Sample Size and the Accuracy of Predictions Made from Multiple Regression Equations

Some rules of thumb are given for estimating the accuracy of predictions based on a multiple regression equation developed from a random sample of a multivariate normal population. The distribution of the prediction error in this case can be approximated usefully by a normal distribution. Formulas are given for the moments of the distribution and for other parameters such as the mean absolute error (MAE). The approximate inflation in MAE (over its asymptotic value) due to estimating the regression coefficients is a simple function of the base sample size and the number of predictors.