PREDICTION OF GROSS TREE VOLUME USING REGRESSION-MODELS WITH NONNORMAL ERROR DISTRIBUTIONS

Previous work in weighted linear regression, where weight functions ar e used to obtain homogeneous variance on a transformed scale, has ofte n assumed that the errors are normally distributed. In a study of four data sets, three of which were actual data sets with unknown error di stributions and one an artificial set with a known error distribution, this assumption is incorrect. Consequently, we tested a transformatio n of the normal distribution, called the S-U distribution, and compare d it with the normal as an alternative, For three of the four data set s studied,the S-U distribution was superior. Prediction intervals and biases for the regression estimators generated using the S-U and norma l distributions were also evaluated, Results for the S-U distribution bettered those for the normal distribution in three of the four data s ets. For the remaining data set, they were comparable.