| Resumo: | The aim of this work is to demonstrate two things: 1) That due to the fact that some social phenomena presents spatial variation, its modeling also presents important methodological implications, and therefore 2) it is advisable to use a regression technique that incorporates data spatiality in its design. A two-step process is followed: First, the Moran autocorrelation coefficient is calculated to demonstrate the spatial concentration of electoral behavior in urban Mexico between 1994 and 2000. Secondly, two regression techniques are compared, OLS and SAM. From the comparison it is concluded that the SAM technique is a more robust technique than the OLS, based on that the former is able to compute more accurate regression coefficients and detects significant regional effects even in the presence of models dealing with the heteroskedasticity problem.
|
|---|
