Abstract:
|
Results from tests for dose-response trend are often used to support conclusions on toxicity of a substance, which may lead to quantitative use of the dose-response curve. A substantial literature has developed, comparing alternative trend tests. Here, power comparisons are displayed for some single- and multiple-contrast tests for trend in mean response for a 5-group design, an assortment of monotone curve shapes, and a maximum effect (compared to controls) equal to 1 within-group standard deviation. We implement a simple algorithm for generating a diverse array of curve shapes. Resort to basic Monte Carlo simulation is avoided due to advances in multiple numerical integration as implemented in current software. Our results are consistent with previously-reported results with respect to dependence of relative sensitivity of alternative tests on curve shape. However, such results may at least narrow the selection of a default procedure.
|