Far East Journal of Theoretical Statistics
Volume 8, Issue 1, Pages 29 - 37
(September 2002)
|
|
INTERVAL ESTIMATION OF THE MEDIAN RESPONSE DOSE IN LOGISTIC
REGRESSION MODELS
Ayman S. Baklizi (Jordan)
|
Abstract: We consider interval estimation of the median response dose in
logistic regression. Intervals based on the asymptotic normality of the maximum
likelihood estimator are usually employed. Sitter and Wu [JASA
88(423) (1993), 1021-1025] showed, through some
simulations and heuristic arguments, that intervals based on Fieller's theorem [D.
Finney, Probit Analysis, 3rd ed., Cambridge University Press, Cambridge, 1971]
are generally superior. In this paper, using an extended simulation design, we
further investigate the performance of these intervals together with intervals
based on the Rao and likelihood ratio statistics. Our simulation results show
that intervals based on the likelihood ration statistic are generally the best
in terms of the attainment of the nominal error probabilities, and symmetry of
lower and upper error probabilities. |
Keywords and phrases: logistic regression, Fieller’s intervals, large sample intervals,
likelihood function. |
|
Number of Downloads: 4 | Number of Views: 818 |
|