SPRT's for the Normal Correlation Coefficient
Julian Köllerström, G. Barrie Wetherill
Journal of the American Statistical Association · 1979 · Volume 74 · Issue 368
DOI: 10.1080/01621459.1979.10481036
Abstract
Hypothesis tests for the correlation coefficient of a bivariate normal distribution are investigated. Some tests using reductions of the data that are not invariantly sufficient were proposed by Kowalski on the grounds that the mathematical properties may be worked out, which is not possible when using sequential probability ratio tests (SPRT's) based on successive invariantly sufficient statistics. These latter tests may be derived from Cox's theorem or from maximal invariance theory. We show how inefficient Kowalski's tests are and, thus motivated, we derive approximations for the mathematical properties of Cox's test with equal variances.
Enable JavaScript for citation tools and additional paper interactions.