The Changing Structure of Unemployment: An Econometric Study

Abstract

H OW do changes in macro-economic conditions affect the structure of unemployment? Instead of making the traditional dichotomy between structural unemployment and aggregate demand unemployment, this study shows how changes in the aggregate level of output affect the structure of unemployment by sex, age, race, occupation, and industry. Once the aggregate effects are measured, changes in the underlying structure of unemployment can be identified, and the problems that would arise if general aggregate demand policies were to lower unemployment can be foreseen. What unemployment rates would remain high? What unemployment rates would be particularly low? Through these questions the nature of the necessary labor market adjustments can be discovered. An econometric model is developed to explain changes in unemployment by sex, race, age, occupation, and industry. Using this model, the paper: (1) shows that much of the change in unemployment can be explained by the rate of growth of GNP, (2) calculates the full employment rate of growth and the rate of growth necessary to reach full employment, (3) isolates the short-run cyclical pattern from the long-run trends in each type of unemployment, and (4) projects the changing structure of unemployment rates that would exist if the total unemployment rate were reduced to 4.0%. Instead of building a model to explain the supply of labor, another model to explain the demand for labor, and then subtracting one from the other to find unemployment, a model is constructed to explain changes in the pool of unemployment. This method eliminates some of the errors that arise from estimating two models and then subtracting, but the method is primarily designed to avoid the problems that arise from the interaction between supply of and demand for labor.' In order to make consistent projections of unemployment (the weighted average of the component unemployment rates must equal the total rate), the model is subject to several constraints. Exactly the same explanatory variables have to be used in the functions for each type of unemployment. A general model is used to explain the different kinds of unemployment even though a better model might be found for each individual unemployment rate. For the same reason, linear functions are used. Log functions could be fitted but they would not necessarily add up to the projected total unemployment rate when translated back into normal unemployment rates.