Missing the Point: Non-Convergence in Iterative Imputation Algorithms

H.I. Oberman; S. van Buuren; G. Vink

Missing the Point: Non-Convergence in Iterative Imputation Algorithms

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

Abstract

Iterative imputation is a popular tool to accommodate missing data. While it is widely accepted that valid inferences can be obtained with this technique, these inferences all rely on algorithmic convergence. There is no consensus on how to evaluate the convergence properties of the method. Our study provides insight into identifying non-convergence in iterative imputation algorithms. We found that---in the cases considered---inferential validity was achieved after five to ten iterations, much earlier than indicated by diagnostic methods. We conclude that it never hurts to iterate longer, but such calculations hardly bring added value.

Original language	English
Title of host publication	First Workshop on the Art of Learning with Missing Values (Artemiss) hosted by the 37 th International Conference on Machine Learning (ICML)
Number of pages	4
Publication status	Published - 10 Jun 2020

Keywords

non-convergence
MICE
Iterative imputation

Access to Document

missing_the_point_non_convergence_in_iterative_imputation_algorithmsFinal published version, 110 KB

Cite this

@inproceedings{eae69001146948c0baacb36ccc2654b1,

title = "Missing the Point: Non-Convergence in Iterative Imputation Algorithms",

abstract = "Iterative imputation is a popular tool to accommodate missing data. While it is widely accepted that valid inferences can be obtained with this technique, these inferences all rely on algorithmic convergence. There is no consensus on how to evaluate the convergence properties of the method. Our study provides insight into identifying non-convergence in iterative imputation algorithms. We found that---in the cases considered---inferential validity was achieved after five to ten iterations, much earlier than indicated by diagnostic methods. We conclude that it never hurts to iterate longer, but such calculations hardly bring added value.",

keywords = "non-convergence, MICE, Iterative imputation",

author = "H.I. Oberman and {van Buuren}, S. and G. Vink",

year = "2020",

month = jun,

day = "10",

language = "English",

booktitle = "First Workshop on the Art of Learning with Missing Values (Artemiss) hosted by the 37 th International Conference on Machine Learning (ICML)",

}

Missing the Point: Non-Convergence in Iterative Imputation Algorithms. / Oberman, H.I.; van Buuren, S.; Vink, G.
First Workshop on the Art of Learning with Missing Values (Artemiss) hosted by the 37 th International Conference on Machine Learning (ICML). 2020.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Missing the Point: Non-Convergence in Iterative Imputation Algorithms

AU - Oberman, H.I.

AU - van Buuren, S.

AU - Vink, G.

PY - 2020/6/10

Y1 - 2020/6/10

N2 - Iterative imputation is a popular tool to accommodate missing data. While it is widely accepted that valid inferences can be obtained with this technique, these inferences all rely on algorithmic convergence. There is no consensus on how to evaluate the convergence properties of the method. Our study provides insight into identifying non-convergence in iterative imputation algorithms. We found that---in the cases considered---inferential validity was achieved after five to ten iterations, much earlier than indicated by diagnostic methods. We conclude that it never hurts to iterate longer, but such calculations hardly bring added value.

AB - Iterative imputation is a popular tool to accommodate missing data. While it is widely accepted that valid inferences can be obtained with this technique, these inferences all rely on algorithmic convergence. There is no consensus on how to evaluate the convergence properties of the method. Our study provides insight into identifying non-convergence in iterative imputation algorithms. We found that---in the cases considered---inferential validity was achieved after five to ten iterations, much earlier than indicated by diagnostic methods. We conclude that it never hurts to iterate longer, but such calculations hardly bring added value.

KW - non-convergence

KW - MICE

KW - Iterative imputation

UR - https://openreview.net/pdf?id=fHSVg6mVqpw

M3 - Conference contribution

BT - First Workshop on the Art of Learning with Missing Values (Artemiss) hosted by the 37 th International Conference on Machine Learning (ICML)

ER -

Missing the Point: Non-Convergence in Iterative Imputation Algorithms

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this