 |
|
|
March 2nd, 2007
Location: Zeus Grill and Seafood
6:30 pm Registration
Incorporating Practicality into Statistically Optimal Designs: An
Application of the Penalized Optimality Criterion
While experimental designs based on statistical design criteria have statistically optimal properties, there may be practical problems associated with their implementation. As a result, investigators often abandon these designs in favor of the designs traditionally used in the field, with the potential for a significant reduction in the quality of the statistical properties. The penalized optimality criterion (Parker and Gennings, JABES, under review) incorporates the investigator*s design preferences using desirability functions to penalize impractical designs. Dose response data from Crofton et al. (2005) will be used to illustrate the methodology. Crofton et al. conducted a study of 18 polyhalogenated aromatic hydrocarbons (PHAHs) on serum total thyroxine (T4). Young female Long Evans rats were dosed with the 18 single agents or a fixed ratio mixture, and serum total T4 was measured via radioimmunoassay. The initial analysis found significant interaction among the chemicals, with evidence of synergy at high doses (Crofton et al., 2005). To address the subsequent question of dose dependency, Gennings et al. (2007) fit an interaction threshold model to the data. The resulting estimate of the interaction threshold was positive and within the observed dose region, however, the corresponding confidence interval was wide and included zero. In order to more precisely estimate the location of the interaction threshold, second stage designs were determined using both the Ds , and the penalized Ds , optimality criteria.
Contact Matteo Bottai for more information.
|
[South Carolina Chapter Homepage] Copyright ©2005 American Statistical Association |
Modified February 2, 2005 |