A person's income is generally explained as a function of human capital variables and variables that measure the strength or structure of the economy. Income is usually studied as a variable dependent on education, union involvement, the income of others, and experience. Regional breakdowns of income in the United States, however, show significant variations in regional wages that may not be accounted for in traditional wage equations. This paper explores the determinants of income in each of the four U.S. Census regions to determine if significant differences exist and to account for the structural and regional causes of such differences. The results could have significant ramifications for policy makers and businessmen. This paper creates a regional income model to be evaluated using both OLS and neural networks. Results from OLS and the Cotton-Neumark decomposition method are compared with those of neural network to evaluate the advantages in methodology. Clearly, some of the variations in regional wages are due to differences in the variables making up the wage equations. Other variations, or residual differences, can be summed up as the regional coefficient that explains regional peculiarities in wage determination. To improve and extend the analysis, the data are also evaluated by a forward-feeding back-propagating neural network. Neural networks offer a significant advantage through their ability to determine the relationship between variables using only input and output data. Non-linear relationships can be modeled with neural networks, accounting for variation in behavior that could not accurately be modeled linearly. The networks will be run in two modes: prediction and classification. In the prediction mode, the neural network is directly analogous to the traditional economic model using OLS. As with the OLS regression, the network is run both with and without the regional dummy variables to measure the influence of a county's location. The neural network results are compared to OLS through R values between the predicted and actual values. Correlation values for various neural networks are in the range of .90 to .95, with exceptional income prediction in low and middle income counties. In the classification mode, the network interprets the county data, including income, and assigns each county to one of the four census regions. How accurately the network assigns counties their known regions illuminates the regional differences. Success is measured through several means.
Download Info
To download:
If you experience problems downloading a file, check if you have the
proper application to
view it first. Information about this may be contained
in the File-Format links below. In case of further problems read
the IDEAS help
page. Note that these files are not on the IDEAS
site. Please be patient as the files may be large.