Banca de DEFESA: MARCOS AURÉLIO MAIA DA SILVA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : MARCOS AURÉLIO MAIA DA SILVA
DATE: 27/10/2026
TIME: 14:00
LOCAL: Ambiente virtual
TITLE:

Spatial Bayesian Model for Age-Specific Mortality in Small Areas with Incomplete Vital
Records

KEY WORDS:

Small-area mortality; Bayesian hierarchical models; Spatial dependence; Death
under-registration.

PAGES: 70
BIG AREA: Ciências Exatas e da Terra
AREA: Probabilidade e Estatística
SUBÁREA: Estatística
SUMMARY:

Estimating mortality rates in small geographic areas involves challenges associated
with the scarcity of events, high variability in counts, and incomplete death records, a problem
aggravated by regional inequalities in the coverage of Brazil’s Mortality Information System.

This dissertation proposes a Bayesian hierarchical model that jointly integrates age smoothing,
spatial dependence, and under-registration correction, providing a unified approach to
addressing these limitations. The model extends the methodology of Schmertmann and
Gonzaga (2018), which combines the TOPALS parameterization with a Bayesian formulation for
under-registration modeling. The main innovation consists of incorporating a Spatial Gaussian
Process with an exponential correlation function based on the Euclidean distance between the
centroids of the areas, replacing the strictly local dependence of CAR and ICAR models with a
continuous structure capable of capturing different correlation ranges. To account for the
heterogeneity of mortality patterns across the life course, the spatial variance and decay
parameters are specified independently for four age groups. Estimation is performed using a
hybrid MCMC algorithm combining Gibbs sampling and adaptive Metropolis–Hastings steps,
implemented in R. The results show a substantial reduction in excessive variability, particularly
at extreme ages and in less populous areas, while preserving consistent patterns in more
stable units. RMSE and spatial correlation analyses indicate greater consistency across areas
after modeling, and the MCMC diagnostics indicate satisfactory convergence for most
parameters.

COMMITTEE MEMBERS:
Interno - 3010614 - ELIARDO GUIMARAES DA COSTA
Externo à Instituição - FERNANDO FERRAZ DO NASCIMENTO - UFPI
Presidente - 1781198 - FIDEL ERNESTO CASTRO MORALES
Interno - 1342138 - RODRIGO MATHEUS ROCHA DE MEDEIROS