Optimal testing strategies for large, sparse multinomial models

摘要: Abstract Much has been written in the literature on testing for independence contingency tables. A number of related topics have studied, including choice test statistic, appropriateness asymptotic results versus exact tests, and use conditional or unconditional analyses. work to date focused relatively small The Hardy–Weinberg equilibrium (HWE) is substantive as well. Most focus loci with numbers alleles. increased genetic markers large alleles (e.g., forensic DNA profiling) quickly dating this previous work. size multinomial vectors under consideration grows quickly. Even samples will produce somewhat sparse data sets expected frequencies, calling into question. In face large, multinomials, we provide a comprehensive comparison statistics strategies (asymptotic exact, unconditional) (i.e., presence HWE) at one locus many Attained significance level power are evaluated. We find that Fisher's most appropriate samples, chi-square goodness-of-fit statistic works well relative categories. log-likelihood ratio performs poorly.