Statistical Research

Eunice Kennedy Shriver National Institute Of Child Health & Human Development

Investigators

Linked publications & trials

Abstract

Statistical Methodological Projects[unreadable] [unreadable] [unreadable] [unreadable] The Density Ratio Model and the Nonparametric Behrens-Fisher Problem[unreadable] [unreadable] Dr. Troendle, along with Dr. Konstantinos Fokianos, has studied the application of a semi-parametric approach to the classical nonparametric two-sample problem. Empirical likelihood is used to fit a density ratio model to the two-sample data which has been transformed using a Box-Cox transformation. The tests are compared by simulation and found to be generally more powerful than the fully nonparametric test. A paper is in press for Statistical Modeling. [unreadable] [unreadable] [unreadable] Two-way Heteroscedastic ANOVA Model[unreadable] [unreadable] Dr. Troendle has provided an empirical likelihood based testing method for the group effect in a general heteroscedastic two-way ANOVA model. The method used an approximation to the empirical likelihood ratio test. The method is shown by simulation to have robust testing properties, making it appropriate for virtually all data types. The method has been used to analyze folate levels among two alcohol intake groups while accounting for gender, an important analysis for the Birth Defects Research Group. A manuscript has been submitted.[unreadable] [unreadable] [unreadable] Tests of Genetic Association Using Triads[unreadable] [unreadable] Stemming from his collaboration with Jim Mills and the Birth Defects Study Group, Dr. Troendle has been studying methods of testing for genetic association with disease. One topic is allowing for genotyping errors in the analysis of triad data (mother, father, case child). A method has been derived, appropriate for general genetic disease models, which allows such tests. A manuscript has been submitted. [unreadable] [unreadable] A second project is to extend current methods based on the conditional likelihood to allow restrictions on the parameter space. In genetic association studies, it is often thought that the relative risks of one versus two copies of a risk allele are not completely free to vary. Different restrictions could reasonably be placed on the parameter space, resulting in more powerful tests. This ongoing project will explore such tests and their usefulness. [unreadable] [unreadable] [unreadable] Multiple Testing With Multiple Doses and Multiple Ordered Endpoints[unreadable] [unreadable] Dr. Troendle has investigated methods of incorporating correlation into tests of multiple doses with multiple ordered endpoints. In clinical trial applications where multiple endpoints are tested using a gatekeeper approach, existing methods rely on Bonferroni methods that do not exploit the correlation among the test statistics. Simulation from multivariate normal or resampling methods can be used to increase power while maintaining familywise error rate control. A manuscript is under preparation.