Confidence Limits for Seeding Effect in Single-Area Weather Modification Experiments

Abstract

A recurring problem in single-area randomized seeding experiments has been the assessment of statistical significance over all experimental units, such as days, when some of the units receive no precipitation. The present paper solves the problem in two ways: 1) the likelihood ratio test of the complete model including both zero and positive observations and 2) an approximate confidence interval for the ratio of mean values over all seeded and non-seeded experimental units. The results are obtained for both log-normal and gamma distributions. They are illustrated by numerical examples from the 1972?74 National Hail Research Experiment randomized seeding experiment.