Retransformation bias in a stem profile model

Abstract

An unbiased profile model, fit to diameter divided by diameter at breast height, overestimated volume of 5.3-m log sections by 0.5 to 3.5%. Another unbiased profile model, fit to squared diameter divided by squared diameter at breast height, underestimated bole diameters by 0.2 to 2.1%. These biases are caused by retransformation of the predicted dependent variable; the degree of retransformation bias depends upon choice of dependent variable in the regression model, variance of its prediction errors, and the bole position of the desired prediction. Retransformation biases were greatest near the merchantable top of large trees. Equations are given that reduce the magnitude of these biases, but accurate variance models are required. Additional biases are identified for more complex transformations of stem profile models.

Citation

Czaplewski, Raymond L.; Bruce, David. 1990. Retransformation bias in a stem profile model. Canadian Journal of Forest Research. 20(10): 1623-1630.