This dissertation studies the role of social institutions in economic development. While other research has examined the role of ethnicity, religion and other types of large-scale social organizations in development, I study the impact of two different types of local social organizations in two very different contexts. The first social institution that I study are the tribes of modern Yemen and the second is the Freemasons of the United States in the nineteenth century. I demonstrate that both have had an important impact on development, with the first affecting a political patronage network that functions through the education system and the second having a direct impact on the development of the American educational system.

The first chapter examines the relationship between political patronage and tribes, a key social institution in the developing world. Patronage is a tool used throughout the world to reward political allies. Here I create a dataset of Yemeni tribes to explore their role in an educational patronage network that accounts for upwards of 6% of the entire Yemeni government budget. My analysis has two key results. First, conditional on a rich set of controls, I find that the number of tribes has a significant impact on the quantity of patronage. This impact is negative between regions, though positive within regions, as regions with more tribes have less patronage while sub-regions with more tribes have more patronage. The contrast between these effects illustrates the differing influence of tribes in local and national politics. Second, I find no evidence that a recent decentralization reform affected this patronage network. This analysis provides insight into how pre-Islamic institutions have an important role in the development outcomes of the Muslim Middle East and why decentralization reforms proposed for countries similar to Yemen, such as Afghanistan and Somalia, may be ineffective in weakening the power of local elites.

The second chapter examines the role that American Freemasonry played in the historical expansion of the American educational system. I demonstrate that 19th century Freemasonry had a significant positive impact on educational enrollment during and after the rapid rise of the `common school' in the late 19th century. And in what is a striking example of the `path dependence' of social institutions, I show that this effect persisted through the expansion of American high schools in the 1910s-1940s even after the waning of the influence of this organization. Interestingly, Freemasonry's impact was particularly significant in areas that were the most heterogeneous - both ethnically and religiously. This, combined with the the further observation that areas with more Freemasons had higher levels of local taxation, suggests that Freemasonry helped communities overcome the common good problem. As Freemasons did not tend to migrate to areas with existing public education systems, this effect is not driven by reverse causality. And I use a panel data set of enrollment to provide evidence that unobserved heterogeneity and endogeneity are not driving the observed relationship.

The third chapter, which is a co-authored project with Bryan Graham and Cristine Campos de Xavier Pinto, develops a new empirical tool of significant utility for empirical economists studying issues such as those faced in the other chapters. It presents a new estimator, based on minimum empirical discrepancy (MD) methods, for a class of data combination problems. In these problems the researcher does not have access to a random sample containing measurements of all required variables, Z=(W',X',Y'). Instead two separate samples are available. The first is drawn from the study population of interest and contains N_s measurements of (Y, W). The second is drawn from an auxiliary population and contains N_a measurements of (X, W). The first step of our procedure involves using MD methods to re-weight the auxiliary sample in order to match study sample moments of W, the variable common to the two datasets. Sample moments from the study and re-weighted auxiliary samples are then combined to estimate the parameter of interest. We show that our estimator's asymptotic variance coincides with the relevant variance bound under two auxiliary parametric restrictions, but only requires one of these two restrictions to hold for consistency (`double robustness'). Our procedure can be used to estimate the average treatment effect on the treated (ATT), the two sample instrumental variables (TSIV) model, counterfactual earnings distributions, and to construct poverty maps. We compare our estimator with leading alternatives in an illustrative study of the effect of National Supported Work (NSW) demonstration participation on earnings and in a series of Monte Carlo experiments.