Odile Sauzet1,2, Janne H Breiding3, Kim A Zolitschka3, Jürgen Breckenkamp3, Oliver Razum3. 1. Bielefeld School of Public Health, Bielefeld University, Bielefeld, Germany. odile.sauzet@uni-bielefeld.de. 2. Centre for statistical consulting, Bielefeld University, Bielefeld, Germany. odile.sauzet@uni-bielefeld.de. 3. Bielefeld School of Public Health, Bielefeld University, Bielefeld, Germany.
Abstract
BACKGROUND: Neighbourhood is a complex structure but of high relevance for health. Its operationalisation remains however a challenge.The aim of this work is to present a new application of the use of semi-variograms as an approach for the evaluation of spatial effects on health. For this, we propose to estimate two parameters providing a measure of an average neighbourhood or spatial effect at city level without having to predefine any notion of physical neighbourhood. METHODS: We present the statistical method to estimate the parameters of this correlation neighbourhood by fitting an exponential model to the empirical semi-variogram at short distances. With a simulation study, we show for which sample size and sampling density the method performs well and illustrate how to use the method with data from a birth cohort using the outcome birthweight. RESULTS: For small sample sizes (500) the method provides reliable estimates if the density of observations is high. For larger sample sizes other parameters influencing the quality of estimates are the maximal distance at which the semi-variograms are estimated. CONCLUSIONS: Given the complexity of spatial scales relative to neighbourhood spatial processes, our approach offers the possibility to incorporate existing approaches to the operationalisation of neighbourhood in quantitative analyses while providing a measure of the part of health inequalities which could be possibly due to unmeasured spatial exposure as well as a measure of their spatial scale.
BACKGROUND: Neighbourhood is a complex structure but of high relevance for health. Its operationalisation remains however a challenge.The aim of this work is to present a new application of the use of semi-variograms as an approach for the evaluation of spatial effects on health. For this, we propose to estimate two parameters providing a measure of an average neighbourhood or spatial effect at city level without having to predefine any notion of physical neighbourhood. METHODS: We present the statistical method to estimate the parameters of this correlation neighbourhood by fitting an exponential model to the empirical semi-variogram at short distances. With a simulation study, we show for which sample size and sampling density the method performs well and illustrate how to use the method with data from a birth cohort using the outcome birthweight. RESULTS: For small sample sizes (500) the method provides reliable estimates if the density of observations is high. For larger sample sizes other parameters influencing the quality of estimates are the maximal distance at which the semi-variograms are estimated. CONCLUSIONS: Given the complexity of spatial scales relative to neighbourhood spatial processes, our approach offers the possibility to incorporate existing approaches to the operationalisation of neighbourhood in quantitative analyses while providing a measure of the part of health inequalities which could be possibly due to unmeasured spatial exposure as well as a measure of their spatial scale.
Entities:
Keywords:
Semi-variogram; Spatial correlation structure of residuals; Spatial effects on health
Authors: Jürgen Breckenkamp; Oliver Razum; Jacob Spallek; Klaus Berger; Basile Chaix; Odile Sauzet Journal: BMC Public Health Date: 2021-07-07 Impact factor: 3.295