The three chapters in this dissertation analyze data from studies done with a focus on understanding marginalized people groups. The first analysis (in Chapter 1) looks at the lives of children in poor households in Ecuador, in particular by estimating the effects of the household-level determinants on educational achievement and on the incidence of child labor. The second two analyses (Chapters 2 and 3) attempt to understand the determinants of homelessness in Costa Mesa.

Chapter 1 finds that the Bono Desarrollo Humano in Ecuador has a positive effect on reducing child labor in the current generation, and perhaps secondary effects into subsequent generations. I also find that child labor and household educational attainment have a negative correlation, even after controlling for other observables.

Chapter 2 utilizes a novel dataset and to study homelessness and finds that the self-reported cause of homelessness is a determining factor. It also finds, as would be expected, that education and the presence of family nearby lessens the intensity of homelessness.

Chapter 3 investigates how a homeless individual's intensity of homelessness is predicted by the individual's discount rate and attitudes toward risk. The ex-ante hypothesis is that an increase in homeless intensity could be driven by an decrease in patience and/or an increase in risk tolerance. The hypothesis is tested by using a novel application of weighted least squares to regress the individual's homeless intensity (HIM) on the risk and discount variables mentioned above, as well as on the clearly exogenous individual-level characteristics. Evidence is found to support the hypothesis, in that a decrease in patience is associated with an increase in the individual's HIM. Increasing levels of risk tolerance are associated with an increase in the HIM as well, and this effect appears to be significant even with the inclusion of observably risky behaviors.

The data for Chapter 1 was obtained from the Ministry of Social Development of the Government of Ecuador. The data for Chapters 2 and 3 come from two surveys the author conducted himself. The first took place in October of 2012 and the second occurred in April 2014.