Hello,
I am using WLSMV estimator in structural equation modelling to test the relationship between latent variables using Mplus. I have missing data and the default method for handling missing data with WLSMV is pairwise which produces unbiased estimates for parameters and standard errors under the assumption that data are missing completely at random (MCAR).
I am wondering whether missing at random (MAR) also enough to use pairwise deletion?
Best,
Pairwise deletion requires MCAR data. If you have MAR data that are not MCAR, you could use multiple imputation instead of pairwise deletion.
Jochen Wilhelm
MAR data is sufficient for multiple imputation. See, e.g.,Schafer, J. L., & Graham, J. W. (2002). Missing data: our view of the state of the art. Psychological Methods, 7(2), 147.
https://slyderse.folk.ntnu.no/medstat/KLMED8006/Shafer.pdf
Since MAR is not testable, it is recommended to include relevant auxiliary variables in the imputation phase to make MAR more likely to hold.
- Badges
- Science topic