Missing at random (MAR) or Missing not at random (MNAR)

[Sunniva]

Dear stata users,

I have ran a linear mixed model in Stata to measure the effectiveness of an intervention, using full information maximum likelihood (ML) to handle missing data. ML providing unbiased estimates under the assumption of data being missing at random. I wonder how I can diagnose the missing data mechanism in my data. In other words, how can I know if my data is Missing at random (MAR) or Missing not at random (MNAR)? Is there any test in Stata for this?

Many thanks

[Rich Goldstein]

there is no test even theoretically; i.e., MAR is a non-testable assumption; you can do sensitivity analyses

[Carlo Lazzaro]

Sunniva:
https://www.guilford.com/books/Missi.../9781593853938, Chapter 3 ficuses on this difference.
As Rich helpfully warned you about, MAR is a not testable assumption, but relies on a sound a priori knowledge of the data generating process that you're investigating.

[Sunniva]

Many thanks, this was very useful.