The issue you're hitting is that two entirely separate people looked at the Census and decided what to type as the transcription. Two different people, with two differing versions of the original (quality of copies, microfiche, actual originals), maybe years apart and certainly people of different ages in different parts of the country. 90% of the time they'll both get it correct, the other 10% they will err (as we all would) or differ slightly.
When looking at transcriptions, you must be aware of their original source and who would've done the transcription, so you can make informed choices as to whether to search on alternative record sets or whether they all come from the same original source. Or, simply check everything ... find some differences/omissions and maybe get lucky in seeing the originals and seeing for yourself.
All data sets are simply an opportunity to potentially find who you're looking for with a few keystrokes, instead of laboriously tracking down originals and seeing it for yourself. The gaps, or errors, or omissions you're left with are then your invitation to "dig deeper/further" if required.
