Abstract
Two bootstrap procedures are introduced into the hybrid of the backfitting algorithm and the Cochrane-Orcutt procedure in the estimation of a spatial-temporal model. The use of time blocks of consecutive observations in resampling steps proved to be optimal in terms of stability and efficiency of estimates. Between iterations, there were minimal changes in the empirical distributions of the parameter estimates associated with the covariate and temporal effects indicating convergence of the algorithm. Crop yield data are used to illustrate the proposed methods. The simulation study indicated that prediction error from the fitted model (estimated from either Method 1 or Method 2) is very low. Also, the prediction error is relatively robust to the number of spatial units and the number of time points.
Original language | English |
---|---|
Pages (from-to) | 809-822 |
Number of pages | 14 |
Journal | Journal of Statistical Computation and Simulation |
Volume | 80 |
Issue number | 7 |
DOIs | |
Publication status | Published - Jul 2010 |
Externally published | Yes |
Keywords
- Additive model
- Backfitting
- Bootstrap
- Spatial-temporal model