A New Test for Spatial Autocorrelation with an Application to Mortality in Swedish Municipalities

• Activity Number: 348
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Henrik Holmberg*+ and Erling Haggstom Lundevaller
• Companies: Umeå University and Umeå University
• Keywords: spatial ; global correlation test

Abstract

Background. When analyzing data in a spatial context, it is best to account for the possibility of finding spatial autocorrelation. Several tests have been proposed. However, the tests are, in general, sensitive to heteroscedasticity and should therefore be uses with caution. Objectives. The aim of this paper is to propose a new test that works well for detecting spatial autocorrelation that is robust against heteroscedasticity and that is easy to compute. Methods. We derive such a test while assuming that the mean function for the model is known, and we perform simulations for this case and for the residuals to investigate the tests finite sample poperties. As a reference, we use the Moran's I test. The application of the test is illustrated by investigating the spatial autocorrelation of mortality in Swedish municipalities. Results. In the simulations, we show that our test works well under heteroscedasticity with structure and also performs well if there is no heteroscedasticity, given that a sufficient number of observations are available.