Variational Bayesian Quantile Regression with Non-Ignorable Missing Response Data
For non-ignorable missing response variables, the mechanism of whether the response variable is missing can be modeled through logistic regression. In Bayesian computation, the lack of a conjugate prior for the logistic function poses a significant challenge. Introducing a new Pólya-Gamma variable a...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-05-01
|
| Series: | Axioms |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-1680/14/6/408 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | For non-ignorable missing response variables, the mechanism of whether the response variable is missing can be modeled through logistic regression. In Bayesian computation, the lack of a conjugate prior for the logistic function poses a significant challenge. Introducing a new Pólya-Gamma variable and employing lower-bound approximation are two common methods for parameter inference in conjugate Bayesian logistic regression. It can be observed that these two methods yield essentially the same variational posterior in the calculation of the variational Bayesian posterior. This paper applies a popular Bayesian spike-and-slab LASSO prior for variable selection in quantile regression with non-ignorable missing response variables, which demonstrates good performance in both simulations and practical applications. |
|---|---|
| ISSN: | 2075-1680 |