The validity of data fusion

Abstract

"In this paper we derived bounds for the correlations between variables not jointly observed, provided that one of the vectors of specific variables is univariate, and suggest a new quality index of data fusion which is built upon these bounds. Using our results, multiply imputed datasets can be produced according to different admissible correlation structures between X and Y by using appropriate algorithms (notice that since data fusion can be viewed as a problem of missing data, multiple imputation procedures are applicable in general). Analyzing the different fused data sets can then reveal sensitivity to the different assumptions about the correlation structure between the variables that have never been jointly observed." (Text excerpt, IAB-Doku) ((en))

Cite article

Kiesl, H. & Rässler, S. (2008): The validity of data fusion. Paper presented at the CENEX-ISAD workshop on integration of surveys and administrative data, Vienna 29-30 may 2008. Wien, 9 p.