Investigating Variance Estimation under Systematic Sampling with US Forestry Service Data

Presented by:
Sarah Maebius (Reed College) Alexander Flowers (Swarthmore College)

The Forest Inventory and Analysis (FIA) program of the U.S. Forest Service monitors and analyzes the nation’s forests to understand their current state and make predictions about future changes. To this end, the FIA collects data from ground plots arranged across the country using a quasi-systematic sampling design. While systematic sampling designs can be more efficient, estimating variance using this design proves challenging relative to estimating variance under simple random sampling. A common solution that the FIA uses is to assume that the data is collected under a simple random sample. In theory, however, this hurts the potential efficiency increase gained from using a systematic sample, since we do not then account for the systematic spread of data across the population. Our goal this past summer was to perform a simulation study to determine if the simple random sample assumption is indeed a problem for FIA estimation and identify potential alternative variance estimation methods. Initial findings suggest there is cause for concern, but further work is necessary to confirm that underestimation exists for the current FIA sampling method and to determine if existing alternative variance estimation methods are less biased and feasible.