Statistical properties of forced-choice psychometric functions: Implications of probit analysis