This article proposes two approaches to the construction of one- and two-sided confidence limits for conformance proportions in a normal variance components model. One approach is based around a generalized pivotal quantity, the other around the modified large-sample estimation method. Simulation studies reveal that both methods have empirical coverage probabilities close to claimed values, making them both satisfactory. However, slightly better performance and ease of computation might make the method based on the large-sample estimation better in practical application. Three real datasets illustrate both approaches, and a bootstrap calibration approach is adopted.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados