UC Berkeley

Demographers have long been aware that the practical application of basic demographic identities often leads to inconsistent estimates due to data quality limitations. Despite the considerable effort to produce demographic estimates and reconcile inconsistent demographic data in different contexts, this issue remains unsolved. This dissertation proposes a Bayesian probabilistic approach that allows for a concurrent estimation of consistent counts of population, mortality, fertility and migration. Old and new methods are combined, thus building upon well-established demographic techniques and statistical methods.

This dissertation addresses these issues in three chapters. Chapter 3 describes a set of methods to estimate and reconcile past demographic data, including measures of uncertainty. These methods are highly flexible and applicable to different contexts, both at national and subnational levels, with varying availability and quality of data. In addition to reconciling past population estimates, the methods detailed in this dissertation are concerned with estimating fertility, mortality and migration based on a set of raw observed data. Chapter 4 presents novel estimates and analysis of population, fertility, mortality and migration for Brazil and its 27 states for the period 1980-2010, based on several methods and data sources. Chapter 5 applies the data obtained in Chapter 4 to the methods presented in Chapter 3 to produce demographic estimates for Brazil and its three selected states.

This dissertation offers several important methodological contributions to the fields of formal demography and statistical demography. The application of the proposed methods for Brazil and its states unveils new perspectives for generating higher quality demographic estimates in the country. Among other findings, the results reveal higher internal migration flows than those estimated with census data, and fertility estimates that differ from previous work.

This dissertation also offers substantial methodological contributions specifically toward techniques of fertility and mortality estimation. Chapter 4 presents a new sensitivity analysis for the Brass P/F ratio method to evaluate the magnitude of bias in the results of its application when one or more conditions of the method are not met. This leads to a refinement in the original method to correct for these biases. This chapter also introduces a method to adjust for bias on recent deaths in the household reported in censuses. The application of the method for Brazil shows that such adjustment greatly improves the efficiency of mortality estimation, particularly at old ages, resolving one of the main limitations of these data.