EFFECTS OF MULTICOLLINEARITY AND AUTOCORRELATION ON THE VARIABLE-EXPONENT TAPER FUNCTIONS

Effects of multicollinearity, autocorrelation and two sampling strateg ies were studied on two new variable-exponent taper equations using Mo nte Carlo simulations. These two equations were derived from Kozak's 1 988 model. Results of the study indicated that predictions are unbiase d even if the taper model has severe multicollinearity and autocorrela tion. On the other hand, the estimations are considerably more variabl e when severe multicollinearity exists. The use of single observations per tree (uncorrelated errors) does not improve the predictive abilit y of the functions. Taper functions based on a uniform selection of sa mple trees predict the response variables well, but provide poorer est imates of the parameters than random selection.