MAR and MCAR data can be handled with multiple imputation or maximum likelihood estimation (sometimes referred to as "full information maximum likelihood" or FIML).

See, for example:

Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the state of the art. Psychological Methods, 7(2), 147–177. https://doi.org/10.1037/1082-989X.7.2.147

https://www.guilford.com/books/Applied-Missing-Data-Analysis/Craig-Enders/9781462549863

The Mplus software can be used to implement either MI or FIML in a very convenient/user-firendly way (see my Youtube playlist: https://www.youtube.com/watch?v=YVJJ_keoWeE&list=PL-kVjeOVYChpcwSchnFMrzMzwmDtkDavW). Many other packages for structural equation modeling also use FIML with missing data (e.g., AMOS, lavaan in R). Multiple imputation is available in many programs including SPSS.

Thank you so much, Mr. Christian Geiser !

Hi Mariel,

Under MCAR you can delete cases with missing data. However, this approach will reduce the sample size and therefore the power of your analyses.

Under MAR deleting cases with missing data will introduce bias in your results.

As Christian Geiser mentioned, multiple imputation and full information (direct) maximum likelihood are the right approaches for both MCAR and MAR.

For multiple imputation, you can consider using the "chained equations/sequential regression/fully conditional specification" approach. Two free software you can use for this approach are R ("mice" package) or IVEware.

Greetings,

Roberto

Thank you, Mr. Roberto. Actually, I am learning to use the python pandas as I had difficulty in MPlus. As for the R and iveware, I haven't tried it yet.