Neural-Network Modeling for Labor-Force Migration: a Competitive-Learning Approach

This study is an examination of the appropriateness of using a competitive learning neural-net algorithm to model labor-force migration. In the past, there have been a number of studies to model labor-force migration using cross-sectional regression analysis. These studies model the propensity to migrate based upon income differential, and they model this behavior by estimating the propensity to migrate between two cities or states given a number of different factors. All of the studies include income as a significant factor to be examined in the human-capital decision to migrate. This study is a new "computational regional science" approach. A neural-network algorithm based on competitive learning, developed and successfully applied to an image restoration problem, is used as a pilot for modeling regional science problems in general using neural networks. It first converts this problem to an optimization problem, which is solved using the algorithm of Phoha and Oldham (1996). The labor force migration problem lends itself easily to an optimization problem, which can then be solved by the competitive learning algorithm. This study assumes that interregional income differentials determine migration probability and should also be equilibrated by migration. The human-capital model of migration is based upon this income maximization thesis. This algorithm adjusts the income levels locally, but also simultaneously minimizes a global optimization function of the differences in income levels between cities. The resulting method, in terms of economic analysis, takes a general-equilibrium approach to migration. Rather than examining the variation in migration given different location factors like income, the model simulates the optimizing behavior of migration by equilibrating income based on an actual set of initial income conditions in a realistic spatial matrix of cities. By simulating this behavior, using both local equilibria information -- which is distance sensitive -- and a global equilibria -- which is not --, the actual equilibrating behavior of income can be compared to the model output to assess the efficiency of actual labor-force mobility in equilibrating economic conditions over space.