Exact Trend Tests for Dose-Response Data: Comparisons Under Different Models

• Activity Number: 423
• Type: Contributed
• Date/Time: Wednesday, August 6, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Man Lai Tang*+ and Hon Keung Tony Ng
• Companies: Harvard University Medical School and Southern Methodist University
• Keywords: trend test ; dose-response data ; exact conditional approach ; exact unconditional approach

Abstract

Exact methods for small-sample dose-response studies with binary response will be investigated. For exact conditional approach, nuisance parameters (e.g., the intercept) are factored out by conditioning on their reduced sufficient statistics (e.g., marginal row totals for the intercept under the logit link). Unfortunately, even for simple binary dose-response study exact conditional approach is inapplicable except for logit model. For exact unconditional approach, nuisance parameters are eliminated by using a 'worst-case' scenario. For example, the p value is a tail probability maximized over all possible values for the nuisance parameters. The applicability of exact unconditional approach can readily extend to more general models including logit, probit, one-hit and extreme-value. Comparisons are conducted using the well-known Cochran-Armitage trend test under different model specifications. Our simple empirical studies clearly show that the exact conditional approach is generally inferior to the exact unconditional approach with respect to actual significance level and exact power.