The DHS Program

Geospatial Covariates: Proxies for Mapping Urban-Related Indicators (English) 

Document Type

Spatial Analysis Reports

Publication Topic(s)

Geographic Information

Language

English

Recommended Citation

Fish, Thomas D., Bradley Janocha, Trinadh Dontamsetti, and Benjamin K. Mayala. 2020. Predicting Geospatial Covariates: Proxies for Mapping Urban-Related Indicators. DHS Spatial Analysis Reports No. 19. Rockville, Maryland, USA: ICF.

Download Citation

RIS format / Text format / Endnote format

Publication Date

September 2020

Publication ID

SAR19

Abstract:

The enumeration areas selected by The DHS Program are classified as either rural or urban. While this classification often plays a role in the analysis of health outcomes, rurality and urbanicity are not defined by The DHS Program. Instead, official urban and rural classifications are determined by the countries themselves. Lack of consensus about this definition leads to inconsistent classifications of enumeration areas as urban or rural. This inconsistency can impair comparative analyses of urbanicity’s relationship to health outcomes. There is no universally accepted taxonomy for urbanicity or rurality, although it is commonly believed that urbanicity models should include both a demographic and spatial dimension. This study uses data from Demographic and Health Surveys (DHS) conducted in three countries in East Africa to study the interaction between urbanicity-related outcomes and geospatial covariates. Specifically, the study sought to determine if urban-correlated indicators can be predicted by various covariates of urbanicity, and to identify the covariates of urbanicity that are more often significant in the prediction of urban-related indicators. A Bayesian model-based geostatistical approach was used to model the relationship between DHS urban-related outcomes and covariates. A spatial model was implemented through a stochastic partial differential equation (SPDE) in the integrated nested Laplace approximation (INLA), and was compared with a Bayesian non-spatial model. Results of the DIC values show the importance of including spatial component in the models. The results were mixed, but promising. Despite the variability in urbanicity definitions across countries, the relationship between the covariates and the selected DHS outcomes illustrates a pattern across the various countries. Land-use and demographic covariates can be used to help make predictions about the health and demographics of residents living in different enumeration areas (EAs). When women’s agricultural employment is the outcome being modeled, nightlights and travel times to hospitals should be included in the model.