A Method of Obtaining Prediction Intervals

Abstract A method is given for obtaining a prediction interval for a variable Y with distribution F Y (y/θ) based on observations x 1, ···, x n from distribution F x (x/θ), where θ is a common parameter to both distributions. The method uses the conditional distribution of y given t, where t is a sufficient statistic for θ in the joint distribution of (X 1, ···, X n , Y). Sample applications are given for the negative exponential and Poisson distributions.