| Convergence theory stems from neo-classical growth theory and hypothesizes that regional incomes will converge over time. Beta convergence, the idea that regions of initial poverty will grow faster than regions of initial wealth, has received a great deal of study and yields mixed results. Typically beta convergence is tested in an OLS regression with income changed regressed against initial income levels. From the geographic perspective, possible reasons for mixed convergence test results are the impacts that aggregation size and spatial effects can have on model performance. Interestingly, the potential impacts of these problems are relatively unexplored in the convergence literature. This dissertation fills that void through an examination of convergence in the United States from 1970--2004 at three levels of aggregations: states, Economic Areas, and counties. First an Exploratory Spatial Data Analysis is conducted in order to determine the magnitude and extent of spatial dependence in the convergence variables. Next, OLS, first order, and second order spatial models are run testing for unconditional and conditional convergence. Results indicate that spatial dependence is a problem in convergence models at all scales, and a spatial model must be used. First order spatial models out- perform second order spatial models. Regression results indicate convergence evidence to be strongest at the smaller levels of aggregation, though model fit tends to be better at larger levels. In the end, the functional Economic Area geography proves to be the most stable unit of aggregation for convergence analysis. |
