Study on psychometric function in forced-choice experiments

A maximum likelihood estimation method is presented to evaluate the observers' performance in visual detection tasks. The percent of correct detection is obtained as a function of visual stimulus through forced-choice experiments. Effective experimental data is selected via deviance analysis. The maximum likelihood estimation with some constraints is employed to fit the psychometric function. The psychometric functions with two and four choices experiments are established respectively, and the stimulus area covers the range from 100 to 400 square millimeters. The mean square error of the fitting is about 0.05 and the R-square is around 0.9. The maximum error of the psychometric functions between two observers is 0.0466. The results show that maximum likelihood estimation achieves a higher accuracy than the least squares nonlinear regression does.