|
Activity Number:
|
514
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #309263 |
|
Title:
|
A Method for Bias-Reduction of Sample-Based MLE of the Autologistic Function
|
|
Author(s):
|
Steen Magnussen*+ and Robert Reeves
|
|
Companies:
|
Canadian Forest Service and Queensland University of Technology
|
|
Address:
|
506 West Burnside Rd, Victoria, BC, V8Z 1M5, Canada
|
|
Keywords:
|
autologistic function ; cluster sampling ; maximum likelihood ; bias ; bias correction
|
|
Abstract:
|
Sample-based MLE of the autologistic function are biased due to the lack of independence of sampling units (clusters). An earlier study (Magnussen and Reeves 2007, Journal of Applied Statistics, in press) quantified the bias and the properties of sample-based MLE estimates in an extensive simulation study. Adding a buffer of one row of ultimate sampling unit around each sampled unit (cluster) with 'missing data' and assuming that these buffered units are mutually independent a less biased MLE of the autologistic function can be obtained. The bias reduction depends on the number of ultimate units (m) in a square sampling unit. For m = 4 adding a buffer lowered bias from 8.1% to 6.1% while for m = 7 it was lowered from 5.3% to 4.3%. We recommend the buffering since it is easy to implement without adding significantly to the computational effort.
|
