Resampling Variance Estimators for the k-Nearest Neighbors Technique

• Activity Number: 456
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Ronald E. McRoberts*+
• Companies: U.S. Forest Service
• Keywords: forest inventory ; satellite imagery ; bootstrap

Abstract

Nearest neighbors techniques are multivariate, non-parametric approaches to prediction based on similarity in the space of ancillary variables between population units requiring predictions and population units with observations. Variance estimators at the population unit and population level are often complex, computationally intensive, and dependent on assumptions that are difficult to verify. Resampling variance estimators based on jackknife and bootstrap methods are potential solutions. Forest inventory and Landsat satellite imagery data were used to investigate the utility of the resampling methods. In general, variance estimates based on bootstrapping were similar to those obtained using analytical techniques. However, jackknifing did not perform well.