Bayesian Semiparametric Multivariate Density Deconvolution.

Area of Science:

• Statistics
• Biostatistics
• Computational Statistics

Background:

• Precise measurements of vector-valued random variables (X) are often unavailable.
• Observations are frequently contaminated by measurement errors (U).
• Existing literature typically assumes known measurement error densities, limiting applicability.

Purpose of the Study:

• To develop robust Bayesian semiparametric multivariate deconvolution methods.
• To address scenarios where measurement error density (U) is unknown.
• To accommodate complex error structures, including those dependent on unobserved variables (X).

Main Methods:

• Utilizes finite mixture models, multivariate normal kernels, and exchangeable priors.
• Employs Bayesian semiparametric approaches for robust estimation.
• Leverages replicated proxies for individuals with unknown measurement error densities.

Main Results:

• Demonstrates theoretical flexibility in capturing diverse data-generating processes.
• Simulation experiments show efficient recovery of the target density (X).
• Successfully applied to estimate joint dietary consumption patterns from contaminated recalls.

Conclusions:

• The proposed methods offer a flexible and robust solution for multivariate density deconvolution with unknown error structures.
• The approach enhances accuracy in estimating distributions from contaminated data.
• Provides a valuable tool for applications in various fields, including nutritional epidemiology.