little's test for missing datas

[kealnit]

Hello,

I'm searching the method to calculate the little's test (1988) for hypothesis of MNAR.

hypot

does anyone has the answer ?

thanks !

[Doc_Duke]

Try this google search

proc MI test mcar site:sas.com

PROC MI may be able to address your question.

[Rick_SAS · Accepted solution]

I don't think SAS implemented Little's test. However, the MNAR statement in PROC MI (SAS/STAT 13.1) performs sensitivity tests for the MAR assumption, which might be sufficient for you. In the book Multiple Imputation of Missing Data Using SAS, the authors say (p. 4) "Little (1988) presents a likelihood ratio test...that the data are MCAR.... Although useful in some special cases, this test is not sufficient to establish that the missing data mechanism is ignorable."

[H]

I will have to check out the Proc MI approach referenced above. Though, I just ran Little's test earlier this week using the following macro:

mcartest

[Rick_SAS]

Awesome.  Who wrote the macro and where can I download it?

[H]

I saw the macro referenced in a couple of SAS User Group papers.  One place it is located is here:

Macro Programs | Applied Missing Data .com

I also saw Yuming Ning listed as the author at another location. Though, I believe when using it, you can only use all numeric or categorical data (or perhaps numeric and categorical coded as integers). I would be interested in seeing the results from the macro compared to the results from a program where the test is built in.

How do you get at the MCAR test in Proc MI??

Side  note, I have your Simulation book Rick and just upgraded to SAS 9.4. If I get a chance I would like to try running the code to simulate multiple categorical variables. I could not get it to run right before. There was an erratum in the book and my version of SAS was  before the 9.3 second maintenance version. It does take much to confuse me!