Schooling, cognitive skills, and household income: An econometric analysis using data from Ghan

Most of the human capital literature pertaining to developing countries focuses on the returns to education in either farm work or wage work. Using data from Ghana, this dissertation broadens the focus by acknowledging that many households are engaged in multiple income-generating activities, and argues that the estimated returns to education need to incorporate the effect that human capital has on all sources of income.;Chapter One examines farm households in Ghana and finds that while the returns to schooling as measured by a farm profit function are small, the farm households accrue significantly larger returns from schooling in their off-farm work. This chapter also analyzes whether the difference in returns to farm and off-farm work affects the allocation of labor across these two activities. Chapter Two expands the sample to include all Ghanaian households, and uses cognitive skills instead of school attainment as an alternate measure of human capital. This chapter finds that cognitive skills have a positive effect on total income and off-farm income, but have no effect on farm profit. The results from both chapters support the hypothesis that focusing on income generated only from farming or wage work will mis-measure the returns to human capital.;An issue faced in Chapter One and Chapter Two is how to model education attainment, which is an individual level variable, in household-level income functions, Chapter Three addresses this by considering competing models of household schooling and the determination of income. This chapter finds evidence to reject the standard assumption that the school attainment of the head of household is the relevant measure of schooling for the household.;The estimators used in all chapters are robust to many violations of the classical linear regression assumptions. In particular, this dissertation discusses the effect that a clustered sample has on parameter estimates and standard errors. Chapters One and Two report standard errors which are Huber-corrected for sample design effects, while Chapter Three reports standard errors which result from a bootstrap method that replicates the clustered sample design.