Examination of İnfant Mortality Risk in Turkey With Spatio-Temporal Bayesian Models

Abstract

In this thesis, it was aimed to determine relative risk (RR) of infant mortality for each province in Turkey between 2009 and 2017 years by including the concepts of space, time and space-time interaction, to obtain risk maps, and to examine the risk factors affecting. Spatio-temporal Bayesian models were implemented to estimate RR with integrated nested Laplace approximation in R software. Structured spatial and temporal random effects on RR were modeled with Gaussian Markov random fields, using intrinsic conditional autoregressive structure and random walk model, respectively. The best model was determined according to deviance information criterion (DIC). The major contribution to variability of RR explained with the best model was from unstructured spatial and structured temporal interaction random effect. From 2009 to 2017 the number of provinces with high RR, decreased. From 2009 to 2017 in each year consistently; significant risk areas clustered in eastern and southeastern Anatolia regions. Effects of gross domestic product (GDP) per capita, percentage of mothers aged under 20 and percentage of mothers aged over 39 on RR of infant mortality were examined with generalized linear model without concepts of space and time and with the best spatio-temporal Bayesian model. As GDP per capita increased, RR decreased for generalized model and spatio-temporal model. Whereas percentage of mothers aged under 20 and percentage of mothers aged over 39 increased, RR increased for generalized model. But percentage of mothers aged under 20 and percentage of mothers aged over 39 had no effect on RR for spatio-temporal model. Spatio-temporal Bayesian model can be more preferable than generalized model, because of having lower DIC than generalized model. Therefore, while determining the factors that may have an effect on RR of infant mortality, it is also important to consider the effects of space, time and space-time interaction.