American Statistical Association New York City Metropolitan Area Chapter Memorial Sloan-Kettering Cancer Center Biostatistics Seminar

Among tools that are commonly used either for nonparametric statistical analyses or for assessing new statistical methods are: ranks, permutations, random number simulation, exact tests, and asymptotic approximations for unknown or difficult to compute distribution functions. Simple, classic examples include the chi-square approximation to the hypergeometric distribution for contingency tables, or rank tests for comparing means. However, one seldom examines how good is the approximation, or how discrete is the assumedly continuous outcome space. We can go really back to basics (combinations and permutations, method of moments, linear contrasts, ordering discrepancies from the null hypothesis, etc.) and do some really big simulations to explore some really simple alternatives.