Quality & Quantity (2006) 40:935–957 © Springer 2006
A Geo-Additive Bayesian Discrete-Time Survival
Model and its Application to Spatial Analysis of
Childhood Mortality in Malawi
Clinical Sciences Research Institute, CSB, UHCW Campus, Warwick Medical School,
Clifford Bridge Road, Coventry CV2 2DX, UK;
Department of Statistics, Stockholm University, SE-106 91 Stockholm, Sweden
Abstract. We describe a ﬂexible geo-additive Bayesian survival model that controls,
simultaneously, for spatial dependence and possible nonlinear or time-varying effects of
other variables. Inference is fully Bayesian and is based on recently developed Markov Chain
Monte Carlo techniques. In illustrating the model we introduce a spatial dimension in mod-
elling under-ﬁve mortality among Malawian children using data from Malawi Demographic
and Health Survey of 2000. The results show that district-level socioeconomic characteris-
tics are important determinants of childhood mortality. More importantly, a separate spa-
tial process produces district clustering of childhood mortality indicating the importance of
spatial effects. The visual nature of the maps presented in this paper highlights relationships
that would, otherwise, be overlooked in standard methods.
Key words: Bayesian inference, discrete-time survival models, geo-additive models, Markov
Chain Monte Carlo (MCMC), Spatial modelling, time-varying effects, under-ﬁve mortality
Investigations on trends in, patterns of, and associations to childhood
mortality rates are worthwhile efforts because mortality in childhood is a
sensitive indicator of the quality of life in society (WHO, 1998). Since causes
of childhood mortality are multifaceted and may operate in many complex
ways, appropriate methodology that address these complexities are called for.
In particular, socioeconomic and demographic patterns of child mortality
vary a great deal from place to place and over time. Standard approaches
such as correlation coefﬁcients and regression analysis may produce sum-
mary statistics and measures of association at one particular site. But it can-
not be assumed that these relationships hold everywhere within a country.
Author for correspondence: Gebrenegus Ghilagaber, Department of Statistics,
Stockholm University, SE-106 91 Stockholm, Sweden, E-mail: Gebre@stat.su.se.