Work in nonparametric multiple comparisons has been expanded to include contrasts as well. Joint work on the modelling of continuous proportional data has proceeded. Models for proportions with covariates have been built for which the conditional distribution of the proportions is Dirichlet, with parameters estimated by an iterative, maximum likelihood computer program. In nonparametric regression, approaches are being investigated which improve the poor predictive behavior of least square estimates when there are outliers present or when the errors are not Gaussian. The properties of several newly-devised estimators are being compared in a computer simulations study. The study of incomplete proportions is continuing. In particular, joint work has begun on the homogeneity of correlated proportions in the case where the correlation is unmeasured or unreported.
