Abstract:
|
The evidential paradigm (EP) uses the likelihood ratio evaluated at two simple hypotheses to measure statistical evidence, with favorable operational characteristic (Royall, 2000). However, requiring a simple alternative hypothesis can limit applicability in studies such as GWAS, as there is generally no a-priori reason for choosing a specific alternative value. Bickel (2012) and Zhang & Zhang (2013) proposed to extend EP to accommodate composite hypotheses through the Generalized Likelihood Ratio (GLR). We derive the operational characteristics for GLR for fixed-dimensional parametric models asymptotically, and we show that GLR is a valid evidence function characterized by low error probabilities. Using the GLR, we conduct an association analysis of lung disease severity in 2813 samples with Cystic Fibrosis. The GLR provides confirmatory evidence of a previously identified association, but also highlights loci that strongly support no association, allowing regions of interest to be more clearly demarcated. The GLR provides a general procedure for conducting pure likelihood-based genetic studies, and avoids identifying statistically significant but clinically meaningless findings
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.