Laura Close, Associate Statistician of Research Division NYS DOL

State Office Campus, Building 12, Rm. 470 C

Albany , NY 12240

Phone: 518- 457-6574

Fax : 518 – 457- 6382

e-mail: usdlmc@labor.state.ny.us

Roger Gerby, Chief Research and Evaluation Unit NYSDOL

Igor Zurbenko, PhD., Professor School of Public Health, University at Albany, New York 12144

Optimization of effectiveness of profiling system

Roger Gerby, Laura Close and Igor Zurbenko

Profiling of groups of population is frequently used in marketing analysis, insurance, public health studies, unemployment analysis and many other areas. The statistical tool usually used in profiling is logistic regression or contingency tables. Both of those approaches are very vulnerable to the right choice of predictor variables over the population. We develop a method of selection of optimal group of variables, which provides the best possible prediction among the considered outcomes. The proposed method has been successfully used for unemployment insurance profiling at the New York State Department of Labor. The analysis performed on Unemployment Insurance beneficiaries shows the effectiveness of prediction can be nearly double compared to traditional approaches. Relevant to the total NY unemployment data available, several groups of best predictor variables have been chosen. The optimal number of predictor variables in those groups was generally around five variables.

References

R. Gerby, L. Close, I. Zurbenko, A review of Worker Profiling, Report of Division of Research and Statistics, NYS Department of Labor.

K.Schlauch, M.Puglisi, Worker profiling and reemployment services. Profiling methods: lessons learned. Federal report on worker profiling. http://www.itsc.state.md.us/ui manage/wpr.html

P.Adriaans, D.Zantinge, Data Mining, Addison-Wesley 1996

D.Hosmer, S.Lemeshow, Applied Logistic Regression, John Wiley, 2000