Confidence Intervals for Binomial Probability Predictions

PDF

Authors
  1. Murdoch, D.J.
  2. Nguyen, P.
Corporate Authors
Defence R&D Canada - Atlantic, Dartmouth NS (CAN);University of Western Ontario, London Ont (CAN) Dept of Statistics and Actuarial Services
Abstract
In this report, we study methods for assessing the accuracy of detection probability estimates as a function of range to the target for sidescan sonar systems. First, we created a hypothetical model for the detection probability. Then, we evaluated the coverage probabilities and lengths of several confidence interval methods for the binomial proportion used in the traditional binning approach, simple logistic regression model and piecewise logistic regression models using binned data and cubic regression splines. Using simulation, we determined that the best approach for modeling the data collected from the sonar system was B-splines with a careful knot selection. The method is demonstrated on real data from the CITADEL sea trial. A package for use with the R statistical software is also documented.

Il y a un résumé en français ici.

Report Number
DRDC-ATLANTIC-CR-2008-248 — Contractor Report
Date of publication
01 Jan 2010
Number of Pages
62
DSTKIM No
CA034197
CANDIS No
533667
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: