Modeling life expectancies : A spatial analysis

Abstract: In the present paper, we examine the effect of socioeconomic characteristics on the life expectancy of men and women in the Stockholm metropolitan area. Detailed individual data allows for a novel approach where observations can be displayed in high resolution. As is often the case with geographical data, the variables display high spatial autocorrelations, which imply that observations in proximity are more, or less, similar than what could be expected under the assumption of independent and identically distributed observations. Presence of spatial autocorrelation makes conventional regression models nonfunctional, and a model that accounts for this is therefore specified. In addition, a distance-band which reflects the distance and association between observations is determined. Lagrange Multiplier tests, AIC, log-likelihood, and the Schwarz criterion suggest that a spatial error model with a 300-meter distance band is appropriate for the data at hand. The findings suggest that: (1) Belonging to a minority group has the strongest effect on life expectancies and (2) the effect is negative for both genders, although the negative impact is stronger for males. Tests for spatial autocorrelation on the residuals suggest that the adopted spatial error model captures nearly all spatial autocorrelation in the data, compared to alternative models.