[Openspace] Distance Decay Weights

[Julia Koschinsky]

on behalf of Larry.

-----Original Message----- 
From: Larry Plummer 
Sent: Tuesday, April 08, 2008 1:29 PM 
To: 'openspace at sal.uiuc.edu' 
Subject: Distance Decay Weights 

Dear Openlisters, 

With apologies, I could use some help revisiting what I gather is an "old" topic.  I 
say old because I have found some previous comments on the openspace list 
from 2005 on this topic.  I'd like to know if the answer has changed since then. 

How would I test the hypothesis that the friction parameter of the spatial 
weights matrix equals 2 (i.e., that the spatial correlation exhibits gravity type 
decay)?  Is there a "definitive" reference on the subject or is this is this still an 
unresolved issue?  I understand that one approach is to specify the weights 
matrix with varying values for the friction parameter and then accept the matrix 
with the highest likelihood that also satisfies the Wald > LR > LM inequality in 
the specification tests.  However, I also note from Dr. Anselin's "Under the 
Hood" paper that, "when both the original values (xk) and the spatial lag (Wxk) 
for the same variable are included [in a regression], tests for the importance of 
distance decay can be performed (distance decay implies gk < bk)."  These two 
approaches seem to get me only part way to a solution to my problem.  

Second, in gauging the goodness-of-fit, how critical is it that the W > LR >  LM 
inequality apply? 

Many thanks! 
Larry 

----------------------------------------------------- 
Larry Plummer, PhD 
Assistant Professor of Management 
(Strategy and Entrepreneurship) 
Clemson University 
Department of Management 
101 Sirrine Hall 
Clemson, SC  29634 
864-656-3763 Tel 
864-656-2015 Fax 
lplumme at clemson.edu