APPLICATION OF REGRESSION-ANALYSIS TO INVENTORY DATA WITH MEASUREMENTS ON SUCCESSIVE OCCASIONS

Successive measurements from permanent forest plots have often been ma de as part of forest inventories. The data have then been used with re gression analysis to develop forest yield prediction models. When ordi nary least-squares (OLS) regression is used with such data sets, the p arameter estimator is unbiased, but the estimator of the covariance ma trix is negatively biased. This prevents the use of normal hypothesis tests of the fitted regression. This paper considers these problems an d various approaches that have been used to deal with them. The use of a two-stage model building procedure is advocated to help find an app ropriate model form to be used for the regression. It is concluded tha t the use of permanent sample plots with many remeasurements should be limited in collecting data for growth and yield model development and used primarily to assist in determining an appropriate form for the g rowth model.