Jude C. Hays
Department of Political Science, University of Illinois at Urbana-Champaign, Urbana-Champaign, IL 61801 e-mail: jchays{at}uiuc.edu
e-mail: franzese{at}umich.edu (corresponding author) In this paper, we demonstrate the econometric consequences ofdifferent specification and estimation choices in the analysisof spatially interdependent data and show how to calculate andpresent spatial effect estimates substantively. We considerfour common estimatorsnonspatial OLS, spatial OLS, spatial2SLS, and spatial ML. We examine analytically the respectiveomitted-variable and simultaneity biases of nonspatial OLS andspatial OLS in the simplest case and then evaluate the performanceof all four estimators in bias, efficiency, and SE accuracyterms under more realistic conditions using Monte Carlo experiments.We provide empirical illustration, showing how to calculateand present spatial effect estimates effectively, using dataon European governments' active labor market expenditures. Ourmain conclusions are that spatial OLS, despite its simultaneity,performs acceptably under low-to-moderate interdependence strengthand reasonable sample dimensions. Spatial 2SLS or spatial MLmay be advised for other conditions, but, unless interdependenceis truly absent or minuscule, any of the spatial estimatorsunambiguously, and often dramatically, dominates on all threecriteria the nonspatial OLS commonly used currently in empiricalwork in political science.
Authors' note: This research was supported in part by NationalScience Foundation grant no. 0318045. We thank Chris Achen,Jim Alt, Kenichi Ariga, Neal Beck, Jake Bowers, Kerwin Charles,Bryce Corrigan, Tom Cusack, David Darmofal, Jakob de Haan, JohnDinardo, Zach Elkins, John Freeman, Fabrizio Gilardi, KristianGleditsch, Mark Hallerberg, John Jackson, Aya Kachi, JonathanKatz, Mark Kayser, Achim Kemmerling, Gary King, Hasan Kirmanoglu,James Kuklinski, Tse-Min Lin, Xiaobo Lu, Walter Mebane, CovadongaMeseguer, Michael Peress, Thomas Pluemper, Dennis Quinn, MeganReif, Frances Rosenbluth, Ken Scheve, Phil Schrodt, Beth Simmons,Duane Swank, Wendy Tam Cho, Craig Volden, Michael Ward, andGregory J. Wawro for useful comments on this and/or other workin our broader project on spatial econometric models in politicalscience. Bryce Corrigan, Aya Kachi, and Xiaobo Lu each providedexcellent research assistance and Kristian Gleditsch, Mark Hallerberg,and Duane Swank also generously shared data. We alone are responsiblefor any errors.
相似文献