Statistical power in forest monitoring

Abstract

Statistical power, the ability to detect a specified effect size (ES) (difference among treatments or change over time), if an effect exists, is low in most monitoring studies. A priori power analysis estimates the power of a particular experimental design and answers the following questions. (1) What is the smallest ES that can be detected? (2) How many samples are needed to detect effect sizes of biological or management significance? (3) What is the risk of wrongly accepting a false null hypothesis (Type II error) at a given risk of wrongly rejecting a true null hypothesis (Type I error)? When changes over time are gradual (as during silvicultural conversion of even-aged, second-growth forests to an uneven-aged condition) and resources are limited, power analysis helps allocate sampling effort to maximize the probability of detecting small effect sizes. The power of any particular experimental design can be enhanced by increasing sample size, increasing acceptable Type I error risk, and, in analysis of variance, specifying planned means comparisons. Alternatively, power can be increased by changing experimental design to reduce residual variance and/or increase ES. Use of parametric and one-tailed statistical tests also increases power. The forest monitoring program at Fort Lewis, a military installation, provides an example of the application of power analysis.