PREDICTING STRENGTH OF MATCHED SETS OF TEST SPECIMENS

Five different methods for a priori estimating bending strength of woo d and wood composite specimens are compared in this paper. They are: ( I)edge-matching, (2) matching specimens by normal distribution, (3) ma tching specimens by log-normal distribution, (4) simple linear regress ion, and (5) multiple linear regression. It was found that the square root of mean square error (RMSE) of percent difference (PD) of predict ed modulus of rupture (MOR) is the key measure in comparing the five m ethods. Multiple linear regression was found to be the best method to predict MOR of a specimen in an edge-matched set. Finally, how to crea te the prediction limits for mean MOR of a subgroup of specimens is di scussed. The prediction limits for predicting MOR make it possible to quantitatively determine the effect of various treatments of wood mate rials.