|
Activity Number:
|
471
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #310335 |
|
Title:
|
Improving the Delta Method
|
|
Author(s):
|
Paul Duty*+ and Nancy Flournoy
|
|
Companies:
|
Montgomery College and University of Missouri-Columbia
|
|
Address:
|
51 Mannakee Street, Rockville, MD, 20850,
|
|
Keywords:
|
confidence interval estimation ; coverage improvement ; statistical theory ; rational functions ; logistic model ; quantal estimation
|
|
Abstract:
|
Confidence intervals for nonlinear functions of parameters, when based on the Delta Method, seem prone to undercoverage. We propose a strategy for constructing confidence intervals that we call the Composition Method. The Composition Method is a completely analytical approach that produces asymmetric intervals that have higher coverage than the Delta Method interval for many nonlinear functions, including rational functions. We show why Composition Method intervals have higher coverage. Analytical comparisons are then made between Composition Method intervals and the Delta Method interval. Finally, we compare the coverage between confidence intervals based on the square root and logarithmic compositions with the coverage of the Delta Method interval for the quantal from a dose-response function.
|
