The objective is to investigate and develop distribution-free methods in areas of application for which standard parametric techniques are inappropriate or too sensitive to violations of underlying assumptions. Much of the work of the Branch lends itself to the nonparametric approach. In sample size studies involving analysis of 2x2 tables, the determination of the minimum detectable risk for a given sample size is often required. Some asymptotic techniques have been developed in the Branch for this, but they must ultimately be validated by an exact technique which is theoretically based on the theory of randomization testing. This technique is now being developed. Another general application is the use of runs tests to evaluate residuals in regression analysis to determine goodness of fit. Research on a particularly apt nonparametric runs test, based on the variance of the length of positive and negative runs of residuals, continues. Investigation is also continuing in the use of randomization testing for comparing proportions with cluster effects.
