Abstract
Two procedures for estimating the mean squared error of prediction of an empirically determined linear prediction equation are examined. The method usually employed makes use of a second validation sample; another method makes use of the calibration sample alone. The mean squared error of estimation is derived for each of the two estimation procedures and a comparison made. A test is provided also for the hypothesis that use of a prespecified subset of predictors results in no increase in the expected mean squared error of prediction.