FIML Requirements
Message/Author
 Hillary Gorin posted on Friday, January 20, 2017 - 8:48 am
Hi Dr. Muthen,

Do you know of any good resources for understanding sample size requirements for FIML and proportion of missing data?

Thank you!
Hillary
 Bengt O. Muthen posted on Friday, January 20, 2017 - 11:52 am
You can check Chapter 10 of our new book:

http://www.statmodel.com/Mplus_Book.shtml

 Hillary Gorin posted on Friday, January 20, 2017 - 1:34 pm
Hi Dr. Muthen,

Thank you very much for these resources! I took a look through them and they are very helpful. I am still a little confused about the proportion of data present that I need for FIML.

For example, I know that 50% of missing data is likely too much missing data for FIML to manage (and too much missing data to successfully impute). Is 1/3 (33%) missing data too much?

Thank you!
Hillary
 Bengt O. Muthen posted on Friday, January 20, 2017 - 6:07 pm
It really depends on how selective the missing data is. You should look at how sample statistics differ for variables without missing for those with 50% or 33% missing(on other variables) versus those without that missingness. 33% missing may still be too high.
 Hillary Gorin posted on Monday, January 23, 2017 - 8:50 am
Hi Dr. Muthen,

Thank you for your response. I am not sure I understand what you mean. I ran descriptives for the group with 33% missing data and those not missing that data. How can I tell how selective the missing data is?

Thank you!
Hillary
 Bengt O. Muthen posted on Monday, January 23, 2017 - 3:37 pm
Check if the means of the variables are different in these 2 groups.
 Hillary Gorin posted on Monday, January 23, 2017 - 5:29 pm
Means are 43 vs. 57. Highest possible score is 102. Is that too high?
 Bengt O. Muthen posted on Tuesday, January 24, 2017 - 6:35 am
You should discuss this with a statistical consultant.