Partial identification in statistical matching with misclassification

详细信息    查看全文

• 作者：Marco Di Zio^a ; ^{dizio@istat.it} ; Barbara Vantaggi^b ; ^{barbara.vantaggi@sbai.uniroma1.it}
• 关键词：Data fusion ; Synthetical matching ; Consistency ; Partial identifiability ; Envelopes of probabilities ; Inference
• 刊名：International Journal of Approximate Reasoning
• 出版年：2017
• 出版时间：March 2017
• 年：2017
• 卷：82
• 期：Complete
• 页码：227-241
• 全文大小：456 K
• 卷排序：82

文摘

The main target of statistical matching is to make inference on variables observed in different sources by using information on common variables. The partial information generally implies that the model is not identifiable. The aim of this paper is to study the case when common variables are affected by a misclassification. The partially identifiable region, that is the class of probabilities extending the conditional probabilities obtained by the information in different sources, is determined. These regions are determined in the general case and under some specific restrictions on the misclassification mechanism. An application to real data is used to show in practice the results achieved in the paper.