Conducting tests for statistically significant differences using forest inventory data
Many forest inventory and monitoring programs are based on a sample of ground plots from which estimates of forest resources are derived. In addition to evaluating metrics such as number of trees or amount of cubic wood volume, it is often desirable to make comparisons between resource attributes. To properly conduct statistical tests for differences, it is imperative that analysts fully understand the underlying sampling design and estimation methods, particularly identifying situations where the estimates being compared do not arise from independent samples. Information from the Forest Inventory and Analysis (FIA) program of the U.S. Forest Service was used to demonstrate circumstances where samples were not independent, and correct calculation of the standard error (and associated confidence intervals) required accounting for covariance. Failure to include the covariance when making comparisons between attributes resulted in standard errors that were too small. Conversely, comparisons of the same attribute at two points in time suffered from exaggerated standard errors when the covariance was excluded. The results indicated the effect of the covariance depends on the attribute of interest as well as the structure of the population being sampled.