Start when you first create the tables or featureclasses.
Now how many of you knew this was possible?
Solves your <null> problems since you will have to provide a valid nodata value, something which has meaning.
Python's math and numpy modules support the concept of NaN (not a number) .
Nan's are omitted in numeric calculations by functions that ignore them... automagically... eg. mean, vs nanmean
Text... never use "" or '' because you can't see them. If you are recording textual information, provide appropriate classes like .... "never measured", "not at home", "forgot", "wasn't me", "Nadda", or even "NONE" as text in all caps to differentiate it from the real None.
Think about this next time you decide to work with tabular data.