I have a dataset with 891 observations and 12 variables. From them, 2 have NA values (V1 has 20% NA and V2 has 77% NA).
First I examined if the data are MAR. So, I have created 2 new variables named V1_NA and V2_NA that act as a flag on whether that variables have NA (so 0 if NA, 1 otherwise). Then I performed Chi Square Independence Test between each of them and the response variable and in both cases I rejected the NULL (p-value < 0.00001), so I concluded that the data are NOT MAR.
I was planning on using KNN imputation but I am not sure if that is appropriate now since the data are not MAR. I have found conflicting views on the subject with some suggesting I cannot use imputation (https://www.theanalysisfactor.com/missing-data-mechanism/) and others suggesting that KNN algorithm is applicable in non MAR data (https://towardsdatascience.com/the-use-of-knn-for-missing-values-cf33d935c637).
So to sum up my questions are:
Is my approach correct to infer that the data are MAR? If not how should I test alternatively?
If my approach is correct for infering MAR, can I use KNN imputation?