Modeling Micro-Spatial Employment Location Patterns: A Comparison of Count and Choice Approaches
Abstract:This article studies employment location patterns in the Puget Sound Region of Washington State at a micro level of geography. Traditional discrete choice modeling using multinomial logit (MNL) models may be problematic at a micro level of geography due to the high dimensionality of the set of alternative locations and the likely violations of the independence from irrelevant alternatives (IIA) assumption. Count models are free from the IIA assumption and, unlike logit models, actually benefit from large numbers of alternatives by adding degrees of freedom. This study identifies the best-fitting count model as the zero-inflated negative binomial (ZINB) model, because this model more effectively addresses the large number of cells with no jobs and reflects a dual process that facilitates the identification of threshold clustering effects such as those found in specialized employment centers. The estimation and prediction results of ZINB are compared with those of MNL with a random sampling of alternatives estimated on an equivalent data set. The ZINB and MNL models largely agree on major trends, with the ZINB model providing more insightful details, but with less capacity to predict large count situations.
Document Type: Research Article
Affiliations: 1: Public and Private Infrastructure Investment Management Center, Korea Development Institute, Seoul, Korea, 2: Evans School of Public Affairs and Department of Urban Design and Planning, University of Washington, Seattle, WA, 3: Department of Civil and Environmental Engineering, Pennsylvania State University, University Park, PA, 4: Department of Civil Engineering, Washington University in St. Louis, St. Louis, MO
Publication date: April 1, 2008