Shift-share analyses are used by geographers, regional scientists, economists, and other social scientists to explore the nature of interconnections in economic and social systems. By examining changes in statistical associations in data among subunits (such as states, economic sectors, or groups of people) as parts of a larger entity (such as a nation, an economy, or a population) over a specific time period, shift-share analysis helps identify the degree to which changes over time are a function of broad variations across the entire entity ("shares") and the degree to which they relate to differential developments among the subunits ("shifts"). Although shift-share analysis explicitly focuses on changes in patterns over time, it has been criticized by many scientists as being overly deterministic and as incapable of elucidating the processes that produce those changes. This project will explore alternate forms of shift-share analysis based in probabilistic statistical approaches. These approaches differ from conventional accounting-system approaches. By basing his work on the fact that probabilistic shift-share analysis is a special case of more general statistical models, the investigator will extend probabilistic shift-share approaches to include interaction terms among variables. He also will evaluate algorithms designed to estimate parameters in probabilistic shift-share analyses, and he will compare the results of probabilistic shift-share methods with results obtained using more conventional approaches. This research will address some important methodological issues that have limited the potential utility of shift-share analysis. This project will explore the statistical foundations of the procedure and will develop it within a probabilistic framework, which should provide a sounder basis for quantitative testing of hypotheses about changes in economic activity, employment, and population.
