SADDLEPOINT APPROXIMATIONS OF MARGINAL DENSITIES AND CONFIDENCE-INTERVALS IN THE LOGISTIC-REGRESSION MEASUREMENT ERROR MODEL

We illustrate that the saddlepoint approximation can lead to accurate inference in the logistic regression measurement error model. The marg inal densities of an asymptotically unbiased estimator of the regressi on parameters are computed with a general saddlepoint approximation, a nd bias-corrected confidence intervals for the regression parameters a re derived.