Multiple imputation and test-wise deletion for causal discovery with incomplete cohort data.

Area of Science:

• Causal inference and machine learning
• Statistical modeling and data analysis

Background:

• Causal discovery algorithms estimate causal graphs from observational data, complementing individual treatment-outcome analyses.
• Constraint-based causal discovery algorithms traditionally struggle with missing data, limiting their application.

Purpose of the Study:

• To investigate and compare the performance of test-wise deletion and multiple imputation for handling missing values in causal discovery.
• To establish conditions for causal structure recoverability using test-wise deletion.

Main Methods:

• Simulated data from benchmark causal graphs were used to compare various imputation methods.
• Methods evaluated include test-wise deletion, multiple imputation, list-wise deletion, single imputation, random forest imputation, and a hybrid approach.

Main Results:

• Both test-wise deletion and multiple imputation significantly outperform list-wise deletion and single imputation.
• Multiple imputation shows particular utility with small datasets containing only Gaussian or discrete variables.
• Performance is mixed when datasets contain a combination of Gaussian and discrete variables.

Conclusions:

• Test-wise deletion and multiple imputation offer viable solutions for incorporating missing data into causal discovery.
• The choice of method depends on the characteristics of the data, particularly variable types and dataset size.