This data is st00pid. It relies on self-reporting, and the questions are coached to get the "required" result.
A while ago I did an experiment, I found the list of people convicted for typical "homeless" crimes in SF, and did a background search on them. About 90% had extensive crime records in other states, far away from CA.
This is not a definitive result for sure, but it's suggestive.