MixedDataImpute: Missing Data Imputation for Continuous and Categorical Data using Nonparametric Bayesian Joint Models

Vilhuber — Mon, 10 Apr 2017 15:55:20 +0000

Author(s):

Many datasets include a mix of continuous and categorical variables with missing values. In a paper published in the Journal of the American Statistical Association, we developed a joint model for such mixed data that can be used for multiple imputation. The approach uses a nonparametric Bayesian mixture model as the imputation engine. The mixture model comprises one set of mixture components with multivariate normal kernels for the continuous variables, and a separate set of mixture components with products of independent multinomial kernels for the categorical variables. The model induces dependence between the continuous and categorical variables in two ways, namely (i) by allowing the means of the multivariate normal distributions to depend on the categorical variables, and (ii) by using a tensor factorization prior that links the two sets of membership components.

External Link:

Paper

CRAN

Nodes:

Duke University / National Institute of Statistical Sciences (NISS)

Tags:

missing data

imputation

EditImputeCont: Simultaneous Edit-Imputation for Continuous Microdata

Vilhuber — Fri, 12 Jun 2015 14:09:24 +0000

Author(s):

Many statistical organizations collect data that are expected to satisfy linear constraints; as examples, component variables should sum to total variables, and ratios of pairs of variables should be bounded by expert-specified constants. When reported data violate constraints, organizations identify and replace values potentially in error in a process known as edit-imputation. In a paper published in the Journal of the American Statistical Association, we developed an approach that fully integrates editing and imputation for continuous microdata under linear constraints. The approach relies on a Bayesian hierarchical model that includes (i) a flexible joint probability model for the underlying true values of the data with support only on the set of values that satisfy all editing constraints, (ii) a model for latent indicators of the variables that are in error, and (iii) a model for the reported responses for variables in error. This R package implements a version of the model that uses mixtures of multivariate normal distributions for the underlying true values and uniform distributions for measurement errors. The package is available on CRAN.

External Link:

Nodes:

Duke University / National Institute of Statistical Sciences (NISS)

Tags:

Bayesian hierarchical model

edit procedures

imputation

constraints

NSF-Census Research Network - imputation

MixedDataImpute: Missing Data Imputation for Continuous and Categorical Data using Nonparametric Bayesian Joint Models

EditImputeCont: Simultaneous Edit-Imputation for Continuous Microdata