Is Tony Smith's "Estimation Bias in Spatial Models with Strongly Connected Weight Matrices" at https://doi.org/10.1111/j.1538-4632.2009.00758.x helpful?
Roger -- Roger Bivand Emeritus Professor Norwegian School of Economics Postboks 3490 Ytre Sandviken, 5045 Bergen, Norway [email protected] ________________________________________ From: R-sig-Geo <[email protected]> on behalf of Josiah Parry <[email protected]> Sent: 13 May 2024 17:12 To: [email protected] Subject: [R-sig-Geo] Maximum sparsity for spatial regression As I'm reading through Modern Spatial Econometrics in Practice, we assume the spatial weights matrix to be sparse. At one point they note that the contiguity matrix for the US counties is 0.18% non-zero. But what % non-zero is too dense? I am wondering if there is any research or papers that document what a recommended upper bound of sparsity should be for one of these models? Is 10% non-zero too much or sufficient? I suspect the answer is, like most things, "it depends." But, thinking of a situation where someone might use a distance band to specify neighbors they might create a bandwidth that can encompass 50% or more of points if using max(knn=1) to specify the distance. I suspect using a kernel or IDW could reduce the weights close to zero making the impact minimal. Nonetheless, I'm curious if others have thought about this or written about it! Thanks, Josiah [[alternative HTML version deleted]] _______________________________________________ R-sig-Geo mailing list [email protected] https://stat.ethz.ch/mailman/listinfo/r-sig-geo _______________________________________________ R-sig-Geo mailing list [email protected] https://stat.ethz.ch/mailman/listinfo/r-sig-geo
