|
Activity Number:
|
248
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #305701 |
|
Title:
|
Exact Unconditional Inference for Multinomial Likelihoods with an Example Using 2x3 Contingency Tables
|
|
Author(s):
|
Gerald Crans*+ and Jonathan J. Shuster
|
|
Companies:
|
Eli Lilly and Company and University of Florida
|
|
Address:
|
389 Arbor Drive, Carmel, IN, 46032,
|
|
Keywords:
|
exact unconditional inference ; hypothesis testing ; nuisance parameters
|
|
Abstract:
|
Exact unconditional hypothesis testing methods for 2x2 binomial data, 2x2 multinomial data, and correlated proportions exhibits a superior power advantage when compared to exact conditional approaches. Comparisons beyond the 2x2 testing scenario have not been explored due to the computational complexity of maximizing null power functions that depend on several nuisance parameters. A technique to compute the maximum of null power functions that result from the comparison of independent multinomial distributions is presented. As a special case, the 2x3 contingency table comparing independent trinomial distributions is considered. Exact critical values of the Pearson chi-square test are computed for n=10(1)70 and alpha = 0.05. Sample-size comparisons between Pearsons chi-square test statistic and the conditional p-value suggest the conditional p-value tends to be superior.
|
