Appendix: Simulation details and code

This is the online companion to the paper ‘An Imputation-Based Solution to Using Mismeasured Covariates in Propensity Score Analysis’, by Yenny Webb-Vargas, Kara Rudolph, David Lenis, Peter Murakami, and Elizabeth Stuart.

You can see the details of the simulations, color figures and code by following the links:

Normal scenario:

• Normal ‘Y’, normal (X,Z,W)
• Normal ‘Y’, normal (X,Z,W), including a simple imputation method
• Normal ‘Y’, normal (X,Z,W), with varying sample sizes for calibration and main data sets
• Normal ‘Y’, normal (X,Z,W), estimating Average Treatment effect on the Treated (ATT)

Non-normal scenarios:

• Normal ‘Y’, mixture (X,Z,W)
• Bernoulli ‘Y’, normal (X,Z,W)
• Bernoulli ‘Y’, mixture (X,Z,W)
• Mixture ‘Y’, normal (X,Z,W)
• Mixture ‘Y’, mixture (X,Z,W)
• Normal ‘Y’, normal (X,Z,W), where Z is a binary variable

You can see the code for Guo, Little and McConnell’s Multiple Imputation for External Calibration, or download the R script we used in the simulations.