mutliple imputation of data missing not at random

[Brett Ory]

Dear statalist users,

I have data missing not at random, and my understanding is that I cannot use Stata's multiple imputation to impute this data because this method assumes that data is missing at random or completely at random. Is that correct?

Thank you for the help!
Brett

[Anders Alexandersson]

Yes, in the sense that for missing data not at random (MNAR) the process that causes missing data must be modeled to obtain valid results. You need to do sensitivity analysis for MNAR. StataCorp has "A note on how to perform multiple-imputation diagnostics in Stata".

For what it is worth, SAS has an MI procedure with an MNAR statement; see the section "Multiple Imputation with Pattern-Mixture Models" on page 5922 in the "SAS/STAT 14.1 User's Guide The MI Procedure" . The SAS implementation is also discussed in a blog.

[Rich Goldstein]

just to note that there are ways to "modify" standard MI results to help with the MNAR situation: (1) adding (subtracting) a fixed amount (sometimes called "delta ajustment"; for an example, see pp. 184-186 of van Buuren, S (2012), Flexible imputation of missing data, CRC Press; note that this is just one example; a general discussion appears in Rubin, DB (1987), "A noniterative samp=le/importance resampling alternative to the data augmentation algorithm for creating a few imputations when the fractions of missing information are modest," Journal of the American Statistical Association, 82: 543-546; (2) weighting; see, e.g., Carpenter, JR, Kenward, MG and White, IR (2007), " Sensitivity analysis after multiple imputation under missing at random: a weighting approach," Statistical Methods in Medical Research, 16: 259-275

[evank23]

This is quite relevant: https://www.stata-journal.com/articl...article=st0440

[Carlo Lazzaro]

See also https://link.springer.com/article/10...273-018-0650-5