Latent Class Multiple Imputation for multiply observed variables in a combined dataset

L. Boeschoten, D.L. Oberski, A.G. de Waal

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientific

Abstract

Both registers and sample surveys can contain measurement error. While
some errors are invisibly present, others become visible when logical
relations in the data are investigated. When a variable is measured in multiple
datasets within a combined dataset, we can get an indication of the errors
which are invisibly present within the separate datasets. We propose a new
method (MILC) based on latent class modelling that estimates the number of
measurement errors in the multiple sources, and simultaneously takes
impossible combinations with other variables into account. We then use the
latent class model to multiply impute the latent “true” variable. Whether
MILC can be applied depends on the entropy R2 of the LC model and the
type of analysis you are interested in.
Original languageEnglish
Title of host publicationQ 2016
Subtitle of host publicationEuropean Conference on Quality in Official Statistics
Place of PublicationMadrid
Pages1-10
Number of pages10
Publication statusPublished - 2 Jun 2016

Keywords

  • latent class models
  • multiple imputation
  • combined dataset

Fingerprint Dive into the research topics of 'Latent Class Multiple Imputation for multiply observed variables in a combined dataset'. Together they form a unique fingerprint.

Cite this