Advancing Estimation Of Timber Products Output In The Lake States Region Of The Northern United States

Abstract

The Forest Inventory and Analysis (FIA) – Timber Products Output (TPO) program has chosen to implement a sampling design for collecting information from primary wood-using facilities across the US. Sample-based approaches are often clear alternatives to surveys, as they offer estimates with increased precision and quantifiable error at lower costs and greater speed. Coulston et al. (2018) has selected a unique stratified random sampling design, which separates mills into a “certainty” or “uncertainty” sample based on a measure of size (MOS). However, the new design is in its early stages and needs for developing the efficiency of the design have been identified. This research assesses the advancement of two key areas: the selection of an effective MOS and the identification of a threshold for allocating mills into a “certainty” sample. When sampling highly-skewed populations, a few large units may account for large portions of the mean estimated and incorrectly accounting for these units can negatively impact the precision of estimation. Systematic identification of a certainty threshold was assessed through methods inspired by the work of Glasser (1962) and Hidiroglou (1985). Estimates produced by these methods were analyzed against historic TPO data to assess for overall impact. MOS also alter the precision, and as the correlation between the variable of interest and the MOS increases the level of uncertainty tends to decrease. Sources for gathering auxiliary mill metrics were explored and relevant attributes were combined to create MOS using three separate techniques: correlation comparison, simple linear regression, and multiple regression. The implementation of different MOS and threshold identification techniques, their impact on sampling efficiency, and potential areas of further research are assessed.