In developing countries, where there are limited resources and social support, children rely primarily on their family for their well-being. Their lives are affected by the configuration of household members, by the type of family structure, and by household decision processes that determine the resources that are allocated for their well-being. This dissertation examines the influence of family structure on the outcomes of children's well-being, using the case of Cambodia, where between 2000 and 2010 the children composed more than a third of the population.

To assess the link between family structure and the well-being of children in Cambodia, this dissertation uses multilevel modeling. It examines three aspects of well-being. The first is the amount of time invested on children. The second examines the nutritional status of children. The third is the risk of infant mortality. In this investigation I focus on two forms of family structure: nuclear and multigenerational. For our purposes, a multi-generational household is one containing one or more grandparents.

To consider the investment of time, I take advantage of time-use data collected by the Cambodia Socio-Economic Survey and assess time spent on childcare individually and collectively by members of the household. My analysis reveals that children in multigenerational households receive more childcare time than those in nuclear households. Moreover, the additional care time these children receive is due to the presence of grandparents. Grandparents supplement maternal care time.

To assess the nutritional status of children in Cambodia, I use three waves of the Cambodia Demographic and Health Survey from 2000, 2005 and 2010. My findings reveal that children in multigenerational households have better nutritional outcomes than children in nuclear households. The gains of children in multigenerational households increased over the decade. This finding suggests that even though national economic development improves the overall health status of children, those coresiding with grandparents are the most advantaged.

To detect possible influences of family structure on infant and child mortality, I explore whether or not the health advantage of children in multigenerational households extends to a survival advantage. Findings reveal that even in the earliest stages of life, the presence of grandparents safeguard infant survival more than the presence of a father in the household.

The human mortality experience has changed fundamentally as a result of the mortality transition. Not only are humans today living longer than their ancestors on average, they also experience greater certainty about the eventual timing of their death. This greater certainty is due to the considerable compression of the distribution of ages at death, which characterizes the mortality transition and results in lower life span disparity at the population level. In this dissertation, I investigate two key issues, which lie at the intersection of the mortality compression and mortality disparities literatures. First, I explore the recent transition from an era of mortality compression to a new era of mortality change, the shifting mortality era, in which life expectancy continues to increase but variability of age at death remains constant. Secondly, I examine differentials in life span disparity between the sexes and across countries over the course of the epidemiological transition. I adopt an age and cause-specific approach in analyzing trends in variability of age at death and differences in life span disparity across groups.

To account for age and cause-specific effects, I take advantage of recent methodological advances in demographic analysis and employ computationally based decomposition, perturbation, and simulation methods. Using an extensive collection of period life tables from the Human Mortality Database (HMD), I am able to explore aggregate level all-cause mortality change over the past one hundred twenty years in ten Western European countries. For some of my analyses, I take advantage of the full collection of historical mortality data available in the HMD, which includes populations in thirty-seven distinct geographic areas on five continents. Using a long series of cause-of-death data from France, which extends back to 1925, I am able to examine how sex-specific trends in variability of age at death were influenced by changes in age and cause-specific mortality during the epidemiological transition.

In documenting sex-specific trends in variability of age at death, I find that, similar to the emergence of the sex gap in life expectancy, a sex gap in variability of age at death developed around the midpoint of the 20th century with females experiencing lower life span disparity in comparison to males. Using decomposition techniques, I demonstrate that the gap emerged because females experienced greater reductions in premature mortality in comparison to males. Cause-of-death decomposition results for France reveal that this is due to (1) declines in infectious disease, which young females had suffered disproportionately, and (2) male disadvantage in external cause mortality across the stages of the epidemiological transition. Examining cross-country differences in life span disparity between Sweden and the United States, I find that a gap in variability of age at death emerged between these two countries with Sweden exhibiting lower variability due to (1) Sweden gaining an advantage in premature mortality and (2) the United States gaining an advantage in old age mortality.

Using measures of variability of age at death, I also examine trends in mortality compression and possible explanations for the recent transition to an era of shifting mortality. To this end, I investigate how initial mortality conditions interact with the age pattern of mortality change to produce mortality compression, expansion, or shifting. I find that proportional mortality change that is fixed at rates unvarying across age does not necessarily lead to a parallel shift in the death distribution. I discover that certain initial mortality conditions are particularly primed for compression and that there is a significant change in the age-pattern of the sensitivity of measures of variability of age at death to proportional changes in age-specific mortality rates over the course of the mortality transition. I demonstrate that the potential for compression still exists in more developed countries despite the current shifting trends and that trends in cancer related mortality may play a particularly important role in determining future trends in variability of age at death.

In this dissertation, I explore the nature and role of risk aversion among older Americans from a variety of perspectives. Risk preferences are important to demographers for several reasons. First, risk preferences are fundamental to most individual-level demographic events, including to financial decision-making, health behaviors, labor market decisions, migration, and marriage and family-formation. Second, there is substantial evidence that risk aversion increases with age. With age also comes increased responsibility in terms of making specific financial and health decisions. In the age of decreasing pensions, older persons must make significant decisions about their financial portfolios and finances in light of pending retirement decisions. In fact, the decision to retire is itself one in which risk plays a role. Third, health behaviors, which are a function of one's riskiness, often display their effects at older ages.

I explore the genetic nature of risk aversion through a number of approaches. Taking advantage of a newly-released database with over two million pieces of genetic variants, I examine the specific genetic nature of risk aversion through two genomic techniques: a Genome-Wide Association study (GWAS) and a Genome-Wide Complex Trait Analysis (GCTA). I provide evidence that risk aversion is a highly complex trait that is a function of a large number of possibly interactive genetic variants. Through the GWAS, I show that the number of genetic variants influencing individual-level differences in risk aversion is numerous and that these variants are likely to be scattered across the genome. The GCTA, while using a separate methodological approach, confirms this finding. I argue that the intricate nature of the genetic underpinnings to risk aversion should be better understood in order to more precisely model economic decisions involving risk preferences.

I also characterize risk aversion from a non-genetic perspective. Using panel data of risk aversion collected over nearly two decades, I use longitudinal methods to explore the extent to which the relationship between hypothetical risk and measurable risky behaviors remain consistent across both time and among individuals. As a follow-up, I examine the specific time period following the 2008 recession to examine any change in the relationship in portfolio allocations relative to stated risk tolerance for individuals after the global financial crisis. I conclude that the relationship between measured risk and risky behaviors remains relatively constant across the 15 years prior to the global financial crisis. The analysis also shows that the relationship between risk and financial assets does in fact change slightly after the global financial crisis, though the statistical evidence is not very strong.

This dissertation provides a contribution to the understanding of the complex nature of risk aversion and is on of the first to characterise it's genetic influences. This research helps to answer questions on the economic, social and biological drivers and consequences of risk aversion among older Americans.

This dissertation is dedicated to an exploration of "Wood's Method" -- a novel approach to

fitting demographic transition matrices to age and sex population count data.

Demographic transition matrices, otherwise known as "Leslie matrices," are extensively

used to forecast population by age, sex, and other characteristics. Our implementation of

Wood's Method simplifies the creation of age and sex population forecasts greatly by

reducing the amount of data necessary to create a demographic transition matrix.

Furthermore, the method can be used to infer a demographic component of change (one of

migration, fertility, or mortality) if the other two components are specified.

In Chapter One, we introduce Wood's Method, as well as showing some illustrative

examples. In Chapter Two, we evaluate the accuracy of Wood's Method by crossvalidating

age and sex specific forecasts for 3,120 US counties. In Chapter Three, we

present a simpler, alternative derivation of Wood's Method with an extensive example and

show some extensions to the method made possible by this new formulation. In Chapter

Four, we use the method to examine migration rates at the US County level and show

important results regarding clustering of migration. Each chapters is independent of the

others, but should be read in order.

To our knowledge, this is the first time Wood's Method has been used for forecasting

human populations. We hope to show its viability as a forecasting and analysis method

and sketch directions for further research.

This dissertation addresses the problem of estimating and projecting orphanhood prevalence and kinship resources available to orphans in Zimbabwe, one of the countries hit the hardest by the HIV/AIDS epidemic. The extended family has been recognized as a major safety net against the negative consequences of the generalized HIV/AIDS epidemic in sub-Saharan Africa. However, little is known about the effect of the epidemic on the quantity of kinship resources available to children. This study contributes to the existing literature by providing a quantitative assessment of the material basis of traditional kin relations in Zimbabwe.

An approach based on formal demographic methods is used to evaluate the effect of the HIV/AIDS epidemic on maternal orphanhood, for the period 1980-2050. The model, which is informed by United Nations estimates and projections of demographic rates, provides insights on the process of orphans' generation. One of the main results is that the number of maternal orphans in the age group 0-17 years is expected not to decline until around 2030. This is related to the fact that the transition to orphanhood is a cumulative process with age, and that there is a lag between the peak in adult HIV prevalence and the one in AIDS-related orphanhood prevalence.

A microsimulation model, whose core relies on SOCSIM, is used to estimate quantities for which analytical expressions are unmanageable, such as kinship resources for double orphans. The model is calibrated to the Zimbabwean setting, using data from the Demographic and Health Surveys (DHS) and estimates and projections from the United Nations. The results of the simulation show a transition, between 1990 and 2010, from fairly high to fairly low levels of kinship resources for young children in Zimbabwe. The proportion of double orphans without any living grandparents will increase until about 2030 and then will decrease. This trend will shift the responsibility for double orphans to uncles and aunts. On average, the number of uncles and aunts per double orphan has been decreasing from 1980 to 2010, but it is expected to increase progressively during the next decades.

Methodologically, the dissertation deals with the problem of calibration and statistical inference for stochastic demographic microsimulations. Traditional forms of parameter tuning are formalized within a Bayesian framework. The approach does not provide a definite answer to the problem of statistical inference for the outputs of demographic microsimulations. However, it provides a first contribution towards the development of a more comprehensive methodology to assess uncertainty in the field of stochastic kinship forecasting.

The dissertation provides a quantitative basis for the study of the stress imposed by the HIV/AIDS epidemic on traditional forms of kin relations. This research raises questions on the social consequences of changes in kinship structure, and on the strategies needed to address the lack of care in the context of a generalized HIV/AIDS epidemic.

Recent genetic studies suggest that many age-related diseases may be attributed not to a single or small number of mutations, but rather to a large number of mutations, each of which is individually slightly deleterious. Following in the tradition of Kimura and Maruyama, we consider mutation accumulation in an infinite population with a large number of mutation types. We compare the distribution of genotypes under two extreme assumptions regarding genetic recombination: no recombination versus ``free" recombination, in which recombination acts more rapidly than mutation and selection. Under a range of assumptions, including realistic mutation rates and demographic fitness measures, we find unexpected similarities in the predictions from the different models. While recombination predictably affects the level of mutant alleles present in the population, the overall shape of the genotype distribution under the two models is quite similar, as are the general behavior of demographic outcomes such as lifespan and hazard rates. Furthermore, the distribution of genotypes under the assumption of no recombination may be well approximated by a Poisson random measure. The qualitative similarities in genotype distributions and demographic characters under these extreme models of genetic recombination suggest that attempts to model recombination in a more realistic manner may not add much to our understanding when viewed from a demographic perspective.

The two models analyzed here, developed by Steinsaltz, Evans, and Wachter, are general enough to connect age-specific effects on demographic characters, such as mortality, to mechanisms of genetic change. While the 2005 model without recombination has a series solution, it cannot be directly evaluated except in the simplest of cases. Sampling from the distribution of genotypes in cases with a large number of mutation types is challenging. In this work we utilize a multiple-try Metropolis algorithm to sample from the distribution of genotypes for spaces containing up to 1000 different mutation types. We consider a variety of test cases, finding scenarios in which typical genotypes contain 100, 350 or even 850 mutations. Our success at accurately estimating genotype distributions and demographic outcomes under assumptions that produce such a large average number of mutations suggests that this model could be utilized under more realistic scenarios, such as mutations associated with age-related disease.

Although the study of mortality is central to demography, comparatively little is known about its lived experience: how population-level mortality translates to individual experiences of loss within families and communities. Given the significance of bereavement for the health and socioeconomic outcomes of survivors, better understanding the impacts of mortality crises on family networks is crucial for predicting their longer-term consequences. The data required to study kin loss, however, is seldom available, leading many demographers to rely on computational approaches such as microsimulation to estimate the impact of these events.

This dissertation presents microsimulation-based approaches to examining the impact of a mortality crisis, in this case COVID-19, on kin networks in the short and long term. The first study presents estimates of monthly excess kin loss by age and type of kin relation in 31 countries during the period of March 2020 to June 2021. These estimates demonstrate a generational pattern of kin loss reflecting COVID-19 age-specific excess mortality risk, and highlight the significant effect of excess mortality on family bereavement. The second study extends this estimation approach to 120 countries over the 2020-2021 period, documenting high rates of excess kin loss in many low-and-middle-income countries higher than or comparable to those observed in high-income countries. It also considers the extent to which differences in country estimates were shaped by both excess mortality and pre-existing kinship structure. The third study projects how "kinship memory'', the estimated share of national populations related to victims of COVID-19 excess mortality, may change over the next century in 120 countries around the world, and considers what this may mean for how this crisis is remembered in the future. These three studies combined highlight the significant burden of COVID-19 excess mortality in terms of bereavement experienced by surviving family members, and demonstrate the importance of computational approaches in helping better understand the experiences of populations for which limited data exists.