© 1971 by Biometrika Trust
Bias and accuracy of parameter estimates in a quantal response model
University of Exeter
For a quantal response situation involving a total of n subjects, expressions are derived for biases correct to order n1 and covariances correct to order n2. These are applied to the standard probit model and it is shown that in this situation, first order biases can be appreciable, even for moderate n, and that the usual asymptotic results for variances can be considerably in error.
Key Words: Asymptotic properties of maximum likelihood estimates Quantal response model Binary response model Probit analysis