Differences in migrants mortality: for a multilevel approach of spatial inequalities using a fine granularity. An application to Switzerland, 1990-2008
Jonathan Zufferey, University of Geneva
This paper aims to deepen the spatial inequalities of mortality among migrants living in Switzerland while some studies showed that the spatial differentials in longevity are modest among the Swiss cantons. Longitudinal data (for the period 1990-2008) on the whole population living in Switzerland linked to death certificates are used to estimate hazard ratios for each area, i.e. the ratio between observed and expected deaths according to a quasipoisson regression model. Bayesian methods are also used for the estimation of this mortality ratio in small areas, allowing us to highlight the heterogeneity at a local scale. First, we compare the spatial heterogeneity between the Swiss citizens (1'030'136 deaths for 98.2 millions pearson-years) and the foreigners (83'850 deaths for 22.4 millions person-years) across the 26 cantons of the country, taking only into account age and sex. The variance of the hazard ratios for the Swiss population is more than twice smaller than for the migrants, what means a more important spatial gradient for the migrants. When controlled for education, marital status, urbanization and origins, the spatial heterogeneity declines since partly captured by the socioeconomic variables but still stays almost twice larger for migrants. Because of the high variability within many of the 26 cantons, in a second time we analyze the spatial mortality gradient among migrants for smaller areas - the 2896 communes of Switzerland - and compare those local estimations to the mortality force at the cantonal level. Important local specificities appears. So there are significant local dynamics influencing mortality such as physical environment or sociocultural factors. This study highlights the large spatial heterogeneity of the migrant mortality at a tiny scale: the local environment seems to have, especially for migrants, an important influence which still have to be tested in a multilevel approach.
Presented in Poster Session 3