Hello,
The percentage of missing values between latent variables varied between 0 and 3.7%, and also missingness is completely at random (MCAR). I have used the pairwise deletion approach to handle missing data. Is this proportion of missing data logical to use pairwise deletion?
I'm not sure I fully understand your question, but here is a hopefully nonetheless useful answer:
The choice of the proper method of missing data handling (e.g., listwise/pairwise deletion, FIML, multiple imputation) depends on the missing data mechanism (MCAR, MAR, MNAR), not on the amount of missingness. If your data is truly MCAR, then listwise and pairwise deletion would lead to unbiased results. However, those methods might still cause a loss of statistical power because they don't make use of all available data points.
Multiple imputation (MI) and/or FIML are preferable because they only require the MAR assumption (which is less restrictive than MCAR) for unbiased results. MI and FIML also retain more statistical power and allow you to incorporate auxiliary variables (missing data correlates) in the analysis that can help you "achieve" the MAR condition.
In your case, it may not matter very much since your amount of missingness appears to be quite small.
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media