Featured White Papers
Content provided in partnership with
Business Services Industry
Reemployment differences among dislocated and other workers
American Journal of Economics and Sociology, The, Jan, 1997 by Mary Ellen Benedict, Peter VanderHart
4. It is possible that quitters may have a job offer in hand or may be planning for retirement when they leave their job, so that their classification is endogenous. To the degree that this is true, the results involving quits should be viewed with caution.
5. Establishment data does not include information on industries in agriculture, forestry, fishing, and private households. Additionally, there were some changes in SIC code assignments in 1987 from earlier years, but this did not affect the 2-digit levels for the industries, except for a small number in SIC 79 (bowling alleys, billiard and pool parlors) and 84 (museums, art galleries and zoos), which were therefore omitted.
6. These results are available from the authors upon request.
7. Fallick (1993) controls for this problem by adjusting growth rates for wages and hours normal to all industry. Industry data used here omits information on wages and hours for a good number of the service industries which makes his procedure impossible here.
8. The use of the multinomial logit approach does leave some issues unaddressed. Specifically, our formulation does not completely separate the effects of sector-specific skills from other related factors. Possible remedies to this problem include a reformulation in which one alternative is reemployment in the same or similar industries, rather than the very broad growth categories in this paper. Unfortunately this formulation would require a great deal of work classifying industries, and would require a much richer source of data than the PSID.
Additionally, this model does not take into account the timing of the transitions, which obscures issues related to the length of unemployment spells. Analysis of the effect of the factors considered here on the length of unemployment spells may add much to our understanding of reemployment of the dislocated. Additionally, our study is unable to control for multiple observations from a particular respondent. Both of these problems can be rectified with a competing risk hazard rate analysis, which is a topic or our ongoing research.
9. The multinomial logit model has the often implausible property of independence of irrelevant alternatives which implies that the odds ratio between two choices does not change when the choice set is expanded. This often causes problems with multinomial logit analysis where the explanatory variables are characteristics of the choice alternatives (rather than characteristics of the decision maker), and where there are a nearly limitless number of alternatives that could be included in the regression, but are not. However, in the regressions contained in this paper, the five labor market outcomes fully exhaust the alternatives that are possible, and thus the independence of irrelevant alternatives criticism is not of great concern.
10. The analysis employs LIMDEP to conduct all estimations (Green, 1986).
11. The base case probabilities represent single white males, with an educational level that is less than a high school degree, and who reside in the Northeast. White males were chosen to better reflect the PSID data.
