Aniella Kabel:

1) The problem of missing data and its imputation is a serious matter that may affect the final results of your analysis. Before you do Little's test or any imputation, please refer to some background literature to at least distinguish MCAR, MAR and MNAR. Personally, I recommend Article When and how should multiple imputation be used for handling...

2) I'm not an expert in Little's MCAR test. But, as I understand it, you should include ALL your covariates to look for ANY interrelations potentially relevant to MAR. Anyway, I tried to perform the test in R, but since its 4.0.0 version I could not find it anymore... (Aniella Kabel please help me on how to do Little's now?). Finally, in case of p

Hi,

It is generally advised to do listwise deletion (LD) if the missingness is about 5%. For the rest, we may run all analyses twice, once with data imputation and once with complete case analysis using LD. More often, the same substantive results occur in each method. If they do not, the dual approach enables us to determine what difference data imputation may make. Regardless, data imputation is preferred as it tends to reduce estimation bias.

If the p-value for Little's MCAR test is not significant, then the data may be assumed to be MCAR, and missingness is assumed not to matter for the analysis.

Here is an article which can help you:

http://www.statisticalassociates.com/missingvaluesanalysis_p.pdf

See the attached file which can help you

Dr. Salahuddin

Thank you all for your answers, I have learned a lot from you!

Michal S. Karbownik , The articles you provided were very helpfull, especially you recent paper with the supplementary Excel file. Unfortunately, I cannot help you with R. I use SPSS version 25.

Bhogaraju Anand , That 5% rule is very helpfull. in addition, comparing a complete case analysis with an analysis with imputation seems a good idea.

Salah Uddin, the article is somewhat complicated, but very informative, thank you.

Do you use all variables or just the independent?

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

P-values are not good stand-alone statistics. Percent of missing does not say much as circumstances vary. However, a simulation that says you can have up to 90% missing would seem to be for unreasonably idealized circumstances.

I suggest that whatever is used, a cross-validation oriented check should be done to see if results seem viable. If you cannot remove some data, run your technique(s) on the remaining data, and find that what you would impute for missing data that you actually held out is acceptably close, then you better try something else. But remember that modeling too closely to one set of circumstances might leave out others, so better not stop at one such test.

Cheers.