The Effect of Data Swapping Procedures on Regression Estimates – Evidence from a Simulation Study

Bartus, Tamás (2014) The Effect of Data Swapping Procedures on Regression Estimates – Evidence from a Simulation Study. Hungarian Statistical Review, 91 (SN18). pp. 3-20. ISSN 0039-0690

Preview

Text 2014_K18_003.pdf Download (215kB) | Preview

Abstract

Data swapping is a technique of statistical disclosure control. Although it has several desirable properties, it biases covariances and linear regression coefficients. The study examines by simulation the extent of biases caused by various data swapping techniques, using anonymized Labour Force Survey (LFS) data of the Hungarian Central Statistical Office (HCSO). It is found that the relative bias associated with data swapping may be held under 10%, and the bias can be minimized by selecting the pairs to be swapped randomly and by manipulating several explanatory variables simultaneously. Interpretation of the results is based on the theory of measurement errors.

Actions (login required)

Edit Item