Probabilistic Graphical Models for Statistical Matching
Eva Endres, Thomas Augustin · 2015
In the information age a massive amount of data is available. It can be of great benefit to use this existing data for secondary analysis instead of collecting new data, which might be time-consuming and expensive. But what can be done if the required variables are not all accessible in one single data set? The solution is given by statistical matching: With the aid of statistical matching, information from dierent surveys can be combined. The initial situation of statistical matching [2, e.g.] are two (or more) data sets, e.g. A and B with nA or nB observations, respectively, that contain information on a set of common variables X, and specific variables Y and Z which are not jointly observed. The observation units in the dierent