Русский
!

Conference publications

Abstracts

XX conference

Spatial models specification and diagnostic

Belyaeva A.V.

Institute of Control Sciences V. A. Trapeznikov Academy of Sciences Russian Federation, 117997, Moscow, Profsoyuznaya ulitsa, 65 Tel: +7 495 334-89-10, fax: +7 495 334-93-40, +7 499 234-64-26 E-mail: BelyaevaAV@gmail.com

1 pp. (accepted)

Spatial autoregressive models is a powerful method to estimate objects which characteristics are depend on their location. The method is used in various fields, for example in real estate objects appraisal. But the main problem is spatial dependence of disturbing factors by which OLS is ineffective. Standard OLS gives biased estimates of residuals variance, inflated R² and inconsistent procedures of statistical inference. Furthermore, residuals spatial correlation is distort properties of tests to determine variables to include in the model and to diagnostic residuals. To eliminate these effects we should: apply effective tests to find spatial autocorrelation (SA) in residuals and apply the methods of estimation and model specification to neutralize the negative effect of spatial disturbances if it is so.

The report is about the first part of the problem. Analysis of test effectiveness and tests comparison on SA in the residuals were conducted. Were compared: Moran test [1, 2] and the procedure to check the SA in residuals in case of SA of the dependent variable based on a modification of the Lagrange multipliers test [3]. Power of tests was compared and influence of the quality of the model specification on their work was studied by using of the statistical modeling. To verify results on real data was built several model variants for the training set and quality of each model was tested in a control set. The main quality criterion was standard deviation of model and market prices of the control set objects. The investigations revealed areas of effective uses for each test.

References

1. Moran P., Notes on continuous stochastic phenomena, Biometrika 37, 17 – 23, 1950

2. Kelejian H.H., Robinson D.P., Spatial correlation: The Cliff and Ord model and a suggested alternative, in: L. Anselin and R. Florax, eds., New directions in spatial

econometrics, Springer (Berlin), 1995

3. Luc Anselin, Anil K.Bera, Raymond Florax, Mann J.Yoon, Simple diagnostic tests for spatial dependence, Regional Science and Urban Economics 26, 77 – 104, 1996



© 2004 Designed by Lyceum of Informational Technologies №1533