Hi Chelsea,

I don't understand your logic. Whether missings are MAR or NRM depends on your theoretical model. MAR does not mean that the missing are random (yes that's confusing and MAR is a real misnomer). It means that the missing process is not related to your dependent variable(s) unconditionally or after controlling for model variables, see

Thoemmes, F., & Mohan, K. (2015). Graphical representation of missing data problems. Structural Equation Modeling: A Multidisciplinary Journal, 22(4), 631-642.

This one paper tells you more about missing data than the whole statistical literature. Especially Figure 3 explains when missingness will become non-random and harmful. As any causal process, nothing works without considerations of plausibility and theoretical assumptions.

All the best with your work,

--Holger

This question was also on:

https://www.researchgate.net/post/If-missing-data-is-not-MCAR-how-to-proceed-further-using-SPSS

Well, I don't suggest that you just "...analyze what is there," or use mean values (thus depressing actual variance, as well as perhaps providing low quality estimates).

Imputation using regression has the advantage of maintaining structure if your model is 'correct,' and you can also estimate the variance of the prediction error. You can study how well a model fits by using a "graphical residual analysis" (which also helps look at heteroscedasticity - please see https://www.researchgate.net/publication/333642828_Estimating_the_Coefficient_of_Heteroscedasticity), and you can try to avoid overfitting by leaving out some cases that you actually have, to see how well you would predict for them - which might be particularly helpful here. (See "cross-validation.") Here you might pick some cases you suspect are like the missing ones to see how well the model performs, though that is somewhat subjective, but definitely something you can document. You might also do a more formal cross-validation. Once you are happy with your model, that should do well for you. (See "ignorable nonresponse," and "nonignorable nonresponse.")

So, if you have a good regression model and good data, or models by subpopulation, I like that idea, when applicable.

Cheers - Jim

If you truly believe the data are missing not at random, in which case, the missingness is related to the missing value itself (not predicted by observed data), the best way to approach it is sensitivity analysis - you may check out pattern-mixture model. Please note that you cannot test whether the data are missing at random or missing not at random.

I have two recommendations for you.

The first paper deployed several imputation techniques for comparison.

The second paper implemented a novel approach for both Missing assessment and treatment of Missing values  .

I think the size of Missing values are discussed very well. . .

"Knn in row" is a new imputation technique with  wide range of applications.

https://ieeexplore.ieee.org/abstract/document/8859963

https://www.researchgate.net/publication/349384200_Investigating_the_impact_of_missing_data_imputation_techniques_on_battery_energy_management_system?_sg=T89qlVyc07IPNPI2AGCHIStg_maiCNHj8UZq6nO7fwgxVSZ7fyEW2Ao8owGbUxSVFnDOBLQkxps3Hzk_hraC1-OJ5wXs9YThAVJy4ZBk.G6mSFbsYwPE2eZv0huCWE92yDYRU78bzWOat5T1tTFmYILtMiYkWLZdw4Bq4o6-pfmGboQtsu9IKbnSTjO80Ug