Authors: |
Gregory A. Reams, Joseph M. McCollum |
Year: |
1999 |
Type: |
Scientific Journal |
Station: |
Southern Research Station |
Source: |
Reprinted from the 1999 Proceedings of the Section on Statistics and the Environment of the American Statistical Association |
Abstract
The USDA Forest Service's Southern Research Station is implementing an annualized forest survey in thirteen states. The sample design is a systematic sample of five interpenetrating grids (panels), where each panel is measured sequentially. For example, panel one information is collected in year one, and panel five in year five. The area representative and time series nature of the sample design offers increased flexibility in providing estimates of annual growth, mortality, removals and change in forest area. Restricting analyses to the most recently measured panel results in many missing cells in standard forest inventory tables. Rather than treat all unmeasured panels as missing, imputed values are used to update plots in each unmeasured panel. Because it is uncertain what analyses the ultimate users of these public-use data may engage in, we evaluate the effect and consequences of excluding important predictor variables from the imputation models.
Citation
Reams, Gregory A.; McCollum, Joseph M. 1999. Evaluating Multiple Imputation Models for the Southern Annual Forest Inventory. Reprinted from the 1999 Proceedings of the Section on Statistics and the Environment of the American Statistical Association