Struggling with Missingness in a data set

[klongway]

Hello,

I am running a logistic regression with four predictor variables, 1 outcome variable, and 6 covariates.  My N is around 2000 participants.

For one of my four predictor variables, there are around 1000 people missing the question.  The question asked was very sensitive and it is normal for people to skip it on surveys per other papers. The situation is similar for one of my covariates, where a question on a sensitive topic is also skipped by a substantial number of people. 

Right now when I run it, my data set ends up cut in half.  My major professor will not let me do multiple imputations and wants me to do Full Information Maximum Likelihood to account for the missing data.

Does anyone have any advice on if I can do this in SAS? 

Thank you!

[Norman21]

From what you describe, the people are not "missing" the question, they are deliberately choosing not to answer it. This is information that might be useful for your model.

You can try recoding this variable to "not answered" (or whatever) then run the logistic regression.

[PaigeMiller]

Some people will fill in the mean of the data in place of the missings and also create a new variable which is 0 if the original variable was not missing and 1 if the original variable was missing. Then you put the original variable and the new variable into the model. This essentially predicts the observations with the original variable missing at a constant offset from the observations when the original variable was not missing.

[Ksharp]

Do you want impute missing value by EM algorithm ?

@Rick_SAS  wrote a blog about it recently .

[Rick_SAS]

The SAS documentation has an example of how to compute FIML by using structural equation modeling with PROC CALIS.