Give Missings a Chance: Combined Stochastic and Rule-Based Approach to Improve Regression Models with Mismeasured Monotonic Covariates Without Side Information

52 Pages Posted: 19 Feb 2011

See all articles by Stephan Dlugosz

Stephan Dlugosz

ZEW – Leibniz Centre for European Economic Research

Date Written: February 1, 2011

Abstract

Register data are known for their large sample size and good data quality. The measurement accuracy of variables highly depends on their high importance for administrative processes. The education variable in the IAB employment sub-sample is an example for information that is gathered without a clear purpose. It therefore severely suffers from missing values and misclassifications. In this paper, a classical approach to deal with incomplete data is used in combination with rule-based plausibility checks for misclassification to improve the quality of the variable. The developed correction procedure is applied to simple Mincer-type wage regressions. The procedure reveals that the quality of years in education is very important: The German labour market rewards general education less than vocational training. Furthermore, using this method, no indication of an inflation in formal education degrees could be found.

Keywords: Measurement Error, EM by the Method of Weights, Wage Regression, Expansion of Educational Degrees, Misclassification, Imputation Rules

JEL Classification: C13, J24, J31

Suggested Citation

Dlugosz, Stephan, Give Missings a Chance: Combined Stochastic and Rule-Based Approach to Improve Regression Models with Mismeasured Monotonic Covariates Without Side Information (February 1, 2011). ZEW - Centre for European Economic Research Discussion Paper No. 11-013, Available at SSRN: https://ssrn.com/abstract=1763271 or http://dx.doi.org/10.2139/ssrn.1763271

Stephan Dlugosz (Contact Author)

ZEW – Leibniz Centre for European Economic Research ( email )

P.O. Box 10 34 43
L 7,1
D-68034 Mannheim, 68034
Germany

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
29
Abstract Views
456
PlumX Metrics