|
Abstract:
|
Objective: To estimate required sample size (number of regions) and power in a spatial analysis to identify spatial correlation and to correct potential bias due to the correlation. Method: Random maps were simulated with lattice sizes being from 16 to 100, squared and non-squared arrangement, and Rooks' and Hexagon neighborhood. Measurements (Y) in the study area were from a linear model, Y=b*X+E. Where b is fixed at 0 to 0.5; E=1/[I-p*W] *u; p is from 0 to 0.5; W is weighting matrix and u is a normal random vector. With 2000 replicated modeling, power to detect spatial correlation and impact of X was examined. Results: Without spatial correlation considered, estimate of b is unbiased; considering spatial correlation yields smaller variance of the estimate. A spatial correlation of 0.3 could be detected with 100 regions, 0.4 be detected with 64 regions and 0.5 with 36 regions.
|
