Migration Decision-making: A Hierarchical Regression Approach

While migration decision-making has long been studied using mover-stayer models and standard regression models, they do not well handle smalland large-scale heterogeneities (migration propensities). The hierarchical regression model can help solve this problem, because it deals with data organized hierarchically and studies variation at different levels of the hierarchy simultaneously. Using Wisconsin’s 5% Public Use Microdata Sample (PUMS) file from Census 2000 for a two-level hierarchy – individual/household level and Public Use Microdata Area (PUMA) level, we take a fresh look at how a hierarchical logit model can improve migration studies by including demographic, socio-economic, and biogeophysical factors. The findings indicate that the hierarchical regression approach provides significant advantages in studying migration decisionmaking. 1 Paul R. Voss and Guangqing Chi, Applied Population Laboratory, Center for Demography and Ecology, Department of Rural Sociology, University of Wisconsin-Madison, Madison, WI53706. Address correspondence to Paul R. Voss, E-mail: [email protected], Tel: (608)262-9526, Fax: (608)262-6022. This research was supported by the Wisconsin Agricultural Experiment Station (Hatch project no. WIS04536). An early version of this paper was presented at the 2004 meeting of the Southern Demographic Association, October 14-16, 2004, Hilton Head Island, SC. We thank Dr. Troy Blanchard for his suggestions on this study. 1. Background and Objective People move for various reasons: obtaining a different job, going to college, following their employer in a business relocation, expiration of a lease, marriage, divorce and so on. Migration is a large concern for policy makers because flows of population can significantly affect local political, social, economic, and ecological structures for both sending and receiving areas (DaVanzo 1981). Regional economists, demographers, sociologists and geographers have made numerous contributions to migration studies since 1960 (Greenwood 1969, 1975). Many early economic studies used aggregated data to treat migration as an equilibrating mechanism that minimizes geographic wage and employment differentials, while later studies have shifted to a microeconomic approach to study why individuals and families move (DaVanzo 1981). Three types of approaches have been widely applied to study migration. The first is the mover-stayer model in which analysts are interested in attributes that differentiate those who move from those who do not move, and the place-to-place flows for those who do move (e.g., Goodman 1961; Spilerman 1972; White 1970). The second is a multivariate regression approach in which scholars are interested in the strength of a set of migration covariates, often among origins and destinations, in order to model migration flows (e.g., Bartel 1979; Greenwood 1969; Mincer 1978; Tunali 2000). The third is a combination of the moverstayer model and Maximum Likelihood Estimation (MLE) (e.g., Frydman 1984; Kennan & Walker 2003; Sandell & Liberg 1992). Each approach has been studied and applied by demographers and regional economists. For example, mathematical demographers have attempted to improve the mover-stayer model to obtain better estimates of parameters, and regional economists have used the model to study employment migration across industries. However, all approaches have some drawbacks, which encourage us to try a fourth approach – hierarchical regression. The hierarchical regression model has already been successfully applied in other sub-disciplines of sociology such as family planning (e.g., Entwisle et al. 1984; Hirschman & Guest 1990) and education studies (e.g., Anguiano JRAP (2005)35:2