A. Georgeson 2
1 University of Plymouth, Plymouth, UK
2 Aston University, Birmingham, UK
In clinical testing of visual acuity and contrast sensitivity, it is often assumed that data reflect sensory performance and observers do not exhibit strong biases for or against specific letters, but this assumption has not been extensively tested. We measured letter identification performance against letter size, spanning the resolution threshold, for 10 Sloan letters at central and paracentral visual field locations, for 10 naïve observers, using the method of single stimuli. Results showed that observers had individually different letter biases consistent across the whole range of sizes. Preferred letters were called more often and others less often than expected (group averages from 4% to 20% across letters, where the unbiased rate was 10%). A “noisy template” model was used to distinguish biases from differences in visibility. A model with varying bias across letter templates fits much better than one with sensitivity variation, but the best model combined both - having substantial biases and small variations in sensitivity across letters. The over- and under-calling decreased at larger letter sizes, but was well-predicted by templates having fixed additive response bias: with stronger inputs (larger letters) there is less opportunity for bias to influence which template gives the biggest response. The neural basis for such bias is not yet known, but a plausible candidate is the letter-recognition machinery of the temporal lobe. In the future, it will be important to investigate whether the observed response biases are likely to have a meaningful effect on clinical measures of visual performance.
