THE STRENGTH OF STATISTICAL EVIDENCE FOR COMPOSITE HYPOTHESES: INFERENCE TO THE BEST EXPLANATION

摘要: A general function to quantify the weight of evidence in a sample data for one hypothesis over another is derived from law likelihood and statistical formalization inference best explanation. For fixed parameter interest, resulting that favors composite ratio using value consistent with each maximizes interest. Since generally only known up nuisance parameter, it approximated by replacing reduced func- tion on interest space. The has both interpretability Bayes factor objectivity p-value. In addition, coherent sense cannot support any entails. Further, when comparing lies outside non-trivial interval within interval, proposed method weighing almost always asymptoti- cally correct under mild regularity conditions. Even at small sizes, simple an substan- tially reduces probability observing misleading evidence. Sensitivity hypotheses' specification mitigated making them impre- cise. methodology illustrated multiple comparisons setting gene expression microarray data, issues simultaneous multiplicity are addressed.