Canadian Forest Service Publications
Stochastic resampling techniques for quantifying error propagations in forest field experiments. 1997. Magnussen, S.; Burgess, D.M. Canadian Journal of Forest Research 27: 630-637.
Available from: Pacific Forestry Centre
Catalog ID: 4823
CFS Availability: Order paper copy (free)
Statistical analyses of forest field experiments often do not address uncertainties in model parameters and intermediate residuals and thus fail to account fully for the uncertainty inherent in the results. Stochastic resampling (bootstrap) provides a tool to integrate all known sources of variation into the final results. We demonstrate stochastic resampling of data from a red pine (Pinus resinosa Ait.) spacing trial with two spacings and four replications. Stochastic resampling resulted in higher among-plot variances of tree size and volume, which consequently lowered the significance level of pairwise t-tests of no spacing effect. The reliability of a direct analysis (no resampling) averaged 0.84. Stochastic resampling lowered the t-test statistics by an average of 18% and their significance levels by about 75%. Resampling reversed the conclusion of 1 hypothesis out of 12 of no spacing effect, and changed the significance level of 2 tests from 0.06 to >0.09. Bootstrapped volume estimates were 1-6% higher than directly computed estimates because of nonlinear transformations of residuals in the volume equations. Resampling techniques with a complete account of all relevant sources of variation hold promise for data analyses in forestry.
- Date modified: