Analyzing Spatial Distribution of Antenatal Care Utilization in West Africa Using a Geo-additive Zero-Inflated Count Model

Abstract

Utilization of antenatal care services measured by the number of visits is a count variable and, for most developing countries, it is often characterized with excessive zeros due to nonattendance, and over-dispersion necessitating the use of special models for analysis. Zero-inflated negative binomial model was applied to the data from five neighboring West African countries and the usual parametric form of the parameters were extended to structured additive predictors in order to examine spatial patterns in a manner that transcends geographical boundary and nonlinear effects of continuous covariates. Inference was Bayesian based on Markov chain Monte Carlo approach. While results of the socioeconomic and demographic variables largely confirm findings from existing literature, there are significant residual patterns that are unexplained by the variables. In particular, there appears to be a tie transcending boundaries especially among regions of Mali, Niger and northern Nigeria where utilization remains persistently lower.

Appendix

Sample codes for estimating parameters in BayesX

The following codes demonstrate how one of the models was estimated within BayesX

• > dataset d

• > d.infile, maxobs = 100000 using C:\Users\zib_anc\data.txt

• > map m

• > m.infile using using C:\Users\zib_anc\West_Africa_map.bnd

• > m.reorder

• > logopen using C:\Users\zib_anc\ results_1\logb.txt

• > bayesreg b

• > b.outfile = C:\Users\zib_anc\results_1\b

• > b.regress anc = age(psplinerw2) + mar_dur(psplinerw2) + prim + sec_high + hus_prim + hus_sec + newspaper + radio + telv + residence + work + poorer + middle + richer + richest + parity23 + parity4 + region(spatial, map = m), family = zip zipdistopt = zinb iterations = 35000 burnin = 5000 step = 10 predict using d

• > b.plotautocor, mean

• > b.getsample