As i continue struggling with outlier detection a problem emerged...
I do my outlier tests in R following Tukey's method. I run the first test and i get some outliers. Lets say 1000 out of a 100k database. I wrote the function to convert their values to NA and the test itself ignores NA values. So when i run the test again i get lets say 690 outliers. When i "remove" them and run a third test i get 300 for example.
Is it sensible to run outlier tests and do what is mentioned above until the test shows 0 outlier? From a modelling standpoint i have read that it is recommended that the data set is "purified" from outliers.
Is it logical to say that the outliers found in the 2nd, 3rd, 4th and so on tests are outliers as much as the ones found by the first test? Asking this because i want to create a separate data base with all outliers found in the one i am analysing. So i am wondering if i should include the ones found in the 2nd, 3rd and so on tests.
Please, explain if i am doing it wrong.